Minerva Access is the Institutional Repository of The University of Melbourne
Author/s:
Hurley, James Crowther
Title:
Endotoxemia: correlation with gram-negative bacteremia and association with outcome
Date:
2013
Citation:
Hurley, J. C. (2013). Endotoxemia: correlation with gram-negative bacteremia and
association with outcome. Doctorate, Department of Medicine (Austin Health), Faculty of
Medicine, Dentistry and Health Sciences, The University of Melbourne.
Persistent Link:
http://hdl.handle.net/11343/38393
File Description:
Endotoxemia: correlation with gram-negative bacteremia and association with outcome
Terms and Conditions:
Terms and Conditions: Copyright in works deposited in Minerva Access is retained by the
copyright owner. The work may not be altered without permission from the copyright owner.
Readers may only download, print and save electronic copies of whole works for their own
personal non-commercial use. Any use that exceeds these limits requires permission from
the copyright owner. Attribution is essential when quoting or paraphrasing from these works.
Endotoxemia: correlation with gram-negative bacteremia and
association with outcome
A thesis submitted for Doctor of Medical Science (by compilation of published papers)
James Crowther Hurley
B Med Sci, MB BS, M Epidemiol, PhD, FRACP
This thesis is submitted in total fulfilment of the requirements for the degree of Doctor
of Medical Science, through the Austin Health Department of Medicine, Faculty of
Medicine, Dentistry & Health Sciences University of Melbourne, September 2013.
-3-
Contents:
1. Front-matter
1.1. Abstract
1.2. The compiled publications that form this thesis
2. Introduction and overview
2.1. Endotoxin
2.1.1. Structure-activity
2.2. Assays for endotoxin
2.2.1. The limulus amebocyte lysate (LAL, ‗limulus‘) assay
2.3. Sepsis: toward diagnostic and prognostic tests
2.3.1. Tests based on clinical criteria
2.3.2. Laboratory based tests
2.3.3. The potential for the limulus assay
2.4. Gram-negative (GN) bacteremia
2.4.1. GN bacteremia types
2.4.2. Pathophysiology of GN bacteremia
2.5. The endotoxin-sepsis ‗disconnect‘
2.5.1. Anti-endotoxin therapies: pre-clinical disconnect
2.5.2. Anti-endotoxin therapies: clinical disconnect
2.5.3. Prognostic value of endotoxemia: clinical experience disconnect
2.6. Meta-analysis: Towards a resolution
3. Aims and objectives of this thesis
3.1. A statement of the problem
3.2. The research questions
4. Materials and methodology
4.1. Literature search methods
4.2. Analytic approach
4.3. Three phases of statistical methods
5. Results and Discussion
5.1. Endotoxemia detection as a diagnostic test
5.1.1. Concordance with detection of GN bacteremia
5.1.2. Impact of GN bacteremia species type
5.1.3. Study limitations and literature reconciliation
5.2. Endotoxemia detection as a test of prognosis
5.2.1. Analysis as ‗two group‘ studies
5.2.2. Analysis as ‗four group‘ studies
5.2.3. Impact of GN bacteremia species type
5.2.4. Study limitations and literature reconciliation
6. Summary and conclusions
6.1. Summary of findings
6.2. Recapitulation of study limitations and strengths
6.3. Future prospects
7. Catalogue of studies and personal correspondence
8. References
9. Appendix: the publications
-4-
DECLARATION
The following declaration, is signed by the student: This is to certify that
i. the thesis comprises only my original work towards the Doctor of Medical
Science except where co-authorship is indicated in (2.)Published articles that
form this thesis,
ii. due acknowledgement has been made in the text to all other material used
iii. the thesis is less than 100,000 words in length, exclusive of tables, maps,
bibliographies and appendices.
Signature: ....................................................................
Acknowledgement
I woud like to thank Prof Geoff Zajac and Prof Lindsay Grayson for the opportunity to submit
this thesis through The Department of Medicine at The University of Melbourne. I would like
to thank and acknowledge the generous donation of data toward the research undertaken here
from the following contributors; Dr D Berger, Klinikum der Universitat Ulm, Germany; Dr P.
Brandtzaeg, Ulleval University Hospital, Oslo, Norway; Dr B Byl, Erasme University Hospital,
Belgium; Dr R Danner, NIH, Bethesda, MD, USA; Dr P Engervall, Karolinska Hospital,
Stockholm, Sweden; Dr KCH Fearon, Royal Infirmary, Edinburgh, UK; Dr EJ GiamarellosBourboulis, Laiko General Hospital, Athens ; Dr B. Guidet, Hopital Saint-Antoine, Paris,
France; Dr P Ketchum (Associates of Cape Cod) and Dr D Bates, Brigham and Women‗s
Hospital, Boston, MA, USA; Dr J. Levin, VA Medical Center, San Francisco, USA; Dr D.
Massignon, Center Hospitalier Lyon Sud, France; Dr E Maury, Hopital Saint Antoine, Paris,
France; Dr MJ McMahon, Leeds Teaching Hospital, Leeds, UK; Pr G. Offenstadt, Hopital
Saint Antoine, Paris, France; Prof Steven M Opal, MD, Infectious Disease Division, The Alpert
Medical School of Brown University, Providence, Rhode Island; Dr. Jan M. Prins, Academic
Medical Center, Amsterdam; Dr JT van Dissel, Leiden University Medical Center, The
Netherlands; Dr SM Willatts, Bristol royal Infirmary, Bristol, UK and Dr M Yoshida, Jichi
Medical School, Japan. This research would not have been possible without their donations. I
thank my wife Lucy Callaghan for her encouragement and patient proof reading. All errors are
mine.
-5-
1. Front-matter
1.1. Abstract
Introduction: This thesis is a compilation of 17 publications; eight provide
background material and nine provide original meta-analysis. The thesis investigates
the association of endotoxemia with gram-negative (GN) bacteremia on the one hand,
and with outcome (mortality) on the other, in the extensive literature experience. This
experience is conflicting at multiple levels. The research questions addressed here
relate to the detection of endotoxemia in various clinical settings with respect to,
(i)
what is its diagnostic relevance versus GN bacteremia detection?
(ii)
what is its prognostic relevance versus outcome (mortality)? and
(iii) whether the disparate study results are reconcilable?
Methods: The research occurred in three phases over an 18 year period as more
data and newer analytic methods became available. Literature searches were repeatedly
conducted for clinical studies that met the following generic inclusion criteria;
endotoxemia detection which also had either data for GN bacteremia detection, GN
bacteremia type or mortality. These inclusion criteria were modified for specific
analyses. I also published a ‗call for data‘ (2001) and corresponded with authors to
obtain additional data. Various analytic methods were used (summary in Table 4.1);
Mantel-Haenszel methods (phase 1 - from 1994), ROC methods (phase 2 - from 2000),
and multi-level modelling (phase 3 - from 2009), to derive summary measures and
measures of heterogeneity and also to address the specific research questions that arose.
The accumulated data and the results were submitted for peer review publication.
Results: Over 300 potentially relevant studies had been published between 1970
and 2012. Of those, 107 studies met the generic inclusion criteria and had data that
were useable including 15 that were restricted to specific GN bacteremia types and 22
for which clarifications of data were received. In relation to the above questions;
(i)
Diagnostic relevance. As a test for GN bacteremia, endotoxemia detection has a
sensitivity and specificity of less than 75%. Surprisingly the correlation of
endotoxemia with GN bacteremia overall is no better in the studies that used a
chromogenic (more sensitive) version of the limulus assay versus the original
gelation (less sensitive) version. However, the association of endotoxemia with GN
bacteremia is variable for different GN bacteremia types and for different study
-6-
settings. Indeed, the type of GN bacteremia is at least as important a determinant
toward endotoxemia detection as is the sensitivity of the assay method used.
(ii)
Prognostic relevance. The mortality risk associated with the detection of
endotoxemia or GN bacteremia either alone or together versus the detection of
neither is either non-significant or borderline (Odds ratio <2) among the 9 studies
undertaken in an ICU setting. For studies undertaken outside of an ICU setting the
co-detection of GN bacteremia and endotoxemia is most predictive of increased
mortality risk versus the detection of neither but there is substantial heterogeneity
associated with this finding. Surprisingly, in the co-presence of E. coli bacteremias,
endotoxemia has no prognostic relevance for mortality in any setting.
(iii) Reconciliation. The disparity among the study results, which is quantifiable as
heterogeneity, is partly attributable to publication bias but more so to variations in
GN bacteremia occurrence, GN co-detection, GN bacteremia species types and
underlying mortality risk in the different study populations. These patient and study
level factors are previously unrecognized confounders in the interpretation of the
clinical significance of endotoxemia detection.
Conclusions: Endotoxemia is not detectable for at least 20% and up to 50% of
patients with GN bacteremia. Moreover, the relevance of endotoxemia detection to
prognosis is dependent on the co-detection of GN bacteremia, the GN bacteremia type
and the underlying mortality risk in the study population. The studies which reached
seemingly disparate conclusions can be reconciled on the basis of these confounding
relationships. There are implications from this research for the design and interpretation
of trials of anti-endotoxemia therapies.
Abbreviations
DOR Diagnostic odds ratio
LAL
Limulus Amoebocyte Lysate
ROC Receiver operating characteristic
SROC Summary Receiver operating characteristic
HSROC Hierarchical Summary Receiver operating characteristic
CI
Confidence interval
GN
gram-negative
-7-
1.2. The published articles that form this thesis
1.2.1.
Original research publications
(Contribution of Hurley JC to each publication is 100% unless stated otherwise)
Code Citation
JCM 1994 Hurley JC. 1994. Endotoxemia: concordance with gram-negative
bacteremia in patients with gram-negative sepsis: a reappraisal with
meta-analysis. J. Clin. Microbiol. 32:2120-2127.
http://jcm.asm.org/content/32/9/2120.full.pdf
JCM 1995 Hurley JC. 1995. Bacteremia, endotoxemia and mortality in gram-negative
sepsis: a reappraisal with meta-analysis. J. Clin. Microbiol. 33: 1278-82.
http://jcm.asm.org/content/33/5/1278.short
EIH&D Hurley JC, & Levin J* (1999). Relevance of endotoxin detection in
sepsis. In Endotoxin in Health and Disease., Helmut Brade, Stephen
1999
Opal Stefanie Vogel, David Morrison (editors), Marcel Dekker
Limited. 841 – 854. * Contribution of Hurley JC to publication is
90%.
Hurley
JC. (2000). Endotoxemia: concordance with gram-negative
AP&LM
2000 bacteremia: a meta – analysis using ROC curves. Arch. Pathol Lab Med.
124: 1157 – 64.
http://www.archivesofpathology.org/doi/full/10.1043/00039985(2000)124%3C1157:COEGNB%3E2.0.CO;2
JER 2003 Hurley JC. (2003). Endotoxemia and gram-negative bacteremia as
predictors of outcome in sepsis: a meta-analysis using ROC curves. J
Endotoxin Res. 9; 271-279.
http://ini.sagepub.com/content/9/5/271.short
JCM 2009 Hurley JC. (2009). The detection of endotoxemia with gram-negative
bacteremia is bacterial species dependent: A meta-analysis of clinical
studies. J. Clin. Microbiol. 47: 3826-3831
http://jcm.asm.org/content/47/12/3826.full
EJCM&ID Hurley JC. (2009). Does gram-negative bacteremia occur without
endotoxemia? A meta-analysis using hierarchical summary ROC curves.
2010
Eur. J. Clin. Microbiol. Infect Dis. 29: 207-215.
http://rd.springer.com/article/10.1007/s10096-009-0841-2#page-1
CC 2012 Hurley JC, Guidet B, Offenstadt G, Maury E. (2012). Co-dependence of
endotoxemia, gram-negative bacteremia and underlying risk as
predictors of mortality: a meta-analysis using L‘Abbé plots. Crit Care
16: R148 * Contribution of Hurley JC to publication is 90%.
http://www.biomedcentral.com/content/pdf/cc11462.pdf
Innate Hurley JC, Opal S. (2013). Prognostic value of endotoxemia in patients
with gram-negative bacteremia is bacterial species dependent: an
Immunity
individual patient data meta-analysis. J. Innate Immunity 5(6) (DOI:
(in press)
10.1159/000347172) (in press)* Contribution of Hurley JC to
publication is 90%.
http://www.karger.com/Article/Abstract/347172
-8-
1.2.2.
Background publications
Code Citation
Lancet Hurley JC. (1993) Reappraisal of the role of endotoxin in the sepsis
syndrome (Medical hypothesis). Lancet 341: 1133-5.
1993
http://www.sciencedirect.com/science/article/pii/014067369393139R
CMR 1995 Hurley JC. (1995) Endotoxin: methods of detection and clinical correlates.
Clin. Microbiol. Rev. 8:268-292.
http://cmr.asm.org/content/8/2/268.full.pdf
Drugs 1994 Hurley JC. (1994) Sepsis management and anti-endotoxin therapy after
nebacumab; A reappraisal. (leading article). Drugs 47:855-861.
http://rd.springer.com/article/10.2165/00003495-199447060-00001#page-1
Drug Hurley JC. (1995) Antibiotic induced release of endotoxin: a therapeutic
paradox. Drug Safety. 12:183-195
Safety ‘95
http://rd.springer.com/article/10.2165/00002018-199512030-00004#
EOID 1995 Hurley JC. (1995) Endotoxemia and novel therapies for the treatment of
sepsis. Expert Opinion Invest. Drugs. 4(3):163-174
http://informahealthcare.com/doi/abs/10.1517/13543784.4.3.163
EOID 1996 Hurley JC. (1996). Meta analysis and investigation of anti – infective
therapies. Expert Opinion Invest. Drugs 6: 159 – 167.
http://informahealthcare.com/doi/abs/10.1517/13543784.6.2.159
JER 2001 Hurley JC. (2003). Endotoxemia and gram-negative bacteremia as
predictors of outcome in sepsis: a meta-analysis using ROC curves:
call for data (letter). J Endotoxin Res. 7(6) 467.
http://ini.sagepub.com/content/7/6/467.extract
AP&LM Hurley JC. (2011). Meta-analysis of Clinical Studies of Diagnostic Tests:
Developments in How the Receiver Operating Characteristic
2011
―Works‖. Arch. Pathol Lab Med. 135: 1585-1590
http://www.archivesofpathology.org/doi/abs/10.5858/arpa.2011-0016-SO
-9-
2. Introduction and overview
This thesis analyses the relationship between the detection of endotoxemia
versus gram-negative bacteremia and also versus outcome as mortality amongst the
results of 107 studies [1-107] derived from the literature (see methods – chapter 4).
This chapter reviews the conflicts within the broader literature underlying the research
and the research questions are framed in chapter 3. The methodological approach used
in this thesis is described and explained in Chapter 4. The eight background
publications (listed in 1.2.2) provide additional background relating to endotoxin
detection (CMR ’95, EOID ’95), sepsis trials (Drugs ’94, Drug Safety ’95) an
alternate hypothesis for sepsis (Lancet ’93), and statistical methods (EOID ’96,
AP&LM ’11). This additional background is not duplicated here.
2.1. Endotoxin
Lipopolysaccharide (LPS, endotoxin) is an outer membrane component of GN
(gram-negative) bacteria (figure 2.1) [108-110]. Each E. coli bacterial cell has
approximately 106 lipopolysaccharide molecules [110]. Endotoxin has potent and broad
ranging biological activities in humans (and in other species) which are mediated
mostly by the lipid-A residue within the molecule (figure 2.2). For these reasons it has
long been identified not only as a potential marker of GN infection [111-112] but also
as a mediator and hence a potential target for specific anti-endotoxin therapies [113115]. Surprisingly however, these potentials have not been realized and the conflicts
among both the studies of interest [1-107] and the broader literature are reviewed in this
chapter.
There are numerous effects induced by administration of endotoxin to
experimental animals and humans [116-118]. Recent review articles list over 20
humoral, cellular, immunological and metabolic effects. The effect of primary interest
in this thesis is lethality. There are three difficulties in studying endotoxin in general
and in quantifying endotoxemia concentrations in particular [111-112]. Firstly, the
biological effects of endotoxin are not a uniform gravimetric property of the molecule
[119]. Hence the quantity of endotoxin is usually interpreted as the amount of
biological activity in comparison to a reference endotoxin preparation assayed in
-10-
Chapter 2: Introduction & overview
parallel.
porin
LPS
Lipid-A
lipoprotein
peptidoglycan
Periplasm
Inner
Membrane
membrane
protein
Fig. 2.1 The location of the lipopolysaccharide (endotoxin) molecule in the
cell wall of Gram negative bacteria
The second issue is the unit of measurement for endotoxin concentrations.
Endotoxin concentrations are often expressed either in weight (ng/ml or pg/ml) or
activity (Endotoxin Units; EU/ml) units in comparison to one of a range of reference
endotoxins [120-123]. The conversion has been often quoted as 1 EU ≈ 100 pg/ml
[124-126] However, the choice of reference endotoxin and also the assay conditions
will impact on this quantification. Standard reference endotoxin preparations are
available. However, even these reference endotoxin preparations have been reported to
be subject to imprecision due to storage [122] and other factors. Also, assay conditions
need to be rigorously standardized and controlled in order to give precise and
reproducible quantification. For these reasons and others (see below), in this thesis, the
interest is mostly in endotoxemia detection as a qualitative rather than quantitative
measure. That is, the presence or absence of detectable levels of endotoxemia with
reference to the assay sensitivity level is the measure of primary interest rather than the
quantitative level. As a consequence of these assay issues, a quantitative level of
endotoxemia is best interpreted on a logarithmic rather than on a linear scale. Within
this thesis this interpretation is most relevant in relation to the sensitivity level of
Chapter 2: Introduction & overview
-11-
endotoxemia detection assays.
The third difficulty is that with so many biological properties of endotoxin, it is
problematic as to which is the most relevant biological activity on which to base a
detection assay [127]. In studying the activity of chemically synthesized lipid-A
analogues, Takada et al [128] proposed a classification of lipid-A bio-activities as
determined by dependence on structural specificity into the following three categories;
1. Lethality in chick embryos, Shwartzman reaction, pyrogenicity
2. Lethality in galactosamine pre-treated mice, interferon and TNF inducing
activities in prepared mice, polyclonal B-cell activation, limulus activity
3. Murine macrophage stimulation, IL-1 generation (in vitro), complement cascade
activation in human serum.
In this thesis, the activity of endotoxin in the Limulus amoebocyte lysate (LAL,
limulus) assay (see 2.2.1) is the biological assay of primary interest as it is the most
sensitive and the most commonly studied assay within the clinical context. However,
when comparing the biological activity of endotoxins from different GN bacteria, the
results of various in vitro assays may correlate poorly in predicting the lethality
response. These differences between these various assays versus the LAL assay may be
a reflection of the physical aggregation state of lipopolysaccharide molecule [129]. The
extent of the discrepancy between different assay methods is not trivial [130-135]. The
biological potency for 11 (Dehus et al [132]), five (Devleeschouwer etal [133]) and ten
(Laude-Sharp et al [130]) different LPS preparations from various sources was
investigated with the finding that the activity of LPS preparations from some (e.g. V.
cholera, P. aeruginosa, P. putida) versus other (e.g. E. coli, Salmonella abortus equii)
sources was overestimated by as much as a factor of 300 in the LAL test versus the
relative activities in whole cell cytokine expression assays or rabbit pyrogen assay. A
further complicating consideration is that the relative ranking of activities of different
preparations and bacterial origins of LPS varies from assay to assay to the point of
being inverted [134-139]. The clinical relevance of endotoxemia originating from
different GN bacteremia types is a research question to be addressed in this thesis.
2.1.1. Structure activity
The term ‗endotoxin‘ was attributed to Richard Pfeiffer who in the 1890‘s
distinguished the toxic properties contained within the GN bacterial cell versus those
-12-
Chapter 2: Introduction & overview
released outside the cell (exotoxins) [140]. Progress in defining the structure-activity
relationship in the mediation of the biological activities of endotoxin was difficult and
the exact structure of the molecule was elusive for several decades. It was only in the
mid 1980‘s that the lipid-A moiety of the lipopolysaccharide molecule of Escherichia
coli was totally chemically synthesized in a form available for structure activity studies
[141]. With these structure activity studies, it became possible to attribute the biological
activities of endotoxin to the lipid-A component of the lipopolysaccharide (endotoxin)
molecule. Recent microbiological studies have identified specific variations in the
structure of lipid-A, the mechanisms regulating the synthesis of lipid-A within gramnegative bacteria and the possible relevance of different lipid-A structures to the
pathogenesis of GN infection [142-143].
In humans, cautious endotoxin challenge studies have been able to replicate
many of the biological effects of endotoxin observed in non-human studies [144].
Interestingly, endotoxin has several pharmacological properties that are unusual for a
toxin. It has long been recognized that human plasma accentuates some of the effects of
endotoxin [145-147]. Moreover, the toxicity of endotoxin is indirectly mediated and
endotoxin itself is not cytotoxic, a phenomenon which is best exemplified in the
C3H/HeJ mouse strain [148-149]. This inbred mouse strain is resistant to the effects of
Table 2.1.1 LD50 Challenge studies in endotoxin responsive (C3H/HeN) versus nonresponsive (C3H/HeJ) mice.
LD50 challenge dose
Challenge
endotoxin
O18 LPS
C3H/HeN
C3H/HeJ
(endotoxin
responsive)
(endotoxin
resistant)
400 mcg
3750 mcg
104 cfu
101 cfu
live bacteria
E. coli 018 K1+
o (virulent E. coli strain)
E. coli 018 K1o (non-virulent E. coli strain)
107 cfu
107 cfu
Adapted from Cross AS, et al., Pre-treatment with recombinant murine
tumour necrosis factor and murine inter-leukin 1 protects mice from lethal
infection. J Exp Med 169;2021 (1989) [148]
Foot notes; Cfu = colony forming units
Chapter 2: Introduction & overview
-13-
endotoxin and this non-responsiveness is a genetically determined trait (Table 2.1.1).
Moreover, the susceptibility to endotoxin can be restored by the transfer of bone
marrow cells from histocompatible C3H/HeN strain mice which have normal endotoxin
responsiveness. This observation implicated the mediation of the effects of endotoxin
by a genetically encoded host receptor. This ultimately enabled the identification of the
gene for the endotoxin recognition (lps), and in turn, the LPS receptor itself (see below).
Endotoxin has potent biological activities in humans [144, 150-155]. In humans,
biological responses are apparent at doses as small as 4 ng/kg [155] whereas doses of 1
mcg/kg (rabbits, sheep [156-157]) or as high as 2 mg/kg (rats, pigs, non-human
primates [158-163]) are required for response in other species (Table 2.1.2). This
potent activity has been subjected to extensive preclinical studies in animals (see
below) in attempts to better define the role of endotoxin as a mediator in the
pathogenesis of GN infections. However, the high sensitivity of humans to the effects
of endotoxin is problematic for endotoxin research in that replication in experimental
animals requires artificial interventions such as the administration of galactosamine to
achieve sensitization to the effects of endotoxin comparable to that seen in humans.
LPS structure. The lipopolysaccharide molecule from all clinically relevant
Table 2.1.2 Comparisons of lethal endotoxin dosing in different species.
Species
Dose
LPS type
Dosing
partial
lethal
Ref
mcg/kg
Rat
15000
E. coli O26:B6
bolus
yes
158
Baboon
1500
E. coli O26:B6
10 min
yes
161
5
Salmonella
hours
yes
159
(mcg/kg/hr)
abortus equii
Sheep
0.3-2
E. coli
bolus
yes
157
Rabbit
1
E. coli
30 min
yes
156
Chimpanzee
0.004
E. coli (EC-5)
bolus
no
160
Human
0.004
E. coli (EC-5)
bolus
no
155
Pig
Adapted from Redl H, Schlag G, Bahrami S. Endotoxemia in primate models.
In Endotoxin in Health and Disease, Brade H, Opal S, Vogel S, Morrison D
(editors), Marcel Dekker Limited. 1999; 795– 808 [162]
-14-
Chapter 2: Introduction & overview
n
O-Specific chain
Core region
Lipid-A
Monosaccharide
Glucosamine
Ethanolamine
Long chain fatty acid
Phosphate
2-keto-3-deoxyoctonic acid, KDO
Fig. 2.2 The components of the lipopolysaccharide (endotoxin) molecule
GN bacteria studied consists of three components; a polysaccharide chain, a core
oligosaccharide and a lipid component, lipid-A (figure 2.2) [110, 137, 145-151]. The
LPS molecule, and in particularly, the lipid-A component, has been described as an
―information rich‖ molecule, with many possible sites for specific recognition by
prokaryotic and eukaryotic proteins. The structure of the polysaccharide chain of the
LPS molecule is highly variable even within species of GN bacteria whereas the lipid-A
molecule is broadly conserved across GN bacteria of different types (see below) [151154].
The polysaccharide consists of a repeating saccharide unit in a chain which is
hydrophilic and antigenic. In Enterobacteriaceae (e.g. Escherichia coli), the
polysaccharide is variable in length and in composition between bacterial strains, and
confers the bacterial strain‘s O-antigen specificity. Less complete LPS structures are
present in some genera. For example, the O-specific antigenic chain is not present in
mucosal pathogens such as Neisseria, Bordetella and Haemophilus and LPS in these
genera have only a core oligosaccharide. The core oligosaccharide can be divided into
an inner core and outer core region and for the genus Chlamydia even the outer core
region is not present and these genera have only an inner core [110, 140]. The inner
core region is at the innermost end of the polysaccharide chain and includes 3 KDO (2keto-3-deoxyoctonic acid) residues. The KDO provides the covalent linkage between
the polysaccharide chain and the lipid component of the LPS molecule.
Chapter 2: Introduction & overview
-15-
Lipid-A contains the key molecular components which determine the endotoxic
activity of LPS [150-154]. In addition, the lipid-A and KDO together have a key
structural role in the bacterial cell and are required for growth. The lipid-A and KDO
synthetic pathway is highly conserved. Hence, this synthetic pathway within the cell
would be an attractive target for anti-bacterial therapies. Studies of chemically
synthesized partial lipid-A molecules have identified the key structural requirements for
optimal biological activity in peptide mediator induction from murine macrophage cell
lines [148-154, 164].
Fig. 2.3 Chemical structure of E. coli lipid A with six acyl chains (from “Review: Structure and
function of lipopolysaccharides”, By Erridge C, Bennett-Guerrero E, Poxton IR. In Microbes
and Infection. 2002;4(8):837-851 [165]
The structure of E. coli lipid-A consists of a diglucosamine with two phosphates
and six acyl (fatty acid) chains (hexaacyl LPS) (Figure 2.3) [165]. Two 3hydroxymyristate (fatty acid) chains are attached directly to each of the two
glucosamines with two secondary (―piggyback‖) chains also attached. The fatty acids
may be laurate (C12), myristate (C14) with sometimes palmitate (C16) found as the
secondary acyl chain, and the primary (glucosamine-linked) acyl chains typically
having 12 carbons. This hexaacyl LPS structure is optimal for recognition by the MD2-TLR4 receptor. LPSs from different gram-negative bacteria may have more or fewer
acyl chains, longer acyl chains, branched acyl chains, unsaturated acyl chains, only one
phosphate, or other modifications. The degree of recognition of the various hexaacyl
and non- hexaacyl LPS structures by the MD-2-TLR4 receptor determines the
mediation of endotoxic biological activity and examples are indicated in Figure 2.4
[165, 166].
-16-
Chapter 2: Introduction & overview
LPS supra-molecular structure activity The structure of the lipid-A monomer
also determines its molecular shape [167-169]. These shapes are variably conical or
cylindrical depending on the ratios between the hydrophobic versus hydrophilic regions.
The most endotoxically active lipid-A monomer structure is conical whereas less active
Fig. 2.4 Selected Gram-negative lipid A and derivative structures. The endotoxic activities
given for each compound are included as a qualitative guide or estimate. (from “Review:
Structure and function of lipopolysaccharides”, By Erridge C, Bennett-Guerrero E, Poxton
IR. In Microbes and Infection. 2002;4(8):837-851 [165]
lipid-A is cylindrical. The activity of endotoxin can be increased by sonication [170].
At concentrations above a critical micelle concentration, the lipid-A monomers
aggregate into supra-molecular structures determined by the molecular shape of the
lipid-A monomers leading variously to the formation of lamellar, cubic or hexagonal
shapes (Figure 2.5). The biological significance of these supra-molecular structures is
uncertain. These supra-molecular structures are sometimes large enough to be viewed
Chapter 2: Introduction & overview
-17-
Fig. 2.5 Shape of LPS molecules and resultant supra-molecular structures. Three commonly seen
conformations of LPS. The ratio between the upper (hydrophilic) and lower (hydrophobic) surface areas has
been proposed to be the primary determinant of supra-molecular structure. From Seydel U. Seydel, H.
Labischinski, M. Kastowsky, K. Brandenburg. Phase behaviour, supramolecular structure and molecular
conformation of lipopolysaccharide. Immunobiology, 187 (1993), pp. 191–211.
under electron microscopy [169].
Lipopolysaccharide-binding proteins The serum protein lipopolysaccharidebinding protein (LBP) binds to the lipid A component of bacterial endotoxin and
facilitates its delivery to the CD14 antigen on the macrophage, where pro-inflammatory
cytokines are released and a cascade of host mediators is initiated. The neutrophil
granular protein bactericidal/permeability-increasing protein (BPI) competes with LBP
for endotoxin binding and functions as a molecular antagonist of LBP- endotoxin
interactions. The relative concentrations of these two proteins expressed as a ratio is
correlates with the quantity of neutrophils at sites of inflammation in various body
cavities [182]. The quantitative level of LBP in the serum of patients with severe sepsis
may have prognostic relevance [65].
The specific nature of this binding between these serum proteins and LPS has
prompted structure activity studies to synthesize peptides to enable anti-endotoxin
based treatments of sepsis. Preclinical studies with these synthetic peptides demonstrate
that they can combine excellent selectivity for LPS binding together with suppression
of LPS-induced cytokine release in vitro and protection from lethal LPS induced septic
shock in vivo [723-725]. Interestingly, the molecular interaction between these peptides
and LPS which results in this neutralization of biological activity is evident as a biophysical change in the LPS supra-molecular structure. With peptide binding, the lipid-A
part of LPS is converted from its ―endotoxic‖ conformation, being the cubic aggregate
structure, to an inactive multi-lamellar structure. Peptides that bind to lipid-A have
direct anti-bacterial activity as a consequence of the key structural role of lipid-A
-18-
Chapter 2: Introduction & overview
within the bacterial membrane structure. This binding affinity is reflected in the
similarity between the mean inhibitory concentration (MIC) and the concentration
required to achieve anti-LPS activity with these peptides [723, 725].
Table 2.1.3 Partial classification of lipid-A structures.
Hexaacyl lipid A structures
Mucosal habitat
Escherichia coli
Klebsiella pneumoniae
Enterobacter cloacae
Serratia marcescens
Proteus mirabilis
Providencia rettgeri
Shigella sonnei, S. flexneri
Haemophilus influenzae
Neisseria gonorrhoeae, N.
meningitidis
Bordetella bronchiseptica
Vibrio cholerae O1
Campylobacter sp.
Soil, water habitat
Stenotrophomonas maltophilia
Burkholderia cenocepacia (some
strains)
Nonhexaacyl lipid A structures
Moraxella catarrhalis
Bordetella pertussis
Helicobacter pylori
Chlamydophila trachomatis
Vibrio cholerae O139
Yersinia enterocolitica
Salmonella enterica serovar
Typhimuriume
Vibrio vulnificus
Animal habitat
Brucella abortus
Yersinia pestis
Francisella tularensis
Legionella pneumophila
Capnocytophaga canimorsus
Bartonella henselae, B. quintana
Leptospira interrogans
Pseudomonas aeruginosa
Burkholderia cenocepacia (some
strains)
Burkholderia pseudomallei
Acinetobacter radioresistens
Enterobacter agglomerans
Strict anaerobes
(None)
Bacteroides fragilis
Prevotella intermedia
Porphyromonas gingivalis
Adapted from Munford RS. Sensing Gram-Negative Bacterial Lipopolysaccharides: a
Human Disease Determinant? Infect. Immun. 2008 76 (2) 454-465 [172]
LPS molecular structure activity Whereas remarkably diverse lipid-A
structures are found in the LPSs produced by different gram-negative bacteria, the
requirements for maximal activation of animal cells are rather restricted (Table 2.1.3)
[171-175]. The LPS structure that is optimally sensed by the MD-2-TLR4 receptor is a
bis-phosphorylated di-glucosamine to which are attached six saturated fatty acyl chains
Chapter 2: Introduction & overview
-19-
with lengths of 12 or 14 (occasionally 16) carbons (Fig. 2.4). LPS molecules that have
this structure are commonly referred to as ―hexaacyl LPSs‖ in contra-distinction to
other lipid-A structures which have four, five or seven (i.e. non-―hexaacyl LPSs‖) fatty
acyl chains.
Hexaacyl lipid-A structures are common among Enterobacteriaceae such as E.
coli whereas non-Enterobacteriaceae, such as Pseudomonas aeruginosa have a nonhexaacyl lipid-A structure. The structural differences in lipid-A among clinical GN
bacterial isolates is a recent finding. The possible clinical relevance of these differences
remains to be determined and is one of the research questions of this thesis.
Endotoxin antagonists The structural specificity for LPS has resulted in
structure activity studies leading to the development of endotoxin antagonists which
have undergone pre-clinical evaluation [176]. These antagonists demonstrate ability to
block the endotoxin activities in vitro and in vivo. Several have been found to have
efficacy in preliminary but not in confirmatory clinical trials undertaken in patients
with sepsis (see 2.5.2).
The LPS receptor The mechanisms for lipid-A binding and recognition are part
of the innate immune system and are genetically encoded [177-179]. These
mechanisms have been found in almost all studied vertebrate genomes including
humans. Inbred C3H/HeJ mice are genetically deficient in the LPS receptor mechanism
and as a result are resistant to the effects of endotoxin. Surprisingly, C3H/HeJ mice are
more susceptible to GN infection across a mucosal barrier such as the urinary tract
(Table 2.1.1).
Lipid-A recognition requires both LPS binding (MD-2) and signal transduction
(Toll-like receptor 4; TLR4) mechanisms. The TLR4 is one of ten TLRs known to exist
in humans. Activation of the LPS receptor leads to a range of intra-cellular activities
mediated by protein kinases and activation of nuclear factor kB leading to increased
TNF production [177-181]. Of note, this LPS binding and signal transduction exhibits
specificity for the different lipid-A structures from various sources of GN bacterial
endotoxin (Figure 2.6). The basis for the ligand specificity between hexaacyl lipid-A
and the TLR4-MD-2 complex has been determined by examination of its crystal
structure [175].
Several examples have emerged to illustrate the possible clinical relevance of
-20-
Chapter 2: Introduction & overview
Fig. 2.6. Effect of structural
modifications on lipid A
bioactivity. Figures shown
represent factors by which
structures are reduced in
bioactivity when compared to
complete E. coli lipid A in
relation to modifications to the
hydrophilic section (top).
hydrophobic section
(bottom).
Reproduced from E.T.
Rietschel, T. Kirikae, U. Schade,
U. Mamat, G. Schmidt, H.
Loppnow, A.J. Ulmer, U.
Zahringer, U. Seydel, F. Di
Padova, M. Schreier, H. Brade.
Bacterial endotoxin: molecular
relationships of structure to
activity and function. FASEB J.,
8 (1994), pp. 217–225 [153]
this variation in lipid-A structure toward the pathogenesis of GN infections. One
example is the change in lipid-A structure occurring with Yersinia pestis (plague)
infection. In association with transition from the vector (the flea; body temperature 2027º C) to the human blood stream (body temperature 37º C) there is a change in lipid-A
structure from six fatty acid (acyl) chains to four. It is believed that as a consequence of
this molecular change this enables the Yersinia organism to evade detection by the host
innate immune system [173, 182]. A second example is the variable lipid-A structure
found in isolates of Pseudomonas aeruginosa from early versus late airway infections
in cystic fibrosis (CF) patients and also versus non-CF isolates of Pseudomonas
aeruginosa [183]. The exact relevance of these structural changes in Pseudomonas
lipid-A toward the chronicity of the airway infection in CF is still being elucidated.
Tolerance A biological property of endotoxin that is most unusual for a toxin is
tolerance [184-207]. Doses of endotoxin in experimental animals that would otherwise
be lethal can be tolerated after pre-treatments by sub-lethal doses [186]. In the 1920‘s,
Chapter 2: Introduction & overview
-21-
bacterial preparations containing endotoxin were in therapeutic use to induce a fever as
a therapy for cancers and syphilis in humans [186]. Humans, as with other species, will
acquire tolerance on repeated dosing and require higher (as much as 200-fold) doses to
achieve a fever response equivalent to that achieved with the initial dose. There was a
series of fascinating experiments undertaken initially in rabbits [188] and later in
humans [189-198] to investigate the relevance of tolerance toward GN infections.
In studies undertaken in the 1960‘s using prisoner volunteers who were
rendered tolerant to endotoxin by repeated doses of endotoxin were subsequently
deliberately infected with Salmonella typhi (i.e. typhoid) or Pasteurella tularensis (i.e.
tularemia) it was demonstrated that tolerance to endotoxin could be induced by
administration of Salmonella endotoxin even during the course of an experimental
typhoid illness. Moreover, ―Despite unequivocal activation of the endotoxin tolerance
mechanisms……, the febrile and toxic course of typhoid fever proceeded unabated.”
[192].
This phenomenon has been extensively studied in both animals and in human
volunteers with more than 200 studies cited in four key review articles [185-187, 199].
The literature is confusing given that the key studies have been undertaken over a
period exceeding 50 years during which time the experimental techniques and the state
of endotoxin knowledge have advanced. There are several features of endotoxin
tolerance which are relevant to the research conducted in this thesis as possible
explanations for disparities between endotoxemia detection and the clinical relevance
of endotoxemia:
(i) Tolerance to endotoxin has been demonstrated to be acquired in humans with sepsis
in a variety of settings including pyelonephritis [200], brucellosis [201], melioidosis
[202], tularemia, chronic Salmonella infection [203] and typhoid [204]. While its
relevance to the outcome of sepsis remains unclear [205-207], there are some
features in the cytokine profile of patients with sepsis that resemble the profiles
observed following the induction of endotoxin tolerance [206-207].
(ii) The development of tolerance in humans and experimental animals is bi-phasic in
that there is an early phase lasting hours and a late phase lasting for upwards of
several weeks [196]. The early phase in human volunteers is most apparent with the
continuous intravenous infusion of endotoxin [195]. This phase confers tolerance to
-22-
Chapter 2: Introduction & overview
endotoxins generically as a class but not to other pyrogens [197]. Enhanced
clearance of endotoxin can be demonstrated in this phase although specific
endotoxin antibodies are not involved [187, 194]. This enhanced clearance results in
a ~50% reduction in the half-life of radio-labelled endotoxin in human volunteers
[193]. For early tolerance, the degree of tolerance is proportional to the intensity of
the febrile and subjective symptomatic response to the initial infusion.
(iii)By contrast, late tolerance is largely specific for the homologous endotoxin used
initially to generate tolerance and develops after at least 72 hours. Late tolerance is
believed to be mediated by antibodies to the LPS molecule [197].
(iv) Confusingly, in addition to tolerance, it is possible to demonstrate hypersensitivity
to endotoxin which is manifest either in the systemic or in the local reactivity.
Systemic hypersensitivity can be manifest as the Danysz reaction which occurs if an
infusion of endotoxin to a human volunteer is either interrupted or administered as a
divided dose (e.g. as a daily intravenous administration) [185]. This systemic
hypersensitivity is manifest as enhanced subjective symptomatic and febrile
responses in the period before the onset of tolerance. Hypersensitivity can also be
manifest to endotoxin administered intra-dermally (e.g. forearm). This
hypersensitivity response is the Shwartzmann response [185].
(v) Indeed in human volunteer studies, the onset of typhoid fever is associated with
systemic hyper-reactivity to both homologous (i.e. S typhosa endotoxin) and
heterologous (e.g. Pseudomonas endotoxin) endotoxins.
(vi) Local hypersensitivity to endotoxin may be demonstrated as a dermal reaction to
intra-dermal injections of endotoxin. In human volunteer studies, the onset of
typhoid fever is associated with hyper-reactivity to homologous (i.e. S typhosa
endotoxin) but not to heterologous (e.g. Pseudomonas endotoxin) endotoxins, in
contrast to the systemic hyper-reactivity [192, 196].
(vii)
Surprisingly and confusingly, in the human volunteer studies, dermal
hypersensitivity could be demonstrated concurrently with systemic tolerance.
Hence with these findings in the human volunteer studies of typhoid it was
concluded that given this ―…ability of man to acquire tolerance to endotoxin during the
typhoidal and tularemic illnesses.... circulating endotoxin could not constitute the
major cause of the sustained pyrexia and toxaemia during these illnesses‖ [192].
Chapter 2: Introduction & overview
-23-
These investigators speculated on two possible mechanisms by which
endotoxemia could relate to the manifestation of typhoid fever [208-209]. Firstly,
release of episodic endotoxemia in ‗spikes‘ could overcome the tolerance. Alternately,
the mechanisms underlying the systemic tolerance and the local tissue hyper-reactivity
are dissociated as noted in the prisoner volunteer typhoid infection studies.
These observations on tolerance were considered to be so remarkable for a
bacterial toxin that ―….it is difficult to believe that it has no clinical significance‖ [192].
Recent studies in the past decade have attempted to define the mediation of tolerance at
the molecular level in an effort to further clarify the clinical relevance of this
phenomenon [207].
2.2. Assays for endotoxin
Among the various assays available for the detection of endotoxin [210-221]
described above, the Limulus amoebocyte lysate (LAL, limulus) assay (described
below) has the most published clinical experience having been available for nearly 40
years [210-213]. Moreover, LAL preparations are widely used to screen for endotoxin
contamination in pharmaceutical product before commercial release [213] and a
convenient LAL assay is commercially available.
There are other endotoxin assays although these have less published clinical
experience. The rabbit pyrogen assay is cumbersome as it requires live rabbits. There is
some published clinical experience with the rabbit pyrogen assay that pre-dates the
availability of the LAL assay [96-99, 103]. Several newer in vitro assays have been
developed in the past 20 years [214-219] and published clinical experience is emerging
with these assays. One is a whole blood agglutination assay which senses endotoxin
through a dimerization reagent (Simply-RED) [216, 217]. A second is the neutrophil
chemo-luminescence assay (NCL) also known as the Endotoxin Activity Assay (EAA)
[214, 215] which is a whole blood assay based on neutrophil activation by immune
complexes formed between a monoclonal anti-endotoxin antibody and
lipopolysaccharide. These two more recently developed assays have been studied in a
limited number of settings with, to date, six publications available [100, 101, 104-107].
This limited experience with these non-Limulus assays is explored in the analysis in
this thesis as a comparison against the LAL assay (see chapter 5).
-24-
Chapter 2: Introduction & overview
Another assay for endotoxin (also not further considered in this thesis) is an
assay based on recombinant factor C (rFC) produced by genetic engineering methods
using the genetic code for the Limulus factor C. This assay reagent will help to meet the
growing need for pyrogen testing in association with increasing drug production by biotechnology methods [218-219].
2.2.1. Limulus amoebocyte lysate (LAL, limulus) assay
Fig 2.7 An 1873 illustration of
the anatomy of the Atlantic
horseshoe crab, Limulus
polyphemus with the circulation
coloured in red. The LAL lysate
is derived from the granules of
the amebocyte which is the
circulating blood cell of this
crab.
From: ‗Recherches sur
l‘anatomie des Limules‘ by M.
Alph. Milne Edwards. 1873,
Paris
The Limulus amoebocyte lysate (LAL, limulus) assay is derived from the blood
cell (amoebocyte) of the Limulus horseshoe crab (Figure 2.7) and is the most sensitive
and specific test available for the detection of endotoxin. The original studies by Dr
Fred Bang (a marine biologist) in the 1960s who found that the coagulation mechanism
of the horseshoe crab (a sea water crustacean) was triggered by marine living gramnegative bacteria (such as Vibrio species) [211]. Of note, this triggering was specific in
that extracts of gram-positive bacteria did not trigger the coagulation mechanism [210,
221]. The similarities between the Shwartzman reaction (a response of the mammalian
coagulation cascade triggered by endotoxin) and the limulus response to endotoxin
findings were recognized by Dr Jack Levin (a hematologist). Collaboration between
Bang and Levin led to the first application of the limulus assay to test for endotoxin in
Chapter 2: Introduction & overview
-25-
human samples in the early 1970s [53, 54, 212, 221].
There are many similarities between the coagulation mechanism of the
horseshoe crab, from which the limulus assay is derived, and mammalian coagulation
cascade. Both consist of enzymatic cascades. With respect to the LAL assay, this
cascade serves as amplification mechanism which, as a consequence, explains the
responsiveness of the assays to concentrations of endotoxin below ng/ml levels [222].
As a consequence of this extreme sensitivity, the assay needs to be conducted under
exacting pyrogen free conditions to avoid contamination with endotoxin from external
sources [223]. Even the type of plastic containers used for specimen collection is
critical [220, 224].
The assay can be adapted to generate a quantitative result, and it is often
calibrated against a reference endotoxin, usually derived from E. coli (e.g. EC5).
However, it needs to be noted that the assay is a bio-assay in that it quantifies biological
activity. Moreover, this activity is reported in units of weight relative to a reference
endotoxin. The difficulties with the interpretation of a quantitative endotoxin assay and
the comparability of quantitative results between different studies is problematic (as
noted above). The limulus assay as generally available is also reactive to fungal glucans
[225]. A more recent modification of the limulus assay which is claimed to specifically
detect fungal glucan in clinical samples associated invasive fungal infections has been
developed [226-228]. However, cross reactivity with endotoxin associated with gramnegative bacteremias has been observed in the few clinical evaluations of this modified
limulus assay for fungal glucan. This is problematic as some patient groups, such as the
neutropenic, are often concurrently at risk of both GN and fungal systemic infections
[229-230]. However a comparison of five commercially available LAL assays in
common use that were not specifically modified to be fungal glucan specific,
demonstrated that the LAL assay is 1000 times more sensitive to endotoxin than to
fungal glucans [225]. These later types of LAL assay are of the type used within the
studies to be analysed within this thesis.
Summations by Elin In 1979 [231] and 1985 [232], Elin summarized the
published clinical experience with samples other than blood (Table 2.2.1 and figures
2.8/.9) together with blood samples (Table 2.2.2 and figure 2.10). These tables record
the literature as surveyed and abstracted by Elin.
-26-
Chapter 2: Introduction & overview
Clinical samples other than blood The literature relating to the application of
the LAL assay to body fluids other than blood was surveyed and abstracted by Elin
(Table 2.2.1). These published studies report the application of this assay to samples of
urine, CSF and joint fluid to enable the rapid detection of GN urinary tract infections,
GN meningitis and gonococcal arthritis, respectively. Samples of urine, joint fluid and
CSF do not require pre-treatment as is the case for plasma for the control of the assay
inhibitors that are found in plasma.
Table 2.2.1 LAL clinical utility for fluids other than blood.
ref
Author
year
N
TP
FP
%
%
100
100
100
58.8
100
97.4
100
95
100
100
100
0
0
1
14
0
0
1
0
0
1
2
99
1
100
100
66.7
50
44
6
82
23
100
96.6
47.8
94.7
22
16
4
5
79
10
DOR
CSF (see Fig 2.8)
233
234
235
236
237
238
239
240
232
241
242
Berman
Dyson
Jorgensen
McCracken
Nachum
Ross
Tuazon
Clumeck
Schiefele
Jorgensen
Thomas
1976
1976
1975
1976
1973
1975
1977
1977
1981
1978
1983
232
145
134
307
112
335
81
64
59
305
53
Elin summary
1972
145
692
9
453
2975
164
231
1751
1697
291
Synovial fluid (see Fig 2.9)
239
243
244
Tuazon
Elin
Cesario
1977
1978
1983
6
46
69
Elin summary
9
4
24
Urine (see Fig 2.9)
232
232
245
246
Jorgensen
Gray
Nachum
Jorgensen
Elin summary
1973
1978
1981
1982
214
168
223
574
33
99
21
312
Foot notes;
All data except DOR as abstracted in Elin ‘79 &‘85 [231-232]
DOR = diagnostic odds ratio; calculated here by adding 1 to cells
containing zero to avoid undefined results
Chapter 2: Introduction & overview
-27-
csf only
1
Sensitivity
.8
Sensitivity
.6
.4
.2
0
1
.8
.6
.4
Specificity
.2
0
Fig. 2.8 HSROC summary of clinical utility of LAL applied to CSF samples using
data as abstracted in Elin ‘79 &‘85 (from Table 2.2.1) [231, 232]. Note:X axis is
intentionally backwards (as specificity = 1- false positive rate)
urine & joint fluid
1
Sensitivity
Sensitivity
.8
.6
.4
.2
0
1
.8
.6
.4
Specificity
.2
0
Fig. 2.9 HSROC summary of clinical utility of LAL applied to urine and joint fluid
samples using data as abstracted in Elin ‘79 &‘85 (from Table 2.2.1) [231, 232].
Note: X axis is intentionally backwards (as specificity = 1- false positive rate)
-28-
Chapter 2: Introduction & overview
Blood samples Application of the assay to plasma samples to be optimally
prepared to remove inhibitors of the assay. The assay and sample preperation demands
strict pyrogen free and temperature controlled conditions [247-250]. Three common
methods for the preparation of plasma are the method of dilution and heating, treatment
with perchloric acid, or extraction with chloroform. The use of chromogenic substrates
[251] cleaved by the limulus enzymes and the adaptation of the assay to microtitre
plates in the mid 1980s further simplified the assay. The impact of the introduction of
this more sensitive version of the limulus assay on the results of the clinical studies of
sepsis is unknown. This is one of the research questions of this thesis.
Table 2.2.2 LAL clinical utility for blood.
(Data as abstracted in Elin ‘79 &‘85) [231, 232]
ref
Author
Blood (see Fig 2.10)
25
Elin '75
29
Feldman '74
31
Fossard '74
31
Fossard '74
44
Jirillo '75
53
Levin '70
54
Levin '72
56
Magliulo '76
57
Martinez-G '73
64
Oberle '74
75
Stumacher '73
39
Harris
19
Clumeck '77
16
Butler '78
84
Usawattanakul '79
70
Scheifele '81
30
Fink '81
Thomas
40
Harthug '83
Thomas
80
Thomas '84
Elin summary
year
1975
1974
1974
1974
1975
1970
1972
1976
1973
1974
1973
1977
1977
1978
1979
1981
1981
1982
1983
1981
1984
N
237
153
240
260
10
96
278
39
82
23
340
169
48
21
108
212
87
56
5
15
400
TP
FP
%
%
35
5
100
100
50
67
55
75
6
100
43
100
75
0
85
73
100
67
100
100
61
19
0
31
32
50
6
7
43
2
45
16
40
6
0
6
26
2
34
67
0
3
79.5
14.5
DOR
2
8
19
18
1
22
14
3
3
4
4
69
27
0
64
7
914
3
1
44
41
Foot notes;
All data except DOR as abstracted in Elin ‘79 &‘85 [231, 232] and is
on a per sample not per patient basis
DOR; = diagnostic odds ratio, calculated here by adding 1 to cells
containing zero to avoid undefined results
Chapter 2: Introduction & overview
-29-
blood
1
Sensitivity
Sensitivity
.8
.6
.4
.2
0
1
.8
.6
.4
Specificity
.2
0
Fig. 2.10 HSROC summary of clinical utility of LAL applied to blood samples using
data as abstracted in Elin ‘79 &‘85 (from Table 2.2.2) [231, 232]. Note:X axis is
intentionally backwards (as specificity = 1- false positive rate)
Elin was experienced in the clinical applications of the assay [127] and an
earlier clinical study by Elin and co-workers published in 1975 reported that the limulus
assay applied to blood samples was not clinically useful [25]. The summary data as
abstracted by Elin is re-analysed here to illustrate the contrast between the utility for
blood versus for non-blood samples. Note that the studies selected by Elin [231, 232]
were limited to those available pre-1985 versus the number available for analysis in this
thesis (see Chapter 5). Note also that Elin used a simple summation methodology which
differs to the methods used in this thesis (described in chapter 4). Elin had used simple
column totals of the numbers of true positive, false positive, false negative and true
negative in deriving simplistic summary measures of sensitivity and specificity. The reanalysis undertaken here of the data as abstracted by Elin helps to contrast research
methodology available to Elin [231, 232] versus the methodologies used in this thesis
which have become available since 1985.
-30-
Chapter 2: Introduction & overview
Table 2.2.3 Summary of LAL clinical utility
(data as abstracted in Elin ‘79 &‘85 [231, 232] and re-calculated using HSROC
summary methods)
number of
studies
Sens
Spec
%
%
DOR
CSF
Elin summary
11
99
99
HSROC summary
95% CI
11
94
87-97
98
97-99
Elin summary (joint fluid)
Elin summary (urine)
3
4
82
79
77
90
HSROC summary
(joint fluid & urine)
95% CI
7
93
72-98
87
74-94
Elin summary
21
79.5
85.5
HSROC summary
95% CI
21
67
46-82
87
78-93
883
227-3427
Urine & Synovial fluid
88
23-328
Blood
13.4
6.5-28
Foot notes;
Elin summary result is from Elin [231, 232]. Sensitivity, specificity summary
estimates have been derived using column totals
HSROC summary point (see figure) Sensitivity, specificity summary estimates have been
derived using data as abstracted in Elin ‘79 & ‘85 using method as described in chap 4
The following points relevant to the clinical experience of the limulus assay for
blood versus for non-blood samples emerge in the contrast between the analyses of the
data as abstracted and summarized by Elin versus the analysis of the same data using
HSROC summary methodology. The SROC plot (methodology described in chapter 4
& APLM2011) is a concise way of contrasting the results of multiple studies of a
diagnostic test. This methodology was not available to Elin in 1985.
(i) As noted both by Elin [231, 232] and also within the HSROC analysis (Table
2.2.1/.2), for each category of fluid (CSF, urine and joint fluid), the summary
sensitivity and specificity for the limulus assay applied to non-blood samples are
higher than for the limulus assay applied to blood samples (see chapter 5).
(ii) Elin concluded that for fluids other than blood, the LAL assay performs as well as
the Gram stain in terms of speed, sensitivity and specificity.
(iii)It should be noted that the summation method used by Elin [231, 232] gives
Chapter 2: Introduction & overview
-31-
disproportionate weighting to the largest studies and provides no measure of
variability (to be further discussed in 2.6 & 4.2)
(iv) There are three observations in the HSROC summaries (Table 2.2.3) that cannot be
appreciated in the Elin data summary. The first is the magnitude of the difference
between blood versus non-blood samples.
(v) Second is the trade-off between sensitivity and specificity that is apparent across
studies that used different breakpoints to define a positive assay result.
(vi) Also not noted in the Elin data summary [231, 232] is that within each category of
fluid (blood, CSF, urine and joint fluid), at least one study gave results that were
atypical versus other studies in that category. Moreover, these atypical studies were
not necessarily the smaller studies within each category (Figures 2.8/.9/.10).
Hence, the published experience with the limulus assay, whether with respect to
applications to blood (see chapter 5) or to non-blood samples (not further considered in
this thesis) is not uniform amongst different teams of investigators. This important
observation cannot be made using the data aggregation method as used by Elin [231,
232] and requires an HSROC method of summary analysis. Moreover, the HSROC
analysis method raises a new avenue of enquiry toward possible explanations
underlying the different findings of different investigators as a post-hoc analysis.
The analysis of the clinical experience of the LAL assay to blood samples in
relation to the detection of sepsis is the subject of this thesis and over fifty clinical
studies [1-107] are analysed in detail (Chapter 5). As noted by Elin, the inconsistent
detection of endotoxemia amongst patients subsequently found to have GN bacteremia
(low sensitivity) was a particular source of confusion for these investigators. The
conflict is starkly apparent in a cursory reading of the titles of three of the earliest
publications in the series of studies applying the LAL test to the detection of
endotoxemia;
“Gram-negative sepsis: detection of endotoxemia with the limulus test.” [54]
“Lack of clinical usefulness of the limulus test in the diagnosis of endotoxemia.” [25]
“Limitations of the usefulness of the Limulus assay for endotoxin.” [75]
There were three difficult to reconcile observations among these early studies of
the LAL assay in the detection of endotoxemia;
1. The inconsistent detection of endotoxemia amongst patients subsequently found to
have GN bacteremia (false negatives) was interpreted as a lack of sensitivity of the
assay and was a major criticism of the assay. It might have been expected that the
-32-
Chapter 2: Introduction & overview
patient group with GN bacteremia might be expected to be the group to most
reliably test positive for endotoxemia (true positive). However, by this definition,
the proportion of true positives among studies reviewed by Elin mostly range
between 30 and 80%. Note the wide range in true positives among the different
studies with this proportion being as low as 50%.
2. Also problematic was the proportion of those who test positive who turn out not to
have GN bacteremia (false positives). The false positives by this definition mostly
range between 5 and 60% and were as high as 50% among studies reviewed by Elin
[231, 232]. Note that the range in false positives proportions is also wide and also
overlaps the range of true positives. To compound this were findings in several
studies of positive test results among patients with blood stream infections due to
gram-positive bacteremia or fungemia amongst the category of false positive tests.
3. An additional concern was that few studies (with the notable exception of [11, 12,
54]) had found the limulus assay, whether as a quantitative or qualitative result, to
be predictive of patient outcome. These discrepancies were particularly difficult to
explain given the presumption that endotoxemia would be expected to be not only a
marker but also a mediator of the systemic effects of GN bacteremia.
Given these criticisms, several investigators had concluded that the limulus
assay was neither sensitive nor specific in relation to testing blood samples. There was
an (unstated) expectation that a more sensitive assay for endotoxemia was needed.
Subsequent to 1985, a more sensitive assay for endotoxin did emerge. This was the
LAL assay modified by a chromogenic reaction [251, 252].
The possible reasons for the range experiences across the literature are one of
the research questions of this thesis. The disparity between the three initial studies [25,
54, 75] is of particular interest given that these were all large and seemingly wellconducted studies and their disparity has remained unexplained to this day. However,
two major obstacles to a resolution of these disagreements within the general literature
experience are firstly, the lack of an objective gold standard or an assay for
endotoxemia with equivalent sensitivity to that of the LAL assay and, secondly, an
inability to standardize the patient group of interest. Possible approaches to resolving
these obstacles are considered in the next two sections.
Chapter 2: Introduction & overview
-33-
2.3. Sepsis: toward diagnostic and prognostic tests
Sepsis is a clinical syndrome for which defining objective diagnostic and
prognostic tests are lacking and continue to evolve [253-256]. Forty years ago Lewis
Thomas said that ―It is our response that makes the disease‖ [257], a comment which
succinctly describes sepsis as being the body‘s systemic response to suspected or
proven infection rather than the infection per se. The clinical similarity of sepsis as a
syndrome, whatever the causative organism, has been generally noted [258-259]. The
difficulties with current defining criteria and testing for sepsis is best illustrated by a
contrast with the field of cardiovascular medicine where in the last ten years, the
availability of the serum cardiac troponin test, a rapidly available serum test of cardiac
injury, has transformed the management of patients presenting with acute coronary
syndromes [260].
The two broad approaches toward the development of diagnostic and prognostic
tests for sepsis; these are tests based on clinical criteria and tests based on laboratory
detection methods [256]. There are two statistical criteria that are commonly applied in
the description and evaluation of diagnostic and risk prediction tools; discrimination
and calibration [261-264];Discrimination. This is a measure of how well the model separates those patients
who do versus those who do not have the disease or outcome of interest.
Discrimination can be evaluated using an ROC curve, being a plot of true positive
proportions versus false positive proportions as defined over a range of break points
with the model. This measure may be derived from a cohort retrospectively,
commonly as the ‗c‘ statistic being the area under the ROC curve (see also
APLM ’11 amongst appended publications for further discussion).
Calibration. This is a measure of how well predicted probabilities of the disease or
outcome of interest agree with actual observed proportions within sub-groups of a
new cohort (usually studied prospectively). This is usually measured and assessed
using the Hosmer-Lemeshow statistic.
The literature in this area as it relates to the development of diagnostic and
prognostic tests for sepsis is complex. A review of 19 scoring systems that were
available as at 1995 for the prediction of mortality in the patient group with sepsis
concluded that
―The calibration of risk prediction methods comparing predicted with
actual mortality across the breadth of risk for a population of patients is
excellent, but the overall accuracy in individual patient predictions is such that
clinical judgement must remain a major part of decision making‖ [263]
-34-
Chapter 2: Introduction & overview
2.3.1. Tests based on clinical criteria
Diagnostic tests. Several clinically based diagnostic and prognostic criteria
have been considered (Table 2.3.1). In 1991, The American College of Chest
Physicians (ACCP) and the Society of Critical Care Medicine (SCCM) convened a
panel of 35 experts in a consensus conference in an attempt to create a standard set of
definitions to rapidly identify and triage those patients who might benefit from novel
anti-inflammatory therapies that were about to enter clinical trials [265-267]. The
resulting consensus criteria were initially termed the Bone criteria and subsequently
evolved into the Systemic Inflammatory Response Syndrome (SIRS) which consists of
the presence of at least two of the following;
Body temperature >38 ºC or <36 ºC,
Heart rate >90 beats per minute,
Respiratory rate >20 breaths per minute of hyperventilation evident with a
PaCO2 <32 mmHg,
White blood cell count (WCC) >12,000/mm, <4000/mm3 or with >10%
immature neutrophils.
SIRS is termed sepsis if infection or microbial invasion into normally sterile
tissues is documented. The criteria have subsequently evolved [256]. Moreover, to
enable the grading of severity of SIRS, the ACCP/SCCM conference defined by
consensus the Multiple Organ Dysfunction Syndrome (MODS) as ―the presence of
altered organ dysfunction in an acutely ill patient such that homeostasis cannot be
achieved without intervention‖ [268].
The main advantage of SIRS and related criteria are their relative simplicity
which led to widespread clinical application. Sepsis, severe sepsis and septic shock are
common in the ICU setting with each occurring in as many as 10% of ICU admissions
[269-274]. However, hospital wide, about 30% of cases are identified in patients
outside of the ICU [274]. In the CUB-rea database, the excess risk of death associated
was 54% in those with versus 28% in matched controls without septic shock giving an
estimated attributable risk of 26% [270]. However, these criteria have four limitations
relevant to the research question of this thesis. These limitations being the criteria are;
too sensitive, as non-infection related conditions are included [254],
not specific as to type of infection; bacterial, whether GN or gram-positive,
fungal or other [258-259],
Chapter 2: Introduction & overview
-35-
being a reflection of the host response, SIRS has an associated morbidity
and mortality which is similar whether or not the SIRS is associated with a
documented infection [272, 273].
As many as a third of episodes are found to occur outside of the ICU setting
[274].
Of particular note, the major limiting factor with the sepsis criteria, as originally
defined, are non-specificity. In one evaluation of 1506 patients who satisfied the criteria
within an urban tertiary care hospital in 1989 and not limited to an ICU population,
only 40 (3%) had GN bacteremia, the sub-group thought to be most representative of
patients likely to benefit from an anti-endotoxin therapy were one to become available
[275].
As of 1995, there were already at least 19 scoring systems that could be applied
to the patient with sepsis or at risk of sepsis in the ICU setting [263]. It should be noted
that the accuracy of any given scoring system needs to be re-established in settings (e.g.
ICU versus non-ICU) or even institutions different to that in which the scoring system
was developed (calibration). A recent study in a general medical (non-ICU) setting
found that other scoring systems displayed better discrimination than a score derived
using the sepsis stages (Table 2.3.1) [285]. The optimal diagnostic test remains elusive.
Other approaches to developing risk prediction tools have looked at more complex
scoring systems (SOFA, MODS) some of which are proprietary (APCHEII).
A concept which has emerged since 2001 is the PIRO grading system (P Predisposing factors; I - Infection; R - Response; O - Organ dysfunction) [286-291].
Slightly different PIRO criteria have been used in different studies. The PIRO grading
system is intended for the stratification of patients of investigational studies into subgroups having similar prognosis, in a manner similar to the TNM (Tumour, Node,
Metastasis) staging system as used to describe oncologic disease. PIRO is not intended
for the rapid identification or triaging purposes.
Limited evidence is emerging of the application of PIRO based scoring methods
in large (>1000 patient) populations with demonstrated predictive performance
comparable to other predictive scores in common use such as the APACHE II and Pitt
bacteremia scoring systems [289-291].
Prognostic tests. There are multiple scoring systems including over 20 scoring
systems that have been developed for the patient in ICU [262-263]. Several have been
-36-
Chapter 2: Introduction & overview
developed from a range of perspectives and are not specific to sepsis. In general, the
scoring systems are one of four general types; generic (e.g. APACHEII, SAPS),
specific to organ system failure (e.g. MSOF, SOFA), specific to populations (e.g.
trauma), and others.
Table 2.3.1 Sepsis; predictive and prognostic tests based on clinical criteria.
Author
(yr)
Diagnostic
Bates
(1990)
Mozes
(1993)
Bates
(1997)
Peduzzi
(1992)
Bates
(1997)
Schechner
(2009)
Prognostic
Sprung
(1990)
Knaus
(1995)
Marshall
(1995)
Bossink
(1998)
Rhee
(2009)
GhanemZoubi
(2011)
GhanemZoubi
(2011)
Criteria
End point
276
Clinical
Bacteremia
Hospitalized 1007
0.72
277
Clinical
Bacteremia
Hospitalized
474
NS
278
Clinical
Bacteremia
0.69
279
Clinical
GNB
276
Clinical
GNB
280
Clinical
Pseudomonas
GNB
sepsis
881
syndrome
Sepsis
465
VAMC
sepsis
881
syndrome
Hospitalized 4114
281
Encephalopathy
Mortality
282
SUPPORT
Mortality
268
MODS
Mortality
ICU
336
0.936
283
SIRS
(modified)
Pitt score
Mortality
Hospitalized
300
Mortality
ICU
134
better than
SIRS alone
0.799
MEWS
REMS
MEDS
SCS
Sepsis
stages
Mortality
Hospitalized 1072
0.70-0.79
Mortality
Hospitalized 1072
0.65
284
285
285
Population
N
Discrimination
Ref
Sepsis
1333
(ICU)
Hospitalized 4301
Foot notes;
Discrimination = DOR unless otherwise specified
VAMC = Veterans Administration Medical Center
NS = not stated
‗significant‘
0.70
0.73
OR = 2.2
0.79
Chapter 2: Introduction & overview
-37-
Controversially, these scoring systems have been used to guide therapy for
individual patients. For example, the APACHEII scoring system was used in targeting
therapy with recombinant activate protein C (rhAPC) to the patient group at high risk as
indicated by an APACHEII score >24 [293]. Of note, rhAPC was recently withdrawn
after a failed attempt to confirm efficacy of this drug in patients with septic shock [300].
The clinical application of disease severity scores toward the management of individual
patients is controversial as they are instruments which are validated for use in
populations, not for individual patients [297]. However, a research question relevant to
the analysis to be undertaken in this thesis is whether the mortality risk of the
population determines the relative importance of various risk factors and the efficacy of
anti-sepsis agents in that population [294 -296, 298, 301, 302].
2.3.2. Laboratory based tests
The lack of an objective ‗gold standard‘ for sepsis testing is an impediment
toward progress in developing and evaluating diagnostic tests generally, not only the
limulus assay. All assays developed to enable the rapid recognition of the presence,
severity and the specific microbial causes of sepsis have faced the same lack of a
reference standard. For example a recent review of 170 potential biomarkers for sepsis
(other than the limulus assay for endotoxin) that have been evaluated, none was
sufficiently practical, sensitive or specific for routine clinical use [303]. Moreover,
relatively few had progressed to be evaluated in clinical studies. It also needs to be
recognized that there is evidence for a strong publication bias for prognostic markers.
For example, amongst published studies of markers used in oncology, it was recently
identified that those that have statistically significant results are disproportionately
published versus ones that do not [304]. This ‗reporting bias‘ is likely widespread.
Diagnostic tests. Table 2.3.2 lists five recent representative meta-analyses of
clinical evaluations of diagnostic tests in patient groups similar to those in which the
limulus assay has been applied. The diagnostic odds ratio (DOR) is used as a single
indicator of diagnostic test performance from each meta-analysis to enable a
comparison across a broad literature. This table illustrates three points. Firstly, the table
summarizes >100 publications. Second, while the DOR‘s for all of these diagnostic
tests were >7, only one was considered by the respective authors to be useful for
-38-
Chapter 2: Introduction & overview
Table 2.3.2 Recent meta-analytic summaries of diagnostic tests.
Pfeiffer a
Tang
Jones
Karageogopolous
Avni
2006
2007
2007
2011
2011
[Ref]
[305]
[306]
[307]
[308]
[309]
Test
Galactomannan
PCT
PCT
β-d-glucan
PCR
based
Invasive
Aspergillosis
Sepsis
Bacteremia
IFI
Candidiasis
Oncology
ICU
ED
ICU/
oncology
not stated
27
18
17
16
49
19952005
19992005
19992004
19952010
19932009
3260
2097
2008
2979
4694
Main author
Year
Reference
standard
Patient
population or
setting
Number of
studies
Publication
years
Number of
patients
Meta-analytic summary
Diagnostic Odds
Ratio
15.5;
8.3 – 28.8
7.8;
5.9 – 10.4
9.9;
5.7 – 17.0
19.2;
10.5– 35.4
229;
85 – 617
Clinical utility
of test?
selected
groups
No
No
Has
limitations
Yes
Foot notes; PCT = procalcitonin; IFI = Invasive Fungal Infection; PCR =
polymerase chain reaction; ICU = Intensive Care Unit; ED = Emergency
department.
a. Diagnostic odds ratio for Pfeiffer [305] calculated as the exponentiation of mean
D (mean log odds ratio = 2.74; 2.12-3.36); The Q* for this meta-analysis
was 0.80; 0.74-0.86.
individual patient management. This illustrates that even with markers or risk factors
that might be considered to be strong markers (OR‘s greater than 5) for an end point in
a population by epidemiological criteria, they are not necessarily sufficiently
discriminatory to be clinically useful for application as diagnostic tests for individual
patients [310-312].
The third point is that each of these meta-analyses has incorporated a broad
range of studies. Different studies have evaluated assays variants, different breakpoints
and different patient groups. Given the heterogeneity in populations to which a test for
Chapter 2: Introduction & overview
-39-
sepsis might be applied, together with the diverse published experience, it needs to be
questioned whether a single summary result from the literature is either achievable or
desirable for any diagnostic test for sepsis. A highly relevant finding from multiple
evaluations with implications toward the potential real world application of any
diagnostic test is the extent of variability in the DOR amongst different studies and
different populations in which the test may be used [313-314].
Prognostic tests. The difficulties in the development of prognostic tests for
sepsis is again well illustrated by a contrast with the field of cardiovascular medicine
where the availability of a test for cardiac troponin, a rapidly available serum test of
cardiac injury transformed the management of patients presenting with acute coronary
syndromes. The prognostic values of cardiac troponins measurement in serum was
established in a sub-study of the Global Use of Strategies To Open occluded arteries
Trial IV (GUSTO-IV) which found that the baseline level of troponin T was
independently related to 30-day mortality [260]. This finding, published in 2003,
prompted the development of more aggressive interventional strategies for cardiac
patients determined by the baseline troponin assay. In the field of sepsis medicine, there
is as yet no simple and unifying test to enable patient risk stratification. Note that the
size of the GUSTO-IV study, 7000 patients [260], is more than the sum of patients
included in studies of sepsis tabulated in the meta-analyses listed in Table 2.3.2 and
also among the studies analysed within this thesis (see Table 5.1.1).
Amongst the various laboratory tests that have been evaluated for sepsis, the test
which has been most studied is the serum procalcitonin (PCT) assay. However, even
with this well-studied assay, its clinical application remains unclear. One recent metaanalysis of 18 studies of PCT as a predictor of bacteremia in critically ill patients [306]
found evidence that diagnostic performance was less apparent in larger versus smaller
studies. The PCT assay has been considered for use as a guide to de-escalate antibiotic
therapy [315]. A multi-center trial to reduce antibiotic use in intensive care units
through PCT monitoring (the PRORATA trial), found that this strategy could reduce
antibiotic use without apparent adverse outcome [316].
The search for clinically relevant diagnostic and prognostic tests for sepsis
continues [317-319] and the most recent approaches have examined gene arrays [320],
molecular markers [321] and cytokine profiles [322]. These approaches are challenging
not least because of the statistical methods required to evaluate the high dimensional
-40-
Chapter 2: Introduction & overview
data that is generated [323].
For example, a recent study of 126 patients presenting with sepsis in the setting
of an emergency department measured a panel of cytokines simultaneously using a
multiplex assay with cytokine patterns determined by principal component analysis
(PCA) and agglomerative hierarchical clustering (AHC) [322]. This study, as in
previous studies, found some associations between individual bio-markers with severe
sepsis or septic shock. However, none of these associations remained significant in
PCA or AHC analysis which examined for signature patterns of cytokine profiles [322].
Recently, an assay to detect cell-free DNA might have strong prognostic value in
patients with severe sepsis in some studies [324] but not all [333, see below]. However,
a systematic review of studies of 12 patient cohorts found that sepsis-related
inflammatory changes are highly variable on a transcriptional level [325]. In particular,
the ‖…arbitrary distinction of separating sepsis into pro-inflammatory and antiinflammatory phases is not supported by gene-expression data. [325]‖
2.3.3. The potential for the limulus assay
The limulus assay has four unique advantages over other diagnostic and
prognostic tests that have been evaluated in the setting of sepsis.
(i) The signal amplification that results from the coagulation cascade mechanism
within the limulus assay enables sensitivity to pg/ml levels of endotoxin, several log
fold higher than might be achieved with other assays [210].
(ii) The assay target is endotoxin, a molecule which is considered to be a mediator of
the adverse effects of gram-negative sepsis [222].
(iii)In at least one setting, that of meningococcemia, there is clear evidence that the
levels of endotoxin in blood are both diagnostic and prognostic [11-12].
(iv) The assay been validated in application to samples other than blood (Table 2.2.1).
Strikingly however, these implied advantages are not generally apparent
amongst the surveys of clinical applications of the limulus assay to blood samples. The
clinical experience of the limulus assay as summarized by Elin (Table 2.2.2) indicates
that the limulus assay applied to blood samples performs poorly compared to its
application to non-blood samples. Moreover, the summary clinical experience of the
limulus assay as applied to blood (as summarized by Elin; Table 2.2.2) appears no
better than that with the clinical experience of a selection of non-limulus assays (Table
Chapter 2: Introduction & overview
-41-
2.3.2). Moreover, the prognostic value of the limulus assay is also highly variable
among the studies that have examined this (see 2.5.3). Possible explanations for these
disconnects could be either the patient population or the use of GN bacteremia as the
reference standard. These two possible explanations are considered in the following
section. A third possible consideration is that endotoxemia levels are dynamic, for
example in one clinical study [21] of 100 patients with sepsis in an ICU setting, the
cumulative percent found to have endotoxemia rose from 20% to 40 % between 0 and
24 hours after study entry.
2.4. Gram-negative (GN) bacteremia
In this thesis, GN bacteremia is used as a reference standard against which
endotoxemia detection is compared. Also GN bacteremia is used to assist in the
interpretation of endotoxemia detection as a prognostic test. However, current methods
for the detection of bacteremia may be imperfect [326-327]. Moreover, the use of GN
bacteremia as a proxy reference standard for endotoxemia is controversial and it
potentially raises four major difficulties.
Firstly, the LAL assay is an assay for endotoxin, not gram-negative bacteria per
se. For example, is the amount of cell bound endotoxin sufficient to be detected in the
LAL assay at the levels of GN bacteremia typically found? Lipid-A has a molecular
weight of 4500 and there is approximately 3.4x106 lipid-A residues per E. coli cell
[328-329]: given this, with a typical bacteremia of 10 circulating bacteria per 1 ml of
blood [330-332], it would be expected that there would be 10-12 g (1 pg/ml = 10-12 g/ml)
of endotoxin for a bacteremia with E. coli. Direct measurements to estimate the amount
of LPS per bacterial cell are between ~5x10-3 fg per E. coli bacterial cell [125] and up
to ~25 fg/cfu of endotoxin per E. coli and Pseudomonas bacterial cell [124]. Given the
above assumptions, and assuming that the total amount of bacterial cell bound
endotoxin is completely available, the amount would still be below the detection limit
of even the most sensitive LAL assay which is generally 5 pg/ml. Three further
complicating considerations in this estimation; the limulus assay is at least 10-fold
more sensitive to LPS molecules in aggregates rather than in monomer formations
[128]; there is a complex interactive effect of plasma on the assay [145-147, 250]; and
this estimation of endotoxemia detection takes no account of non-viable bacterial cells
-42-
Chapter 2: Introduction & overview
and cell fragments that accompany a GN bacteremia [326, 333, 334].
In support of the above estimates, there are several experimental rabbit [335,
336] and canine [337] models of GN bacteremia with either an E. coli or Pasteurella
bacterial challenge in which the quantitative blood levels of GN bacteremia and
endotoxemia have been concurrently measured. In these models, GN bacteremia at
levels of 10,000 cfu/ml corresponds to endotoxemia levels of 10 ng/ml [336], 500
ng/ml [335] and 50 EU/ml (~5ng/ml) [337]. Note however, that these levels of
bacteremia are some 1000 times higher than those seen clinically in infections in human.
The second major difficulty is that GN bacteremia is not a single entity (see
section 2.4.1). Amongst various GN bacteremia types there is a well defined ranking of
mortality risk with the risk being lowest for E. coli and highest for GN bacteremias
such as Pseudomonas aeruginosa (see section 2.4.2). Hence, in this light, the recently
described differences in the lipid-A structures (see 2.1.1) among the different GN
bacteremia types provide an impetus for studying endotoxemia detection and as may
serve as a basis for exploring the differences in mortality for the GN bacteremia types.
Could this be a potential confounder in the relationship between GN bacteremia and
both the detection of endotoxemia and the occurrence of mortality? This is one of the
research questions of this thesis.
The third difficulty in using GN bacteremia as a proxy reference standard is that
the mortality attributable to GN bacteremia is itself difficult to define. This issue is
discussed further in 2.4.1.
Finally, GN bacteremia accounts for only a minority of patients with sepsis
syndrome (Figure 2.11). On the other hand and to counter the above difficulties, the
following are reasons that GN bacteremia can serve as a valid proxy reference standard;
(i) GN bacteremia would appear to be the most clinically relevant proxy for
endotoxemia,
(ii) The diagnostic and prognostic tests for sepsis are problematic and while
they continue to evolve, GN bacteremia serves as a population marker,
(iii)GN bacteremia is itself an objective and important clinical end point,
(iv) GN bacteremia has been extensively studied in settings such as in the ICU
and elsewhere in the broader clinical literature enabling a derivation of a
benchmark range (see section 2.4.2) for mortality risk with GN bacteremia,
(v) Methods to enable the rapid detection of GN bacteremia would help to
target specific patient management strategies,
Chapter 2: Introduction & overview
-43-
(vi) in fluid other than blood, such as urine & CSF, there is a quantitative
correlation between levels of GN bacteria and endotoxin,
(vii) there is insufficient experience with the other assays for endotoxin for
the purposes of validating the results of the limulus assay,
(viii) The detection of GN bacteremia has commonly been used as a proxy
marker for comparisons with the results of the limulus assay [1-107],
(ix) GN bacteremia is common, especially in the ICU setting,
(x) Along with an expectation of concordance with detection, there is an
expectation of correlation with level of bacteremia. In this regard, there is
some evidence that the quantitative level of bacteremia correlates with
outcome (see below) [340, 341].
Fig 2.11 Overlap of serious GN
infections amongst 1200 patients
having blood cultures taken for
suspected bacteremia. From Bates et
al. Projected imapct of monoclonal
anti-endotoxin antibody therapy. Arch
Intern Med. 1994;154:1241-9 [275].
GN infection
(n=298)
7
35
GN bacteremia
(n=40)
5
22
Sepsis syndrome
(n=34)
2.4.1. GN bacteremia types
There are two broad categories of gram-negative infections; pathogenic and
opportunistic. The limulus assay has been studies in settings of infections with both of
these categories of gram-negative infections although the findings have not been
uniform. The opportunistic gram-negative infections have been most studied among the
107 studies that have been analysed within this thesis.
Pathogenic. On the one hand are infections due to specific pathogenic gramnegative bacteria as exemplified by Salmonella typhi [1, 16, 56, 77], Neisseria
meningitidis [9, 11-12, 40, 68], Burkholderia pseudomallei [73] and Yersina pestis [1415] which are the causative organisms of Typhoid fever, Meningococcemia,
Melioidosis and Plague, respectively. These pathogenic bacteria have defined
-44-
Chapter 2: Introduction & overview
pathogenesis mechanisms. These infections are usually community acquired by
previously healthy individuals, often occurring in outbreaks. Some outbreaks may be
attributable to a point source or a single clone. Indeed, these infections are each
clinically distinctive. For example, the recognition of outbreaks such as Typhoid fever
and Plague in the era pre-dating the germ theory of disease in the late 1800‘s led to the
formulation of Koch‘s postulates.
In relation to pathogenic GN bacteria, one of the most extensively studied is
meningococcal infection. In this condition the level of endotoxemia is quantitatively
related to prognosis [11, 12]. In addition to endotoxin and viable bacteria there are high
levels of bacterial fragments in the blood stream [334]. The studies of endotoxin
tolerance in the context of typhoid fever pathogenesis in human prisoner volunteers
have also been mentioned above (see 2.1.1). However, there are no endotoxemia levels
from these studies.
Opportunistic. On the other hand are GN bacteremias as commonly found
within the contemporary hospital context. Common examples are Escherichia coli,
Klebsiella pneumonia, and Pseudomonas aeruginosa. As these infections are usually
acquired by hospitalized patients whose host defences are compromised by specific
defects (e.g. neutropenia, invasive devices, urinary tract abnormalities) or by the
severity of their underlying disease (e.g. burns, malignant disease, diabetes), these
bacteria are often considered opportunistic and hence, relatively non-pathogenic [361].
These types of GN bacteremias are common in the contemporary hospital context
presumably as a consequence of patient resuscitation and supportive therapy which
have prolonged the survival of the compromised patient. This supportive therapy
includes the use of invasive devices (e.g. urinary catheters) which breach host defence
mechanism and predispose to subsequent GN bacteremia.
In contrast to the category of pathogenic bacteria considered above, the clinical
prediction (i.e. recognition ahead of the results of blood culture testing) of opportunistic
GN bacteremias in the contemporary hospital context is challenging as the clinical
features are non-specific and are difficult to distinguish from non-bacteremic GN
infections and indeed severe infections due to other micro-organisms such as grampositive or even severe viral infections. Several attempts have been made to identify
those who have GN bacteremia among those patients at high risk. This cannot be
readily achieved with current methods. For example, a predictive model for GN
Chapter 2: Introduction & overview
-45-
bacteremia developed among patients presenting to a tertiary hospital (Table 2.3.1) was
found to have poor discrimination with an AUC of 0.726 [270]. In this regard, an
ability to detect specific types of GN bacterial infection would assist in the targeting of
specific antibiotic therapies. This is especially so in settings where GN infections are
common and clinically important. For example, a test for the rapid detection of
Pseudomonas bacteremia in the ICU and oncology setting would be desirable [362].
2.4.2. Pathophysiology of GN bacteremia
How much does the presence of gram-negative bacteria (or indeed any bacteria)
in the blood stream influence the outcome of sepsis? The answer to the question is
surprisingly elusive [361-369]. The attempts to address this question have implications
for how to approach the same question within this thesis for endotoxemia.
There is no question that bacteremic infections with pathogenic GN bacteria
occurring in an outbreak setting in previously health people, as for example with
Neisseria meningitidis, have high attributable mortality. Indeed even community
acquired GN bacteremias (90% enterobacteriaceae) are associated with an adjusted RR
of 1.5 (1.2-2.0) for mortality versus patients with negative blood cultures [370].
On the other hand, for bacteremic infections with opportunistic GN bacteria in
the intensive care unit for example, the mortality attributable to the bacteremia versus
that due to the underlying illness is difficult to define. Viewing this from an alternate
perspective, among patients with systemic inflammatory response syndrome (SIRS),
the morbidity and mortality is similar whether or not the SIRS is associated with
documented infection, being a documented GN infection or another type of infection
[273, 371, 372]. This raises the presumption that for the patient group with SIRS
without a documented infection there is a missing mediator(s) which accounts for the
similar prognosis between those with, versus those without, a documented infection.
Endotoxin and endotoxemia has long been suspected to have this role. However, the
evidence for these ‗missing mediators' being endotoxemia is better established for
pathogenic Neisseria meningitides. This specific infection is cited as exemplorary of
‗intra-vascular‘ versus ‗extra-vascular‘ sepsis (see below) [373].
In attempting to estimate the attributable-mortality associated with bacteremia,
there are at least three inter-related factors to consider; the source of the GN bacteremia,
-46-
Chapter 2: Introduction & overview
the type of the GN bacteremia and the method of statistical adjustment in matching for
severity of underlying patient illness in deriving the estimate.
Source of the GN bacteremia. The site of origin of the GN bacteremia is an
important prognostic determinant. In particular, most influential is a site of origin being
non-urinary tract versus urinary tract [374-376]. Among E. coli bacteremias, those of
urinary tract origin have a better prognosis versus E. coli bacteremias originating from
non-urinary tract sites such as an intra-abdominal or pulmonary sites [374-376]. In this
regard, bacteremias with E. coli more commonly originate from urinary tract sites than
do bacteremias with Pseudomonas and GN bacteria other than E. coli [230].
Meningitis is a special case of an infection occurring at an extra-vascular site for
which there is some evidence that disease severity is proportional to the CSF
concentration of endotoxin [377]. However, this relationship is likely to be
compounded by the correlation between the concentration of bacteria in CSF and
outcome [378].
From an alternate perspective, does bacteremia increase mortality for an extravascular site of infection? This question has been most commonly investigated in
relation to pneumonia acquired in the intensive care unit. In a meta-analysis [379] of
five studies, the presence of bacteremia (not necessarily GN) is associated with an
increased risk of mortality with an OR of 2.07 (95% CI 1.16-3.7) [380-386]. However,
this question remains open as other large studies of pneumonia acquired in the intensive
care unit (and not all included in the meta-analysis [379]) either have [380, 407] or
have not [385, 386] been able to demonstrate a significant increased risk of mortality
associated with the presence of bacteremia.
‘Intra-vascular’ versus ‘extra-vascular’ sepsis. Munford [361] recently
reviewed a range of observations bearing on the role of GN bacteremic infections in the
mediation of sepsis and proposed a hypothesis based on the compartment in which the
the infection and the sepsis process is primarily localized; that is intra-vascular versus
extra-vascular compartments. The prototype of intravascular sepsis is that associated
with a Neisseria meningitidis infection [9, 11-12, 40, 68].
The prototype of extra-vascular sepsis is that due to a primary (e.g. in the
abdominal cavity or lungs) opportunistic infection with Escherichia coli for example.
GN bacteremia if it occurs in association with extra-vascular sepsis is presumed to
Chapter 2: Introduction & overview
-47-
contribute little to inducing the toxic responses. Munford cites a number of
observations to support this intra-vascular versus extra-vascular hypothesis [361].
Munford [361] cites four large case series of severe sepsis or septic shock in
which the difference in case fatality rates for those with versus those without culture
documentation is less than 10 percentage points [272-273, 387, Saez-Lorens ‗95]. For
example, in a survey of admissions to French ICU‘s over a two months period, the
presence of bacteremia was associated with a significantly elevated risk (OR = 1.7; 1.12.8) for early (< 3 days) mortality but this elevated risk was apparent only in a univariate analysis and not in a multi-variate analysis when other relevant prognostic
factors were controlled for [272].
Munford states that the simple presence of bacteria in the intra-vascular
compartment may not in itself be a sufficient trigger for the serious systemic
inflammatory processes in every instance [361]. For example, among observations
bearing on intra-vascular infection, there are several case reports of accidental infusions
of blood products and intravenous fluids later found to be contaminated with GN
bacteria [388-393]. Munford points out these accidental infusions are not invariably
associated with mortality [361].
However, this intra-vascular versus extra-vascular hypothesis may be overly
simplistic. For example, a contrary observation is a recent meta-analysis (not cited by
Munford [361]) of four identical time-series cohort studies of switching from open to
closed infusion containers in four Latin American countries which could be viewed as
an estimate of attributable mortality of bacteremia in the setting of a ‗natural
experiment‘ [394]. This meta-analysis identified that switching to closed infusion
containers was associated with a reduction (RR 0.77; 0.68-0.87) in all cause mortality
of ICU patients in association with a striking reduction (RR 0.33; 0.24-0.46) of
bloodstream infections (not necessarily gram-negative).
The type of the GN bacteremia Among critically ill patients with
nosocomially acquired GN bacteremias [395-400] including either E. coli [397],
Klebsiella [398], Acinetobacter baumannii [400] and Pseudomonas [400] bacteremias,
the difference in mortality between bacteremic case patients versus matched controls
(i.e. attributable mortality) is commonly less than 15 percentage points. However, as
stated above, the impact of the type of GN bacteremia is likely confounded by the
-48-
Chapter 2: Introduction & overview
urinary tract versus non-urinary tract origin.
An additional factor related to the GN bacteremia is the level of bacteremia
(cfu/ml) as previously it had been noted that a higher quantitative level of bacteremia is
related to an increased risk of mortality [340, 401]. Recently, with the continuous
monitoring of blood culture bottles for the detection of growth during their incubation
in modern bacteremia detection methods, the time to positivity (TTP) has come to be
used as a surrogate quantitative marker of bacteremia. Shorter TTP is thought to
indicate higher cfu/ml. In studies of patients with E. coli [402, 403] and Klebsiella
bacteremias [404], shorter TTP was independently related to increased mortality risk.
However, confounding this relationship is that shorter TTP was also related to a nonurinary tract focus [402, 403] and higher Pitt bacteremia score [404].
A multivariate analysis of risk factors for death among a cohort of 832
bacteremic patients with sepsis from a survey of 24 French hospitals revealed that five
factors were found to be independently associated with an increased risk of death
including age, underlying patient illness severity, presence of severe sepsis or shock
whereas a urinary tact source of infection and the type of bacteremia (GN versus nonGN) were both associated with a reduced risk of death [405]. These authors note that
there was a relatively low incidence of difficult to treat bacteremias with Klebsiella and
Pseudomonas in their survey. By contrast, a multivariate analysis of risk factors for
death among 2286 bacteremic patients with sepsis from a survey of 18 Korean hospitals
found that pathogens other than E. coli were associated with an increased risk of death
after adjusting for illness severity and respiratory site of infection [406]. However,
confounding the comparison of these cohorts, Klebsiella and Pseudomonas bacteremias
were more common amongst the Korean cohort than the French cohort [405].
The method of statistical adjustment The execution and interpretation of
studies that attempt to estimate attributable-mortality associated with bacteremia or any
other documented infection is not simple [364, 381]. There are two main methods to
control for severity of underlying illness in attempting to define attributable-mortality.
The first is ‗control by design‘ such as a simple case-control type study design [399,
408]. However control is difficult to achieve by this method given the large number of
potential confounding factors for which matching is sought and this method is not
commonly used. The alternate approach to matching is ‗control by analysis‘ in which
the severity of underlying illness is estimated for each patient using a risk score and the
Chapter 2: Introduction & overview
-49-
score serves as a basis to match case and control group patients for severity of illness
[407]. In the analysis of attributable-risk, the method of risk factor adjustment is crucial
to the estimates obtained and there are at least five difficulties with this approach;
1. the most relevant severity score to choose may be conjectural (see 2.3),
2. the development of a risk score is not simple and factors that determine its
suitability may be unique to the population at hand. An important considertion is
whether allowance was made for interaction effects in the development of any risk
score,
3. the problematic choice of timing at which the severity score is made in relation to
the presence of the documented infection of interest (see below),
4. of all the important factors that need to be controlled to obtain an unbiased estimate,
not all will be known for all patients. For example the site of origin of the
bacteremic infection may not be clear, and
5. there may be additional factors that are presumably influential but remain as yet
unknown. For example, the timing of administration and appropriateness of
antibiotic therapy has emerged in more recent studies as additional significant
factors but has infrequently been recorded in earlier studies [411-413].
The presence of a documented infection could influence the parameters used within
a severity of illness score. This influence introduces an element of confounding and
‗circularity‘ to the matching process which may result in ‗overmatching‘. This has been
circumvented by various techniques including causal analysis [409], daily scoring [410]
and introducing scoring at a time point several days prior (e.g. three day recalibrated
ICU outcome -TRIO) to the bacteremia [368]. An example of the complexity in this
area and how adjustment for risk factors is critical to the results of an analysis are the
seemingly contradictory results obtained from three analyses of data from French ICU‘s.
The OUTCOMERA database is a large multi-center database derived from 12 French
ICU‘s. Two analyses undertaken by the same authors to estimate the impact on
mortality risk of causative pathogen and infection site in different patient populations
over the years 1997-2004 and 1996-2009 had used either a matched conditional logistic
regression adjusted for TRIO score in one [368] versus a multivariate adjusted hazards
model [367] in the other to adjust for underlying and competing mortality risks.
Amongst other findings, the estimate of the impact of bacteremia (not necessarily GN)
acquired in the ICU patient population differed for these two studies being OR 3.02
(2.17-4.22) [368] versus HR 1.59 (1.22-2.06) [367]. Moreover, in the former analysis
[368], the type of blood stream infection was highly influential in mortality risk being
-50-
Chapter 2: Introduction & overview
highest for GN mono-bacterial (OR 6.1; 3.0-12.2) and fungal (OR 8.8; 1.7-46.7) versus
coagulase negative Staphylococci (OR 1.6; 0.74-3.0) whereas is the other analysis
[367] neither the causative organism nor the presence of bacteremia were found to be
associated.
A third analysis of a different multi-center (11 centers) French ICU data base found
that the population attributable fraction of mortality due to blood stream infection (not
necessarily GN) was only 1.7% (0.9-2.5%) but that this estimate was sensitive to
adjustment for any competing attributable mortality fraction for infection occurring at
multiple sites [369].
With the above listed difficulties in mind, it is of interest to examine the mortality
risks of GN bacteremia as reported in the literature. For the data analysed in this thesis,
there are two main settings (ICU versus non-ICU) and three main categories of
opportunistic GN bacteremias (Escherichia coli, Enterobacteriaceae other than E. coli
and non- Enterobacteriaceae; e.g. Pseudomonas species) which are of particular interest.
Both the range in the mortality risks in the literature, and also their relative rankings in
mortality for these categories both within and between studies are of interest here. GN
bacteremia outcome data from 31 studies is summarized for hospital-wide (open; n=15;
Table 2.4.1.1/.2; Figures 2.12/.13), oncology studies (n=5; Table 2.4.1.3; Figure
2.14), and nosocomial and ICU (n=8; Table 2.4.1. 5, Figure 2.15). The studies in these
tables have been obtained by an opportunistic searching of the literature for
publications of large numbers of patients with outcome data for several GN bacteremia
types and it is not meant to be a systematic summary of the literature.
These opportunistic summaries the literature presented in Table 2.4.1.1/.2/.3/.4/.5
and Figures 2.12/.13/.14/.15 illustrate the following points relevant to this thesis.
a. The studies are mostly contemporaneous with the studies of interest [1-107]
within this thesis.
b. GN bacteremia has a high mortality, being up to 50% for some GN
bacteremia types in some studies.
c. Note that the logit scale is used in each figure. With this, the 95%
confidence interval is symmetric about the mortality proportion. The study
precision is reflected in the width of the confidence interval is expected to
be wider for smaller groups.
(continued p 63)
Chapter 2: Introduction & overview
-51-
Table 2.4.1.1 Gram-negative bacteremia: open studies (pre 1975)
Main author
Altemeier
Dupont
Myerowitz
Setia
Year
1971
1969
1971
1977
Ref
[726]
[727]
[338]
[339]
Country
USA
USA
USA
USA
Survey years
1955-67
1958-66
1967-69
‘74 – ‗75
Population
Hospital
Hospital
Hospital
Hospital
Age
Adult & ped
Adult
Adult
Adult & ped
(5% < 20 yrs)
(16% < 39 yrs)
Community:
Nosocomial
NS
18:82
50:50
58:42
census
NS
NS
NS
NS
Mortality
proportions
n
/N
%
n
/N
%
n
/N
%
n
/N
%
Escherichia coli
45
93
48
80
190
42
10
58
17
6
32
19
All non-E. coli
a
Enterobacteriaceae
Pseudomonas species
73
110
66
101
201
50
19
73
26
6
19
32
30
39
77
45
67
37
12
21
57
5
12
42
12
16
75
89
130
68
21
77
27
Anaerobic bacteria
Mixed GN bacteria
b
All GNB
Staph epidermidis
c
Additional pre 1975 studies not shown in the table are;
[Bryant 1971; 728]; surveyed the years 1965-68 for which the numbers and
mortality proportions were Escherichia coli 17/83(21%); all non-E. coli
Enterobacteriaceae 23/76 (30%); Pseudomonas species 32/45 (71%).
[Hodgin 1965; 729]; surveyed the years 1963-64 for which the numbers and
mortality proportions were Escherichia coli 12/32(34%); all non-E. coli
Enterobacteriaceae 15/32 (47%); Pseudomonas species 4/7 (57%).
[Watt 1967; 730]; surveyed the years 1958-65 for which the numbers and mortality
proportions were Escherichia coli 12/30(37%); all non-E. coli Enterobacteriaceae
4/8 (50%); Pseudomonas species 22/26 (85%).
Foot notes;
n = number of fatalities; N = number of patients; % = mortality percentage
a. Includes Serratia species and Proteus species,
b. Includes; Fusobacterium species, Bacteroides species)
c. Staph epidermidis, a gram-positive, serves as a reference group
-52-
Chapter 2: Introduction & overview
Table 2.4.1.1 Gram-negative bacteremia: open studies (post 1975)
Main author
Kreger
Bryan
Roberts
Geerdes
Year
1980
1983
1991
1992
Ref
[340-341]
[342]
[343]
[344]
USA
USA
Canada
Germany
Survey years
‘65 – ‗74
‘77 – ‗81
‘84 – ‗87
‘79-‗89
Population
Hospital
Hospital
Hospital
Hospital
Adult
Country
Adult
Adult & ped
Adult
(5% < 20 yrs)
(~9% < 20 yrs)
(2% < 20 yrs)
26:74
48:52
Not stated
47:53
NS
NS
In hospital
30 day
Age
Community:
Nosocomial
census
Mortality
proportions
n
/N
%
n
/N
%
n
/N
%
n
/N
%
Escherichia coli
37
189
20
160
559
29
45
205
22
36
257
14
All non-E. coli
Enterobacteriaceae a
Pseudomonas species
36
181
20
216
455
47
49
147
33
9
54
17
22
60
37
52
102
51
46
90
51
13
40
33
Anaerobic bacteria b
14
41
34
8
25
32
Mixed GN bacteria
36
105
34
56
101
56
87
202
43
33
75
44
All GNB
151
612
25
434
1186
37
514
31
76
422
18
75
25
22
101
22
Staph epidermidis c
NS
NS
19
Foot notes;
n = number of fatalities; N = number of patients; % = mortality percentage
a. Includes Serratia species and Proteus species,
b. Includes; Fusobacterium species, Bacteroides species)
c. Staph epidermidis, a gram-positive bacteria, serves as a reference group
Chapter 2: Introduction & overview
-53-
Hospital wide pre-1975
species
Hodgin '65
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Watt '67
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Altemeier '67
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Dupont '69
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Bryant '71
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Myerowitz '71
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
10
20 30
50
70 80
90
mortality %
Fig. 2.12 Mortality proportions from hospital-wide studies pre-1975 (as in Table
2.4.1.1). Note the 95% confidence intervals have been calculated and represented
on the logit scale by methods described in section 4.2 (page 87). The vertical lines
at 19% & 43% represent the IQ range for all GNB (see Table 2.4.2) to behcnmark
across Figures 2.11/.12/.13/.14.
-54-
Chapter 2: Introduction & overview
Table 2.4.1.2 Gram-negative bacteremia: open studies (post 1975)
Main author
Leibovici
Weinstein
Garrouste-
Retamar
Orgeas
Year
1998
1997
2000
2012
Ref
[346]
[345]
[347]
[348]
Country
Israel
USA
France
Spain
Survey years
‘88-‗94
‘92-‗93
‘95-‗96
‘06-‗07
Population
Hospital
Hospital
Hospital
Hospital
Age
Adult & ped
Adult
Adult
Adult
39:61
48:52
0:100
341:560
In hospital
30 day
In hospital
14 day
(19% < 19 yrs)
Community:
Nosocomial
census
Mortality
proportions
n
/N
%
n
/N
%
n
/N
%
n
/N
%
Escherichia coli
116
735
16
14
116
12
6
19
32
24
203
12
All non-E. coli
Enterobacteriaceae a
Pseudomonas species
125
512
24
31
125
25
11
22
50
11
83
13
94
275
34
8
48
17
5
12
42
10
32
31
2
14
14
7
18
39
18
8
21
19
123
15
Anaerobic bacteria b
Mixed GN bacteria
105
287
37
27
79
34
0
3
0
All GNB
440
1804
24
58
334
17
22
56
39
Staph epidermidis c
30
166
18
4
73
5
2
12
17
Foot notes;
n = number of fatalities; N = number of patients; % = mortality percentage
d. Includes Serratia species and Proteus species,
e. Includes; Fusobacterium species, Bacteroides species)
f. Staph epidermidis, a gram-positive bacteria, serves as a reference group
Chapter 2: Introduction & overview
-55-
Hospital wide post-1975
species
Kreger '80
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
all
Setia '77
CN Staphylococcus
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
all
Bryan '83
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
all
Roberts '91
CN Staphylococcus
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Geerdes '92
CN Staphylococcus
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
all
Leibovici '98
CN Staphylococcus
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
all
Weinstein '97
CN Staphylococcus
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
all
Garrouste-Orgeas '00
CN Staphylococcus
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
all
Retamar '12
CN Staphylococcus
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
3
5
10
20
30
50
70
mortality %
Fig. 2.13 Mortality proportions from hospital-wide studies post-1975 (as in Table
2.4.1.2). Note the 95% confidence intervals have been calculated and represented
on the logit scale by methods described in section 4.2 (page 87). The vertical lines
at 19% & 43% represent the IQ range for all GNB (see Table 2.4.2) to behcnmark
across Figures 2.11/.12/.13/.14.
-56-
Chapter 2: Introduction & overview
Table 2.4.1.3 Gram-negative bacteremia: intervention studies
Main author
Ziegler
Ziegler
Greenman
Year
1982
1991
1991
Ref
[349]
[350]
[351]
Country
USA
Survey years
1970s
1980s
1980s
Population
Hospital
Hospital
Hospital
Age
Adult & ped
Adult
Adult
not stated
not stated
not stated
30 day
28 day
28 day
Community:
Nosocomial
census
Mortality
proportions
n
/N
%
n
/N
Escherichia coli
70
87
All non-E. coli
Enterobacteriaceae a
Pseudomonas species
40
71
44
23
Anaerobic bacteria
%
n
/N
68
24
b
Mixed GN bacteria
All GNB
60
191
31
77
200
38
171
Staph epidermidis c
Foot notes;
n = number of fatalities; N = number of patients; % = mortality percentage
a. Includes Serratia species and Proteus species,
b. Includes; Fusobacterium species, Bacteroides species)
c. Staph epidermidis, a gram-positive bacteria, serves as a reference group
%
Chapter 2: Introduction & overview
-57-
Table 2.4.1.4 Gram-negative bacteremia: oncology studies
Singer a
Singer a
Year
1977
1977
Ref
[352]
[352]
Country
USA
USA
‘73-‗73
‘73-‗73
Oncology;
Oncology;
hematological
solid tumour
Age
Adult
Adult
% neutropenic
70%
9%
not stated
not stated
Main author
Survey years
Population
census
Mortality
proportions
n
/N
%
n
/N
%
Escherichia coli
9
35
26
6
51
12
All non-E. coli
Enterobacteriaceae b
Pseudomonas species
11
24
46
11
36
31
12
22
55
6
8
75
13
25
56
11
22
50
Anaerobic bacteria c
Mixed GN bacteria
All GNB
Staph epidermidis d
Foot notes;
a. This study (Singer, [352]) is presented for two sub-groups.
b. Includes Serratia species and Proteus species,
c. Includes; Fusobacterium species, Bacteroides species)
d. Staph epidermidis, a gram-positive bacteria, serves as a reference group
-58-
Chapter 2: Introduction & overview
Table 2.4.1.4 Gram-negative bacteremia: oncology studies (continued)
Main author
Ehni
Elting
Gonzalez
Klastersky
Year
1991
1997
1999
2007
Ref
[354]
[355]
[356]
[357]
Country
USA
USA
Spain
MASCC a
(multicenter)
‘85-‗86
‘80-‗93
‘89-‗96
‘95-‗05
oncology
oncology
oncology
oncology
% neutropenic
100%
75%
100%
100%
census
7 days
at cessation
of antibiotic
30 days
NS
Survey years
Population
Age
Mortality
proportions
n
/N
%
n
/N
%
n
/N
%
n
/N
%
Escherichia coli
3
9
33
10
108
9
20
70
29
13
72
18
All non-E. coli
Enterobacteriaceae b
Pseudomonas species
5
19
26
9
112
8
4
17
24
2
20
10
5
17
29
14
90
16
19
48
40
13
42
31
Mixed GN bacteria
11
114
10
15
48
31
5
29
17
All GNB
37
380
10
48
151
32
30
168
18
4
158
3
13
102
13
8
138
6
Anaerobic bacteria c
Staph epidermidis d
NS
Foot notes;
a. MASCC [357] is a multicenter study (multinational association for
supportive care in cancer)
b. Includes Serratia species and Proteus species,
c. Includes; Fusobacterium species, Bacteroides species)
d. Staph epidermidis, a gram-positive bacteria serves as a reference group
Chapter 2: Introduction & overview
-59-
Oncology
species
Singer (hematology) '77
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Singer (solid) '77
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Ehni '91
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Elting '97
CN Staphylococcus
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Gonzalez '99
CN Staphylococcus
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Klastersky '07
CN Staphylococcus
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
1
3
5
10
20 30
50
70 80
90
mortality %
Fig. 2.14 Mortality proportions from oncology studies (as in Table 2.4.1.3). Note the
95% confidence intervals have been calculated and represented on the logit scale
by methods described in section 4.2 (page 87). The vertical lines at 19% & 43%
represent the IQ range for all GNB (see Table 2.4.2) to behcnmark across Figures
2.11/.12/.13/.14.
-60-
Chapter 2: Introduction & overview
Table 2.4.1.5 Gram-negative bacteremia: Nosocomial and ICU studies
Main author
Bryan
Rello
Valles
Cohen a
Year
1984
1994
1997
2004
Ref
[358]
[359]
[360]
[287]
Country
USA
Spain
Spain
Multiple
‘77 – ‗81
‘88- ‗90
1993
‗89-‗99
Nosocomial /
Nosocomial /
ICU
ICU
ICU
ICU
10% < 2 years
Adult
Adult
Adult
NS
NS
In hospital
NS
Survey years
Population
Age
Community:
Nosocomial
census
Mortality
proportions
n
/N
%
n
/N
%
n
/N
%
/N
%
Escherichia coli
10
16
63
5
8
63
3
15
20
3893
19
All non-E. coli
Enterobacteriaceae b
Pseudomonas species
38
57
67
1
11
9
10
44
23
1886
30
19
25
76
7
16
44
13
44
30
1534
32
Anaerobic bacteria c
2
4
50
2
4
50
429
33
1
11
9
21
75
28
4998
18
Mixed GN bacteria
All GNB
72
106
68
18
45
40
50
191
26
Staph epidermidis d
1
2
50
5
21
24
11
144
8
Foot notes;
n = number of fatalities; N = number of patients; % = mortality percentage
a. Data in Cohen [287] is median derived across mean results of between 7 and
10 studies
b. Includes Serratia species and Proteus species,
c. Includes; Fusobacterium species, Bacteroides species)
d. Staph epidermidis, a gram-positive bacteria, serves as a reference group
Chapter 2: Introduction & overview
-61-
nosocomial & ICU
species
Bryan '84
CN Staphylococcus
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Rello '94
CN Staphylococcus
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Valles '97
CN Staphylococcus
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Danner '91
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Guidet '94
Escherichia coli
non-E coli-Enterobacteriaceae
Ketchum '97
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
Opal '00
Escherichia coli
non-E coli-Enterobacteriaceae
non-Enterobacteriaceae
1
3
5
10
20 30
50
70 80
90
mortality %
Fig. 2.15 Mortality proportions from nosocomial & ICU studies (as in Table 2.4.1.5).
Note the 95% confidence intervals have been calculated and represented on the
logit scale by methods described in section 4.2 (page 87). The vertical lines at 19%
& 43% represent the IQ range for all GNB (see Table 2.4.2) to behcnmark across
Figures 2.11/.12/.13/.14.
-62-
Chapter 2: Introduction & overview
Table 2.4.1.5 Gram-negative bacteremia: Nosocomial and ICU studies (continued)
Main author
Danner
Guidet
Ketchum
Opal
Year
1991
1994
1997
2000
Ref
21
38
47, 48, 264
65
USA
France
USA
USA
‗84-‗87
‘92-‗93
‘93 – ‗94
1990s
Adult
Adult
Adult
Adult
14 days
28 days
28 days
28 days
Country
Survey years
Population
Age
Community:
Nosocomial
census
Mortality
proportions
n
/N
%
n /N
%
n /N
%
n
/N
%
Escherichia coli
1
9
11
6
16
37
6
23
26
9
39
23
All non-E. coli
Enterobacteriaceae b
Pseudomonas species
1
5
20
4
6
66
8
27
30
7
33
21
1
2
50
0
0
2
5
40
14
31
45
Anaerobic bacteria c
Mixed GN bacteria
All GNB
Staph epidermidis d
Foot notes;
n = number of fatalities; N = number of patients; % = mortality percentage
a. These studies all have endotoxemia data and are included for analysis in
thesis
b. Includes Serratia species and Proteus species,
c. Includes; Fusobacterium species, Bacteroides species)
d. Staph epidermidis, a gram-positive bacteria, serves as a reference group
Chapter 2: Introduction & overview
-63-
Table 2.4.2 Inter-quartile range of mortality for GN bacteremias amongst 19
studies [339-348, 352-360]
percentile
25th
75th
N*
Median
E. coli
19
20
14
29
non-E. coli Enterobactereiaceae
19
25
17
33
non-Enterobactereiaceae (e.g. P. aeruginosa)
19
40
31
51
Overall§
57
29
19
43
Abbreviations;
* N = number of study groups
§ - Overall is the IQ range for all GNB and is shown in Figures 2.12/.13/.14/.15
d. This non-systematic review provides observed mortality proportions across
the three main categories of opportunistic GN bacteremias and in various
hospital based settings and eras.
e. The observed mortality proportions for the studies published since 1975 (i.e.
as in Table 2.4.1.2/.3/.4 [339-348, 352-360]) can be used to enable
comparisons of mortality risk in the literature versus the studies to be
analysed in this thesis [1-107]. To this end, the inter-quartile range (IQR) of
the mortality proportions for the three GN bacteremia types as reported in
19 studies (20 groups) published since 1975 [339-348, 352-360] is shown in
Table 2.4.2. Each IQ range will be used to benchmark the results obtained
in section 5.2.3.
f. The overall IQ range is indicated in Figures 2.12/.13/.14/.15 and is shown
to enable comparisons across the four figures.
g. The mortality for bacteremias with Staph epidemidis among studies
undertaken in an ICU setting is typically 18% [287]. Note that Staph
epidemidis is not a GN bacteremia but is a common isolate and is used here
for reference purposes only.
h. The largest study [287] in Table 2.4.1.1/.2/.3/.4/.5 is a summation derived
using robust (median) methods from multiple small ICU studies in the
literature. The median mortality risks as reported in this overview [287] are
within each respective IQ range derived here (Table 2.4.2).
i. There are six studies published prior to 1975 [338, 726-730] and these
studies were not used in deriving any IQ range here as these studies predate
the studies of interest [1-107] within this thesis. Note however that the
mortality proportions for patients with GN bacteremia in the pre-1975
studies are generally higher than for studies published after 1975.
Interestingly, the ranking in mortality proportions within the three categories
in these pre-1975 studies is generally maintained in that the mortality for E.
coli bacteremias is the lowest and that for non-enterobacterieaceae including
Pseudomonas species is the highest.
-64-
Chapter 2: Introduction & overview
j. The lower overall mortality since 1975 presumably reflects improvements in
antibiotic and supportive therapies. There is mixed evidence from single
center studies that the mortality risk associated with GN bacteremia has
decreased further with a decrease from 20-30% to 15-20% in one study
[353] whereas for another study unchanged from 16-24% versus 15-19%
[344] for pre versus post 1985 periods, respectively,
k. The mortality risk for poly-microbial GN bacteremias (not shown in Figures
2.12/.13/.14/.15) is higher and broader than that for the three categories of
monomicrobial GN bacteremias. Poly-microbial GN bacteremias are not
further considered in this thesis,
l. The most common GN bacteremia is usually E. coli bacteremias being, in
some studies, >4 times more common than Pseudomonas. However, this is
not universal and in some studies this ranking is reversed.
m. In most studies, the mortality is lowest for E. coli bacteremias with rates in
some studies being similar to mortality rates seen with bacteremias with
Staph epidemidis. In contrast, Pseudomonas bacteremias have a high
mortality risk, sometimes double that for bacteremias with Staph epidemidis.
n. Even within this limited survey, there are several [65, 345, 347, 354-356,
357-360, 727] studies in which the relative rankings of either mortality or
frequency or both were atypical and these atypical studies were not
necessarily the smallest studies. For example, the ICU studies (Table
2.4.1.5) have a more even distribution of the three GN bacteremia types.
o. Even within an oncology study, there can be differences in mix and
mortality risks for sub-groups with hematological versus solid tumour
malignancies [352].
p. Survival curves of patients with GN bacteremia were available for three
single center studies [342, 343, 356]. For two large studies, mortality
following the first positive blood culture was 12% versus 26% at day 5
versus day 30 [343] and 20% versus 30% at day 5 versus day 21 [342]. In a
third smaller series found that 50% of day 30 mortality had already occurred
by day 7 [356]. This observation is relevant to studies of prognosis reported
in this thesis (see 5.2).
q. There are three studies of anti-endotoxin therapies [349-351] tabulated here
(Table 2.4.1.3). These studies were not used in deriving the IQ range shown
here but are shown for comparison and are discussed further in 2.5.2.
r. Among the studies of interest in this thesis [1-107], the four studies with the
largest number of GNB [21, 38, 47, 65] are tabulated here (Table 2.4.1.5 &
Figure 2.15) to enable a comparison with other studies in the literature.
Chapter 2: Introduction & overview
2.5. The endotoxin-sepsis ‘disconnect’
The term ‗sepsis disconnect‘ has been used to describe the failure to translate
anti-endotoxin (and other immuno-modulatory) agents from promising pre-clinical
evidence to proven clinical therapy [416] (this disconnect is explored in 2.5.1). One
expert in the area of endotoxin and sepsis expressed despair at the low return from ―a
quarter of a century of research involving tens of thousands of patients and several
billion research dollars‖ [418]. Likewise, a meta-regression analysis asked why
―despite promising pre-clinical testing and the expenditure of several billion dollars,
anti-inflammatory agents designed to inhibit specific host mediators failed to show
benefit in 22 clinical trials involving over 10,000 patients‖ [298]. This meta-regression
questioned whether variability in the underlying risk of death between preclinical
versus clinical studies could be an explanation for the disconnect. However, there is no
simple explanation for this ‗disconnect‘.
Moreover, there are numerous paradoxical discrepant findings between the preclinical studies on the one hand and the randomized controlled clinical trials of the
same agent on the other (described in 2.5.1) [298, 416-426]. Hence the term
‗endotoxin-sepsis disconnect‘ could be extended to positive results from the early
clinical studies with anti-endotoxin therapies which raised expectations for progress in
the development of these therapies [422]. Failure to replicate these findings in later
studies, also represents a paradoxical ‗disconnect‘ [423, 424] (this disconnect is
explored in 2.5.2). Some additional literature is cited and discussed further in
EOID ’95, Drugs 1994 and Drug Safety ‘95.
The term could be even further extended to describe conflict amongst clinical
observations regarding the prognostic value of endotoxemia detection in various
settings (this disconnect is explored in 2.5.3 and chapter 5.2).
These three examples of ‗disconnect‘ are reviewed here to the extent that they
have a bearing on the analysis of the clinical relevance of endotoxemia detection, GN
bacteremia and outcome in gram-negative sepsis as raised within this thesis. Antiinflammatory agents that have been developed to block mediators (e.g. anti-TNF, antiPAF) whose release is stimulated by endotoxin and hence are ‗down-stream‘ are not
considered here although these also each exemplify a paradoxical disconnect.
-65-
-66-
Chapter 2: Introduction & overview
2.5.1. Anti-endotoxin therapies: Pre-clinical disconnect
Despite 40 years of research with at least six candidate anti-endotoxin therapies
each having shown promise in pre-clinical and early clinical trials, no anti-endotoxin
approach to the treatment of sepsis has emerged with proven clinical benefit. There is
an extensive literature dealing with the pre-clinical and clinical evaluations on this
subject which is complex and difficult to summarize. The extent to which these various
anti-endotoxin therapies bind to endotoxin, neutralize its activity and mediate mortalitybenefit in experimental studies are each contentious [423-426]. For example, a review
article which collates the evidence for and against one extensively studied hypothesis, that antibodies to the inner core raised by immunization with enterobacterial deep
rough mutants confer broad spectrum protection during gram-negative bacterial sepsis,
cites over 200 references published up to 1997 [427]. The most paradoxical suggestion
is that some of the anti-endotoxin monoclonal antibodies and recombinant proteins
under study had been contaminated by endotoxin in the production process [428-429].
This endotoxin contamination from the source bacteria could account for the difficulties
in replicating the protection in later pre-clinical studies.
Another issue that is difficult to bridge between pre-clinical versus clinical
studies is the possibility that anti-endotoxin therapies may be more effective in some
patient sub-groups if only these could be identified [275]. For the purpose of this thesis,
a question of particular interest in this contentious area of research is the evidence
pertaining to the relationship of mortality benefit to reductions in levels of circulating
endotoxemia.
Some of the anti-endotoxin therapies which have been developed are
summarized in Tables 2.5.1 / 2.5.2 & 2.5.3. The agents listed in these tables have antiendotoxin activity which is mediated either through binding to lipid-A or through
blocking the effects of endotoxin presumably at the level of the receptor. However,
even these pre-clinical binding and mechanism of action studies are controversial and
in some cases have been difficult to reproduce [460]. Not shown in these tables is a
range of other biological agents designed to inhibit downstream immunological
mediators whose release is triggered by endotoxin which have been evaluated in
randomized clinical trials of the patient group with sepsis. Recent commentaries [419,
Chapter 2: Introduction & overview
-67-
461] and a systematic review of 45 reviews of animal studies of therapeutic
interventions for sepsis found evidence of over-extrapolation of the preclinical research
findings and an under-appreciation of the deficiencies in study methods [462]. Recently
it has been proposed that the strength of inference from animal studies to the clinical
context could benefit from the application of systematic review methods that address
the risk of bias [462].
The study design aspects of the pre-clinical studies in which anti-endotoxin
efficacy have been evaluated are worthy of a reflection. The following quote from LJ
Berry, a former President of the International Endotoxin Society, describes the complex
challenges in developing appropriate animal model systems in which to test antiendotoxin therapies.
“Few if any biological materials have been the object of as many
scientific investigations as bacterial lipopolysaccharide or endotoxin……..It
poses challenges to chemists, physiologists, immunologists, microbiologists,
pharmacologists, endocrinologists, biochemists, anatomists, or structural
biologists, pathologists, internists, surgeons, infectious disease specialists and
others. A substance with appeal as broad as this is rare in science….In these
introductory remarks, I want to address three issues (a) the question of dosages
that investigators of endotoxin have used, (b) the importance of route of
endotoxin administration in natural and experimental situations, and (c) the
importance of the variation of [animal] species in the type of responses elicited
by endotoxin…….One of the most serious offenses in endotoxin research has
been the use of excessively high doses. Everyone should be cognizant of the
points made here before assuming that observations made under one set of
conditions are typical of the biological effects of endotoxin under all
circumstances.” [LJ Berry, Introduction, in, Handbook of Endotoxin, Vol 3
Cellular biology of endotoxin. pages xvii-xxi, Editor LJ Berry, Elsevier 1985]
[463].
These comments illustrate that in designing experimental models to evaluate
anti-endotoxin agents for clinical use, an optimal animal model needs to be more than
simply an ‗intoxication‘ model [464-466]. Moreover, any beneficial effect would
presumably need to be additional to that provided by supportive and other therapies
such as antibiotics. A complex study design which more closely reflects the sepsis
dynamic in the clinical setting is desirable but difficult to achieve. One widely studied
example is a canine model of septic shock which is described here [337, 466-475].
Controlled models of septic shock have been developed in dogs, in rats and also
in primates by workers at the National Institutes of Health (NIH). These models have
-68-
Chapter 2: Introduction & overview
generated unexpected and paradoxical observations bearing on bacteremia,
endotoxemia and outcome [124-126, 337, 467-475]. In the dog, implantation of an
intra-peritoneal infected clot induces bacteremia with various selected gram-negative
and gram-positive challenge bacteria together with cardiovascular changes
characteristic of septic shock leading to lethality [126, 470-471]. The outcomes of any
therapeutic interventions are studied under controlled conditions simulating a clinical
trial including randomized treatment assignment and investigator blinding [337]. This
model enables the study of the relative effects of intervention with antibiotics,
cardiovascular support and anti-endotoxin therapies on hemodynamic changes, survival
time, quantitative levels of bacteremia as well as quantitative levels of endotoxemia as
measured with the chromogenic LAL assay. A series of paradoxical observations have
emerged from studies with this model and are of interst to the research questions of this
thesis bearing on the concordance or not of bacteremia and endotoxemia and outcome.
1. Firstly, in comparative studies of a relatively avirulent strain of E. coli versus
respectively P. aeruginosa [123], Staphylococcus aureus [126] or a more virulent
strain of E. coli [469], similar quantitative levels of bacteremia were observed
whereas the associated hemodynamic changes and shortened survival time were in
each case more severe versus those observed in association with the avirulent E.
coli strain, as would be expected [123, 126, 469]. Surprisingly, despite these
expected differences, in each case, the levels of endotoxemia were either three
[469]- to ten [123]-fold lower or even undetectable [126] versus the levels seen
after challenge with the avirulent E. coli strain. Interestingly, endotoxin extracted
and purified from the virulent and avirulent E. coli strains were equal with respect
to potency and endotoxin amount per bacterium and hence the dissociation between
endotoxemia and disease severity in this experimental model could not be explained
on this basis [469].
2. Moreover, following intra-peritoneal challenge with strains of E. coli which have
(O6:H1:K2) or do not have (O86:H8) virulence factors for human disease, survival
time was shorter and the associated hemodynamic changes were more severe after
challenge with the virulent strain, as might be expected. However, there are three
paradoxical observations as follows; (1) bacteremia occurred earlier and more
frequently after challenge with the avirulent E. coli strain, (2) levels of endotoxemia
were three-fold higher after challenge with the avirulent E. coli strain and (3)
challenge with heat killed bacteria at a ten-fold higher dose was associated with a
Chapter 2: Introduction & overview
-69-
reversal of the effects on survival and hemodynamic change seen with live bacterial
challenge: the survival was now significantly shortened after challenge with the
killed non-virulent bacteria versus the killed virulent bacteria. Despite this reversal,
the levels of endotoxemia were again three-fold higher after challenge with the
killed avirulent versus the killed virulent E. coli strain [469].
3. A more strikingly paradoxical observation arose when the following putative antiendotoxin interventions were studied; two anti-endotoxin therapies, lipid-X [467]
and a monoclonal anti-endotoxin antibody (HA-1A) [337], reconstituted human
HDL lipoprotein [473], and plasma exchange [472]. These interventions were each
separately studied versus control treatments in this model for the effect on survival
and the associated levels of bacteremia and endotoxemia following the E. coli
challenge. The effects of these interventions on the levels of endotoxemia resulted
in a lowering [473], or no effect [337], or were not measured [467, 472].
Surprisingly, in each case, the therapies either failed to increase survival or even
paradoxically resulted in worsened cardiovascular effects or shortened survival
despite similar quantitative levels of bacteremia versus control treated animals.
4. Studies of the protective effect of various monoclonal antibodies (MAB) against an
E. coli challenge in the canine and other models also generated somewhat
unexpected results bearing on bacteremia and endotoxemia. Both an isotype
matched core region specific MAB and an O-side chain type specific MAB
conferred protection additional to that provided by antibiotic and supportive therapy
in this canine model [468]. The mechanism of protection differed, for the O-side
chain specific MAB being through enhanced clearance of the bacteremia and
endotoxemia whereas the core region specific MAB partially decreased circulatory
collapse associated with the endotoxemia but without any measurable reduction in
the bacteremia or endotoxemia compared to control treated dogs.
5. Studies by this NIH investigator group of the protective effect of recombinant
granulocyte colony stimulating factor (rG-CSF) versus control treatment generated
results bearing on bacteremia and endotoxemia and outcome in a canine pneumonia
model induced by an E. coli intra-tracheal challenge [475]. The prophylactic
administration of rG-CSF reduced levels of endotoxemia and improved survival in
association with enhanced peripheral blood neutrophil counts in this canine model
of E. coli pneumonia compared to control treatment. The quantitative bacteremia
counts were similar in the treatment and control groups.
-70-
Chapter 2: Introduction & overview
Table 2.5.1 Studies of anti-endotoxin interventions : Prophylaxis studies.
Author
year
Pollack
Baumgartner
McCutchan
Cometta
‘72 &
‗76
‗83
‗85
‗83
‗92
Bennett-Guerrero
‗97
McCabe
ref
430 &
431
432
433
434
435
436
Agent
Set
N
core Ab
H
182
core Ab
J5 IVIG
J5 IVIG
J5 IVIG
core Ab
H
43
ICU 262
Onc 100
Surg 329
Surg
301
Mort GNI
↓
ND
↓
↓*
↔
↔
ND
ND
ND
↔
↓
↓
Foot notes;
Mort = mortality. Onc = oncology patient group, Surg = surgical patients group,
GNI = GN infection incidence
ND = No data
↓ = decrease, ↓*= decrease noted in a subgroup, ↔ = no significant change
6. Finally, studies of E5564, a competitive lipid-A antagonist was studied by this NIH
group in an E. coli challenged rat model of sepsis [476]. This model demonstrated
that higher doses of E5564 were required to protect the rats from intra-vascular
versus extra-vascular E. coli challenge. The paradoxical finding of this study was
that despite this protection, E5564 increased the limulus reactivity in plasma.
Table 2.5.2 Studies of anti-endotoxin interventions: Treatment studies.
Author
year
Meningococcal disease
J5 study group
‗92
Derkx
‗99
Levin
‗00
ref
Agent
Set
N
Mort
437
438
439
J5 PC
HA-1A
rBPI21
ICU
ICU
ICU
73
267
892
↔
↔
↓*
Foot notes;
Mort = mortality
ND = No data
↓ = decrease, ↓*= decrease noted in a subgroup, ↔ = no significant change
Chapter 2: Introduction & overview
-71-
Table 2.5.3 Treatment studies of anti-endotoxin interventions (unrestricted).
Author
year
ref
Agent
Set
N
Polyclonal anti-sera
Ziegler
Calandra
Grundmann
Schedel
Behre
Behre
‗82
‗88
‗88
‗91
‗92
‗95
349
440
441
442
4
443
JS PC
J5 PC
IVIG
IVIG
IVIG
IVIG
H
ICU
ICU
H
Onc
Onc
212
71
46
55
21
52
Monoclonal antibodies
Ziegler
Greenman
Fisher
French Reg
McCloskey
Angus
Daifuku
Greenberg
Greenberg
Albertson
‗91
‗91
1990
‗94
‗94
‗00
‗92
‗91
‗92
‗03
350
351
444
445
446
447
448
449
450
451
HA-1A
E5
HA-1A
HA-1A
HA-1A
E5
MAB-T88
E5
E5
ECA-Ab
H
H
Other endotoxin agents
Bion
‗94
Willats
‗95
Reinhart
‗04
‗07
Bennett-Guerrero
Tidswell
‗10
Unpublished
‗12
Dellinger
‗09
Heemskerk
‗09
Pickkers
‗12
Mort GNI
↓
↔
↔
↓
↓
↓
ND
↔
ND
ND
ND
ND
200 §
212
34
600
2199
1090
9
↓*
↓*
ND
↑
↔
↔
ND
ND
ND
ND
ND
ND
ND
ND
39
411
ND
↔
ND
ND
8
SDD
ICU
59
452 Taurolidine ICU 100
453
iHSA
ICU 143
454
E5564
Surg 152
455
E5564
ICU 235
456
E5564
457
PLE
ICU 1379
458
ALP
ICU
36
459
ALP
ICU
36
↔
↔
↔
↔
↔
↔
↔
↔
↔
ND
↔
ND
ND
ND
ND
ND
ND
ND
H
H
ICU
ICU
§
Foot notes;
SDD = selective digestive decontamination, iHSA = human serum albumen,
ALP – Alkaline phosphatise.
Mort = mortality. Onc = oncology patient group, Surg = surgical patients
group, H = hospital wide group, GNI = GN infection incidence
ND = No data
↓ = decrease, ↓*= decrease noted in a subgroup, ↑ = increase, ↔ = no significant
change
§ Two studies are noted here to have a high proportion of non- E. coli
Enterobactericeae relative to E. coli amongst the GN bacteremias;
Schedel [442], of 55 GN bacteremias 30 were non- E. coli
Enterobactericeae and 12 were E. coli.
Ziegler [350], of 200 GN bacteremias 71 were non- E. coli
Enterobactericeae and 87 were E. coli.
-72-
Chapter 2: Introduction & overview
The summation of these paradoxical observations made over the course of these
studies lead this group of investigators to consider whether endotoxemia has continued
importance once the inflammatory response and the clinical manifestations have been
initiated in that possibly…. ―tolerance to endotoxemia may develop within hours
([162]), and persistently high levels of endotoxin may not be central to the
manifestations of the sepsis syndrome‖ [125].
Another research group has also examined the two anti-endotoxin interventions,
bactericidal/permeability increasing protein (BPI) which is an endogenously produced
human endotoxin neutralizing protein, and a monoclonal anti-endotoxin antibody (HA1A) with a lethal E. coli challenge. The plasma endotoxin levels were significantly
reduced compared to no treatment after the use of BPI but not with HA-1A. However,
whilst both treatments attenuated cytokine release, neither improved survival [477].
2.5.2. Anti-endotoxin therapies: clinical disconnect
The impetus for the development of anti-endotoxin therapies had been supported
by a range of clinical observations (Table 2.5.1) [478, 479]. In the 1970‘s, several
clinical studies suggested that improved outcome of patients with sepsis correlated with
the presence in patient serum of high levels of antibodies to core antigens of
lipopolysaccharide as expressed in a mutant E. coli J5 strain [430-432]. These
observations led to studies using various preparations of anti-sera prepared from
volunteer donors specifically immunized using the mutant E. coli J5 strain [433-437].
Anti-sera production is an expensive and impractical undertaking and later studies used
immunoglobulin preparations produced commercially or specific monoclonal
antibodies [349-351, 444-451].
The results of these studies are contentious as there were a number of different
study designs leading to difficulties in interpretation [478, 479]. Various preparations of
anti-sera or immunoglobulins were used, some containing both IgM and IgG, other
preparations only IgG. Some studies used a therapeutic approach and some chose to
target high risk patient groups for prophylaxis. The end point of interest varied, in some
studies being the occurrence of GN bacteremia (e.g. studies of a prophylactic approach;
Table 2.5.1) versus in other studies being the outcome as mortality in studies (e.g.
studies of a treatment approach; Table 2.5.2/.3). The target patient groups under study
also varied. Some anti-sera studies used normal (non-immunized) immunoglobulin
Chapter 2: Introduction & overview
-73-
preparations as the intervention for control group patients. However, the choice of
normal immunoglobulin preparations for control group patients is problematic as it is
possible that even these may have efficacy in patients with sepsis with a presumed antiendotoxin mechanism [4, 481, 482] mediating possible mortality benefit [481],
although not seen in all studies [482].
A small number of studies listed in Table 2.5.1/.2/.3 have used the detection of
endotoxemia as either a criterion for study entry or as a quantitative outcome of
intervention [4, 441, 442]. In this regard, the strongest evidence that quantitative levels
of endotoxemia are related to outcome comes from patients with meningococcemia.
However, three studies of anti-endotoxin therapies have failed to find any significant
mortality benefit in this condition (Table 2.5.2). One study that examined the effect of
endotoxemia detection on outcome using a logistic regression model found no
significant interaction effect from the intervention [438] and another study examined
levels of endotoxemia but did not report the impact on outcome [439].
Polymyxin B The anti-endotoxin intervention with the strongest evidence for
endotoxin binding and neutralization in pre-clinical studies and with the most published
clinical experience is polymyxin B [483-486]. Polymyxins are from a group of cyclic
cationic polypeptide antibiotics and these have well characterized lipopolysaccharide
binding (presumably lipid-A) associated with inhibition of endotoxin activity as
measured in vitro and in whole animal studies. Toxicity limits the clinical use of
polymyxin B as an antibiotic [485]. However, polymyxin B can be bound to a solid
phase such as a hemo-perfusion column [485]. This enables hemo-perfusion as a
method for the removal of circulating endotoxemia through exposure to immobilized
polymyxin B without the systemic toxicity. This method is being actively explored in
clinical trials.
It is notable that the endotoxin antagonist property of polymyxin-B is dependent
on the bacterial origin of the LPS. Amongst the preparations of LPS from eight
different GN bacteria, Polymyxin was most inhibitory against the LPS of E. coli and
least inhibitory against the LPS of N. meningitidis. A paradoxical finding from this
study was that while polymyxin at low concentrations was inhibitory to the effects of
endotoxin, at concentrations greater than 50 mcg/ml polymyxin acted synergistically
with endotoxin in inducing IL-1 secretion [484].
-74-
Chapter 2: Introduction & overview
Cruz et al [487, 488] systematically reviewed the polymyxin-B hemo-perfusion
literature published to 2006 and found 28 publications (Table 2.5.4). The results of
these studies were significantly heterogeneous. All but two of the 28 publications had
come from groups in Japan and many were small non-randomized and unblinded
studies. There were 17 studies that had reported the effects of polymyxin-B hemoperfusion on levels of endotoxemia amongst which there had been possible ―overrepresentation of two centers representing several of the 17 studies in the metaanalysis‖ [488]. A sensitivity analysis undertaken to allow for possible duplicate
reporting, with removal of the duplicate studies, had found results similar to the results
Table 2.5.4 polymyxin-B hemo-perfusion
(data as abstracted in Cruz et al 2007 [488])
Reduction in
endotoxin levels
All studies
Center duplication
number of
studies
17
6
(pg/ml)
Effect size
95% CI
21.2
17.5 - 24.9
heterogeneity
(p value)
<0.001
16.4
8.9 - 24.0
<0.001
Mortality
(risk ratio)
number of
studies
15
Effect size
95% CI
0.53
0.43 – 0.65
heterogeneity
(p value)
0.07
studies with n>20
7
0.56
0.46 – 0.68
0.17
Center duplication
8
0.61
0.46 – 0.82
0.03
All studies
Foot notes;
1. The heterogeneity p-value indicates the variability between study results compared
to the amount that would be expected due to random variation
2. The reduction in endotoxin levels represents a decrease estimated by Cruz et al to be
between 33% to 80% from pre polymyxin-B hemo-perfusion levels.
3. Does not include the most recent polymyxin-B hemo-perfusion study (Cruz et al
2010 [491]) or study of Reinhart ([453] see Table 2.5.2)
Chapter 2: Introduction & overview
-75-
with all studies included and, moreover, the associated significant heterogeneity
remained. Whilst the Japanese studies suggest that levels of endotoxemia were reduced
by polymyxin-B hemo-perfusion [489], this was not found in one small (n=36 [490])
European study. A second small European study in 24 septic patients of whom 11 had
high endotoxin activity (EA) as measured using a chemo-luminescent assay [494] had
found reductions in EA with polymyxin-B based hemo-perfusion and none of the 24
patients died.
Cruz et al concluded that there was a need for confirmatory studies in adequately
powered studies with prospective, blinded and randomized study designs, two of which
are summarized here [491, 494].
A multi-center European study of post-surgical patients with severe sepsis or
septic shock secondary to intra-abdominal infection, found no difference in mortality; 5
of 17 (29%) patients receiving polymyxin-B hemo-perfusion versus 5 of 18 (28%)
patients receiving conventional therapy [494]. This study used evidence of a GN
infection or detectable endotoxemia as criteria for patient inclusion in this study.
However, the endotoxemia levels measured before and after the polymyxin-B hemoperfusion were not significantly different.
A subsequent study, the EUPHAS study (Early Use of Polymyxin Hemoperfusion
in Abdominal Septic shock) was of a similar design, a similar patient group (postsurgical patients with severe sepsis or septic shock secondary to intra-abdominal
infection) and was double the size (64 patients) of the previous European study [491].
EUPHAS was conducted in 10 Italian ICU‘s over a 3 year period (2004-2007) with
clinical severity indices and mortality as measures of outcome. The study was
terminated after an interim analysis revealed that a statistically significant and
seemingly impressive reduction in mortality had become apparent between the groups
after the enrolment of 64 patients. The reduction in 28 day mortality achieved in this
study was as follows; 11 of 34 (33%) patients with polymyxin-B hemo-perfusion
versus 16 of 30 (53%) patients with conventional therapy [491].
However, there are four criticisms of the EUPHAS study [721, 722]. One is that
even at 64 patients, it is a small study. Also, the number of bacteremias found in the
two groups was unequal; 16 among the patients receiving polymyxin-B hemo-perfusion
versus only 3 among the patients receiving conventional therapy, a highly significant
and unexplained imbalance. A third criticism is that the study results were statistically
significant only according to an analysis based on the relative survival time and not
-76-
Chapter 2: Introduction & overview
absolute survival. That is, survival had been prolonged by a few days only (adjusted
hazard ratio 0.36; 95% CI – 0.16-0.8) but not by a Fisher‘s exact test (p = 0.13).
Finally, a criticism of the EUPHAS study is that being a preliminary study,
endotoxemia levels were not measured. These study deficiencies are to be addressed in
two follow up multi-center studies, the EUPHRATES (Evaluating Use of Polymyxin
Hemo-perfusion in an RCT of Adults treated for Endotoxemia and Septic shock) which
is to be undertaken in North American ICU‘s together with the EUPHAS2 which is to
be undertaken in Italian centers.
Another European study of endotoxemia adsorption using hemo-perfusion
randomized 143 septic patients with suspected gram-negative sepsis [453] to receive
either standard therapy alone versus standard therapy plus an extracorporeal endotoxin
absorber. In this study [453], the endotoxin absorber was immobilised human serum
albumin (iHSA), not polymxin. There was a non-significant trend toward lower EA
with the extracorporeal endotoxin absorber treatment but no difference was found in the
primary end point overall (APAHEII score) or survival [453].
2.5.3. Prognostic value of endotoxemia: clinical experience disconnect
The prognostic value of endotoxemia detection in the setting of the critically ill
patient has been studied in over thirty studies (literature cited in [497]). Conflicting
conclusions became apparent from the earliest studies undertaken [25, 54, 75].
The prognostic value remains unresolved despite 17 large studies including over
2000 patients [11, 21, 23, 25, 37, 38, 48, 54, 65, 74, 75, 86, 89, 91, 95, 102, 104, 106].
On the one hand, in six studies endotoxemia was predictive of septicemia onset or
severe illness [11, 54, 75, 91, 95, 106] and hospital mortality [11, 54, 91] among studies
of hospitalized patients not restricted to an ICU setting. On the other hand, 13 studies
including seven among patients restricted to ICU settings found the detection of
endotoxemia either did not predict organ dysfunction or mortality [23, 25, 37, 74, 75,
48, 91], predicted mortality but not organ dysfunction [65], predicted organ dysfunction
but not mortality [21, 38, 104, 106], or predicted mortality only when the level of
endotoxemia was combined within a lipo-polysaccharide cytokine composite score
[102]. Of note, in only three [11, 54, 95] of these 17 studies did the mortality difference
between the groups positive versus negative for endotoxemia exceed 20 percentage
points.
Chapter 2: Introduction & overview
-77-
In the course of searching the literature for studies meeting the inclusion criteria
(see chapter 4), over 200 studies were found that failed to meet the inclusion criteria of
which 176 are listed amongst the references [498-673]. Amongst these studies are
studies undertaken in the setting of patients with the following conditions or
procedures; peri-operative (n=33 [498-530]), pancreatitis (n=7 [531-537]), liver and
alcoholic disease or liver transplantation (n=35 [538-572]), urological (n=8 [573-580]),
trauma or critical illness (n=21 [581-602]), pediatric (n=11 [603-614]), sepsis and
specified infections (n=22 [615-636]), treatment studies (n=6 [637-642]), and other
including metabolic, heart failure [643-669] and veterinary and animal models (n=31
[298, 337, 467-477, 670-673]). The most common reasons for exclusion of these
studies from the analysis undertaken within this thesis were the lack of blood culture or
mortality data.
There is an extraordinary breadth of populations, end points and associations
among these excluded studies [498-673]. For example, there are studies of endotoxemia
arising in populations of hemodialysis patients [599, 600] and elderly people [646],
studies of endotoxemia arising where the exposure is mastication [645], malaria [632,
633] and marathon running [661, 662], or studies where the association is heart failure
[665, 668] and diabetes [663]. These excluded studies are not further referred to in this
thesis except to state as an observation that the scarcity of GN bacteremia, and
mortality data among these studies itself raises questions regarding the overall
significance of endotoxemia in the broader context in which endotoxemia has been
studied and which itself represents something of a disconnect. On the basis of
observations such as these, it has been questioned in an editorial as to whether
‗endotoxinaemia‘ rather than ‗endotoxemia‘ would more accurately describe the
uncertainty regarding the clinical significance of endotoxin in the circulation [674].
Given the plurality of observations above, do Koch‘s postulates as originally
proposed or as the modified Molecular Koch‘s [675, 676] postulates help us to evaluate
endotoxemia as a pathogenicity trait? Koch‘s postulates originally arose in the context
of highly pathogenic organism-disease relationships such as tuberculosis and anthrax.
The original postulates are somewhat blunt and are not so simple to interpret in the case
of infections with viruses or opportunistic bacteria where ‗pathogneicity‘ depend on a
hosts‘ compromised or non-immune state. The modified Molecular Koch‘s postulates
arose in the molecular age to identify pathogenicity elements that could be transferred
into or out of bacteria by gene transfer [676] in the setting of a controlled laboratory
-78-
Chapter 2: Introduction & overview
experiment. However, these Molecular postulates are not simple to interpret in the case
of clinical observations in which bias and confounding operates to an uncertain and
often unmeasurable degree.
2.6. Meta-analysis: Towards a resolution
The inter-relationships between endotoxemia, GN bacteremia, patient outcome and
the effects of anti-endotoxin interventions have been extensively studied with
conflicting observations arising from multiple sources. Conflicting estimates in this
area are not unusual and additional examples of difficulties toward deriving a summary
estimate exist. For example, studies of the efficacy of appropriate antibiotic therapy
toward treatment success, a seemingly fundamental intervention for the ICU patient
[411], have generated conflicting results. For example a systematic review of 51 studies
of appropriate antibiotic therapy found evidence of significant benefit toward outcome
in 28 studies versus 21 studies which did not [412]. A more recent systematic review of
70 studies of this topic found considerable heterogeneity among all 70 studies (I2
>70%) [413]. Likewise, a meta-analysis of 17 studies of the value of combination
therapy for GN bacteremia found studies with evidence for, against, and neutral being
found in two, one and 14 studies respectively [414]. Likewise, among 21 randomized
controlled trials in a meta-analysis of adjunctive steroid use in sepsis, there were three
studies which demonstrated significant benefit with steroid use, one study that
demonstrated significant harm and the remaining 17 studies were neutral [415].
Given the many fundamental uncertainties that remain in our understanding of
the relationship between the detection of endotoxemia, the detection GN bacteremia,
the type of GN bacteremia and patient outcome, it needs to be asked how these can be
resolved. On the one hand, it is unlikely that further animal model research could
resolve these uncertainties given the key role of patient factors and underlying risk in
the inter-relationships. On the other hand, it is unlikely that a single clinical study
would definitively resolve the uncertainties in this area given the breadth of patient
groups and GN bacterial types that such a study would need to include [677-679]. To
paraphrase Einstein, “Can a solution be derived at the same level at which the problem
arose?”
The first step towards resolving the uncertainties is to measure the extent of
Chapter 2: Introduction & overview
-79-
variation among the study results. The second step is to determine whether the
published study results are reconcilable with each other and with overall variability in
the broader literature. Meta-analysis is a method by which these steps can be achieved
[680-683]. The methodology used in this thesis is described elsewhere in this thesis
(see Chapter 4.2 and also EOID 1997 & APLM 2011) but the logical development is
described here.
It is first necessary to clarify what is a meta-analysis and what is a systematic
review. The distinction between these is clarified here in part to address a common
misconception as reflected in the comment of a reviewer of one of the compiled
publications who questioned ―… the potential for bias as a result of singleton
authorship. This is definitely not the gold standard for systematic reviews or metaanalyses in 2010….‖ In contrast to this reviewers‘ opinion, the definitions for these
terms within The PRISMA (Preferred Reporting Items for Systematic Reviews and
Meta-Analyses) statement 2009 [684] and are as follows;……
―A systematic review is a review of a clearly formulated question that
uses systematic and explicit methods to identify, select, and critically appraise
relevant research, and to collect and analyze data from the studies that are
included in the review. Statistical methods (meta-analysis) may or may not be
used to analyze and summarize the results of the included studies. Metaanalysis refers to the use of statistical techniques in a systematic review to
integrate the results of included studies.‖
The PRISMA statement [684] includes a 27 item checklist which has been
derived as a consensus document. The ―…..aim of the PRISMA Statement is to help
authors improve the reporting of systematic reviews and meta-analyses….. the PRISMA
checklist is not a quality assessment instrument to gauge the quality of a systematic
review.‖ The method of data abstraction is one checklist item and the use of two data
abstracters is given as an example rather than as a prescription. Hence authorship,
singleton or otherwise, is not one of the 27 checklist items.
A systematic review indicates a complete review of the available literature and
is usually undertaken to achieve an overall or average effect for a defined intervention
in a defined patient group toward the clinically relevant end-points. This is commonly
undertaken for the purpose of developing and defining an evidence based treatment
-80-
Chapter 2: Introduction & overview
policy. This is not the intention for either this thesis or any of the compiled publications
that are appended. Confusingly, meta-analysis is often the method by which each
quantitative summary is derived within the systematic reviews. Hence meta-analysis is
the method and systematic review is a formal undertaking in summarizing a collection
of study results.
Meta-analysis can go beyond a simple summary to study the sources of
variability in study results. Indeed where study results are highly variable within a
systematic review, a summary effect size will be problematic and its value somewhat
moot. However, deriving a quantitative summary is a necessary step toward both the
recognition of heterogeneity among study results and its quantification. As the reanalysis of the data abstracted by Elin (see Tables 2.2.1/.2) illustrates, finding
differences among a panel of studies can provide additional insights into a collection of
studies that cannot be derived from the individual studies in isolation or by a simple
tally summation of all study results.
Meta-analysis: an historical example. An interesting historical example is
now discussed to illustrate that meta-analysis can play a broader role outside of a
systematic review and which is relevant to this thesis. In 1904, Karl Pearson (who had
described Pearson‘s correlation co-efficient and the chi-squared test) published in the
BMJ an analysis of the impact of anti-typhoid vaccination on the incidence and
mortality associated with typhoid in the British Army [685]. The previous analyses of
the separate British army experience had been inconclusive and the issue was
contentious. This analysis by Pearson is of interest to this thesis for five reasons; it is
one of the first recorded meta-analyses undertaken (although the term ‗synoptical table‘
rather than ‗meta-analysis‘ was used to describe the analysis), the analysis concerned
only the British Army experience, the analysis is derived from observational data under
field conditions rather than randomized trial data, and the question at hand relates to a
gram-negative bacteremic infection (typhoid) with both the occurrence of typhoid fever
and mortality as the end points of interest.
Moreover, this historical example is of particular interest because the main
finding of Pearson‘s analysis was not simply a quantitative summary of the data. The
main finding was that ―Many of the groups …are far too small to allow of any definite
opinion being formed at all, having regard to the size of the probable error involved‖
[685]. That is, the data from the five different army deployments in India and South
Chapter 2: Introduction & overview
-81-
Africa were too heterogeneous to be represented by a single unifying effect size
associated with vaccine use and that more study was required. This was an important
conclusion which was not otherwise identifiable without meta-analysis.
The issue of typhoid vaccination remained contentious. There was a critical
reply from Sir Almouth Wright and a subsequent exchange between the two occurred
appearing in the next six issues of the BMJ over the months of November and
December, 1904. However, Sir Almouth Wright did acknowledge the deficiencies in
the British army medical data that the analysis had identified. The epilogue to the
typhoid vaccine study is that typhoid vaccination was adopted at the time despite the
analysis of Pearson. It took another 50 years and the discovery of chloramphenicol, an
effective treatment for typhoid, before randomized trials of typhoid vaccination were
deemed to be ethically feasible.
This historical example illustrates the broader capability of meta-analysis
beyond that of simply deriving a summary result within a systematic review.
Sometimes, a more appropriate finding than a summary result is a measure of the
amount of heterogeneity among the available evidence. Methods to quantify
heterogeneity and to attempt to explain it using meta-regression have recently been
developed for this purpose [313, 314, 686, 687]. The current state of these methods and
their applicability to the research questions within this thesis is described in 4.2.
Meta-analysis: limitations. The limitations of meta-analysis are several and
need to be considered. A frank article entitled ―Meta-research: the art of getting it
wrong‖ [688] appeared in the journal Research synthesis methods” in 2011. The author
of this article was JPA Ioannidis. He has authored many overviews of the meta-analysis
technique and its impact on science as well as overviews on the sociology and
epidemiology of medical authorship and publication. He has co-authored many original
meta-analyses of topics including diagnostic tests and also an author of one of the first
meta-analyses of studies of corticosteroid use in meningitis.
JPA Ioannidis summarized the evolution in seven published meta-analyses of
studies of corticosteroid use in meningitis as follows;
“1994; no question about benefits, but beware of harms;
1997; definite benefit only for some bacteria, limit to 2 days to avoid harm;
2003; definite benefit only for children, no increase in harm;
2003; correction: actually benefit is seen also in adults;
-82-
Chapter 2: Introduction & overview
2007; benefit in high-income countries, but not in low-income countries;
2009; clear benefit, give to all, this is it;
2010; no benefit at all;”
and conceded that the conclusions ―have been all over the place‖. However,
whatever the true value of corticosteroids in meningitis may ultimately prove to be,
Ioannidis opined that these meta-analyses were able to reveal that the two most
influential and highly cited original studies (both published in the New England Journal
of Medicine) found treatment effects that in retrospect appear implausibly large.
It is not unprecedented that the finding of a large meta-analysis is overturned by
unpublished data. Two recent examples are given here of where the influence of
publication bias on the summary findings. One is that of a recent systematic review and
meta-analysis eleven studies of plasma TREM-1 for sepsis diagnosis in systemic
inflammatory patients [689]. Another example is the finding of one meta-analysis that
found excess mortality related to cefepime [713] was subsequently refuted [714] after a
large body of unpublished studies controversially appeared from the manufacturer of
cefipime [715-718]. Hence, while publication bias is one of the most important threats
to the validity of the conclusions of a meta-analysis, the technique at least offers an
objective method for identifying this threat.
-83-
3. Aims and objectives of the thesis
3.1. A statement of the problem
Studying the relationships between the detection of endotoxemia versus gramnegative bacteremia and also versus outcome as mortality is problematic for the
following reasons;
the structure activity relationship of lipid-A (endotoxin, lipopolysaccharide) in
the mediation of the various biological effects of lipid-A is not simple,
the structure of lipid-A as found in different GN bacteria is not uniform,
the recognition of lipid-A by the innate immune system is critically dependent
on a hexa-acyl structure,
there is a range of GN bacteremia types of interest among which there is a range
in mortality risk which is variable from study to study,
the defining criteria for sepsis are problematic in being overly sensitive and
these continue to evolve,
there are several patient groups in which these relationships have been studied
(ICU, pediatric, oncology etc),
GN bacteremia, sepsis, and presumably endotoxemia are conditions for which
the attributable mortality is difficult to distinguish from that due to the
underlying patient risk,
while the limulus assay is the most widely used assay for the detection of
endotoxemia, this assay is problematic in that;
o it measures only one of several separately mediated biological activities
of endotoxin,
o the clinical experience with the assay is highly variable,
o the version of the assay in use prior to 1985 relied on a gelation (Glimulus) end point whereas the version in common use since 1985 relied
on the conversion of a synthetic chromogenic substrate (C-limulus).
Overall, approximately half of the studies of endotoxemia detection used
the more sensitive C-limulus assay and half used the original G-limulus
assay,
o it is not the only assay that has been used for the detection of
endotoxemia, and
o levels of endotoxemia are best interpreted on a logarithmic rather than
on a linear scale.
apart from studies of meningococcal disease, the evidence that quantitative
endotoxemia detection correlates with outcome is conflicting,
in treatment studies it is unclear whether clinical benefit is either dependent on
-84-
Chapter 3: Aims & objectives
or follows from a reduction in endotoxemia levels,
there is conflicting evidence that methods that block or inhibit endotoxin are
clinically useful,
some of the clinical studies that are in conflict are large and they would be
difficult and expensive to reproduce, and
there is a substantial ‗disconnect‘ between preclinical evidence on the one hand,
which is mostly strong and reproducible, versus the clinical experience on the
other hand, which is variable to the point of being unreproducible.
In approaching this complex and multiply studied topic, it will not be sufficient
simply to define the relationships between the detection of endotoxemia versus gramnegative bacteremia and also versus outcome as an average or as a typical finding. A
new paradigm is required as it is of greater interest to define the extent of variation
between study findings as a basis toward identifying possible reasons for conflict.
Section 4.2 attempts to describe meta-analysis as the new paradigm to achieve this.
3.2. The Research Questions
The research described here addresses three primary questions related to the detection
of endotoxemia in various clinical settings using the limulus (LAL) assay with respect
to;
1.
What is its diagnostic relevance versus GN bacteremia?
2.
What is its prognostic relevance versus GN bacteremia? and
3.
Are the disparate study results reconcilable?
There are six considerations in approaching these questions;
1.
GN bacteremia is used as the reference standard for interpreting the diagnostic
relevance of the detection of endotoxemia
2.
Mortality is used as the reference standard for interpreting the prognostic
relevance of the detection of endotoxemia
3.
Does the diagnostic relevance of endotoxemia detection depend on which of the
endotoxin detection assays is used? the type of GN bacteremia isolates found? the study
setting?
4.
Does the prognostic relevance of endotoxemia detection improve on that
provided by the detection of GN bacteremia?
5.
Does this prognostic relevance of endotoxemia detection depend on the types of
GN bacteremia isolates? the underlying population risk in each study setting?
6.
Can the disparate findings among the published studies be reconciled with the
broader literature such as studies with incomplete data or studies that used non-limulus
assay for endotoxemia detection?
-85-
4. Materials and methodology
4.1. Literature search methods
Search strategy. A review of the literature was performed repeatedly as detailed in the
compiled publications (listed in section 1.2.1). This search was most recently done in
June 2012 (Figure 5.0.1). The literature review included a search of the Medline
database back to 1966 (but included publications prior to that date [97, 99]), with the
search headings being "septicemia" (prior to 1992), "bacteremia" (from 1992 on),
"endotox*," and "human." The search was not limited to reports published in the
English language [19, 34, 44, 82, 92] and included conference proceeding abstracts [7,
62]. This was supplemented with a manual search of the references from each report
retrieved, review articles, and textbooks.
Unpublished data were sought directly from the authors of studies who had
published results in a form which did not meet the above criteria to seek a clarification
that would have enabled inclusion (studies listed in section 7). This included the
publication of a call for data (JER 2001) in addition to direct contact with the primary
authors for additional unpublished data or clarification of published data, and these
sources of data have been acknowledged in each publication where relevant. This
correspondence is included within section 7 of this thesis.
Generic inclusion criteria. The following were the minimum criteria used in
selecting studies for inclusion:
1. study design, which included a direct comparison of blood culture and LAL
assay methods applied to blood from patients with suspected sepsis in various
clinical settings;
2. size, with at least four patients studied; and
3. data presentation, such that each patient could be classified into one of four
possible categories: Category 1, blood cultures positive for GN bacteria, and
endotoxemia positive; Category 2, blood cultures positive for GN bacteria, and
endotoxemia negative; Category 3, blood cultures negative for GN bacteria, and
endotoxemia positive; and Category 4, blood cultures negative for GN bacteria,
-86-
Chapter 4: Materials & methods
and endotoxemia negative.
Generic exclusion criteria. Studies of animal models, studies of nonbacteremic infections, data insufficient or incorrect format (i.e. no comparison with GN
bacteremia), or duplicate publications were not included.
The main data set of studies meeting the generic inclusion and exclusion criteria
are listed in Table 7.1. Additional inclusion and exclusion criteria were applied as
detailed within the individual publications in the compilation as relevant to which of the
three research questions (i, ii, iii) was under examination within the individual
publications in the compilation (see publication listed in 1.2.1).
Supplementary studies. (listed in Table 7.2). As part of updating this thesis,
studies that were excluded from the main data set of studies were re-examined. A list of
supplementary studies was compiled using the following criteria to select studies from
the list of excluded studies; studies in which the patients had received an anti-endotoxin
intervention, studies that had used a method for endotoxemia detection using an assay
other than the LAL assay (e.g. rabbit pyrogen, endotoxin activity assay – EAA, or
simply red endotoxin-SRE), and studies that had provided outcome data in relation to
the detection of endotoxemia that was not extractable into the 4x4 format (as in (3)
above). The outcome data of these supplementary studies and the main data set studies
(which have a 4x4 format) were all re-extracted into a 2x2 format. This enabled a
comparison of the results of the main data set to the additional data provided by these
supplementary studies as a sensitivity test.
Quality scoring. Quality scoring is not recommended for studies of diagnostic
tests [690] and their use to adjust the results of meta-analyses in general is controversial
[691]. Instead, the approach taken here was to stratify the studies in relation to the
following four equally weighted criteria;
Whether the study design specified patient inclusion criteria in sufficient detail
that would enable a replication of the study with respect to study setting and
target patient population,
Whether it was explicit that each patient test for endotoxemia was undertaken
concurrently with a test for bacteremia,
Whether patient outcome as mortality was specified with details of census time,
Whether the list of GN bacteremia isolates was given.
Studies that met > 2 of these criteria were stratified as higher in quality of study
Chapter 4: Materials & methods
-87-
design.
4.2. Analytic approach
The principal questions of this thesis require the following; firstly, deriving
estimates for the average or typical effect size from the range of studies, and secondly,
deriving estimates of precision and also heterogeneity among the range of studies. The
analytic methods used in this thesis evolved over the time period of the publications
(1994-2012) in three phases. Over this time period increasingly sophisticated analytic
techniques for the analysis of proportion data became available in the STATA®
software used in this thesis. These techniques are described in the order of the three
phases as they evolved. These phases are to be described in detail below but first it is
necessary to describe the statistical issues and approaches to the analysis of proportion
data [692-694].
Analytic issues. Throughout this thesis, proportion data is extensively presented
and analysed arising in the following contexts;
as single proportion data,
as paired proportion data (e.g. as in meta-analysis),
as correlated proportions (e.g. sensitivity and specificity resulting from a
diagnostic test applied at different breakpoints), and
as non-independent proportion data (e.g. mortality proportions from multiple
clusters of infectious diseases).
To begin with, the simplest approach to the analysis of proportion data could be
to convert and undertake the analysis as percentages rather than as the original
proportion data. For example, the proportion 1/20 could be converted and analysed as a
percentage value (5%). However, conversion of proportions to percentages is a form of
data reduction as it fails to fully use all the information in the data (numerator and
denominator). In particular, a statistical inference can be made from a proportion but
not from a percentage.
Four basic issues relevant to the proportion data as found in this thesis are
discussed here. In particular;
(i)
Occasional small denominators (e.g. <20 patients),
(ii)
Occasional zero numerators (e.g. 0 of 20 and also 20 of 20)
(iii) Variable precision of estimates due to variable group size (e.g. 2 of 20
versus 20 of 200).
-88-
Chapter 4: Materials & methods
(iv) Numerators being discrete integers (e.g. 2 of 20 patients).
Single proportion data. To illustrate some of the analytic issues and the available
solutions, an example is given here for the calculation of the 95% confidence interval
associated with an observed single proportion with a zero numerator and small
denominator using the proportion 0/20. The simplest formula to estimate the probability
and derive the standard error is the Wald method as follows;
o given k successes of n trials (in this instance 0 of 20),
o the estimated probability is p = k/n (in this instance 0 = 0/20) with
o standard error = √(p(1 - p)/n), being in this instance 0 = √ (0(1 - 0)/20).
With an observed proportion such as 0/20, there are three possible inferences. One
inference is that the occurrence of the event is impossible, in which case the true value
is zero with zero standard error and a 95% confidence interval of zero (which is what
the values derived in this instance using the Wald method would imply). Several more
complex formulae exist to derive estimates of standard error for a single proportion
which have better (non-zero) coverage in this context. These methods include the
Wilson, Agresti-Coull and Jeffreys methods which in the case of 1/20, yield 95%
confidence interval of 0-0.161, 0-0.189 and 0-0.116, respectively. Note that all of these
intervals are asymmetrical being truncated with zero as the lower 95% confidence limit.
The second inference for an observed proportion such as 0/20 is that the
observed proportion is simply one of a finite number of all possible outcomes. With a
denominator as small as 20, all possible outcomes can be enumerated and a 95%
confidence interval can be derived using a hyper-geometric distribution. For the
observed proportion of 0/20, this is calculated as a result equal to or more extreme than
1/20 which yields a confidence interval as 0-0.168 being a 97.5% one sided confidence
interval. This leads to a special case solution based on exact testing inference which
enables the comparison of a limited number (e.g. two or three) of small denominator
proportions (e.g. 0/20 versus 2/10). The most common example of this is Fisher‘s exact
test which arises in the context of conducting a chi-square test where one or more cell
has an expected frequency of five or less. The result for the Fisher‘s exact test is
derived using permutation methods. In this example (0/20 versus 2/10) the result differs
for the Fisher‘s exact test (p = 0.103) from that for the incorrectly applied Pearson chisquare test (p = 0.038). An extension of the exact test is exact logistic regression (see
Chapter 4: Materials & methods
-89-
below).
A third and more plausible inference for our observed proportion of 0/20 is that
the event is uncommon in which case the zero numerator observed simply is a
reflection of the relatively small sample size (being a sample size of 20 in this instance).
A solution that accommodates this uncommon event inference is the use of a continuity
correction within a logit transformation of proportion data. This method overcomes
several of the issues (i - iv) identified above. The logit (or logistic) transformation is
used in logistic regression (see below). A logit transform of a proportion is the log of
the odds. Hence for p = k/n this is logit = log(k/(n-k)). Any zero here either in the
numerator (k), or in the denominator (n-k) will require the addition of a continuity
correction (e.g. 1 or 0.5) to avoid an undefined logit.
Note that the use of a continuity correction is consistent with the inference that
the event is actually uncommon rather than zero. Hence, for an observed proportion
such as p=0/20 and using 1 as a continuity correction, the logit transform of the inferred
proportion (‗p‘=1/20) is log(1/19). The standard error of a log odds (by the delta
method) = 1/√(‗p‘(1 – ‗p‘)*n) = 1/√ (.05*.95*20). The inferred proportion and the
derived 95% confidence interval on back transformation is now 0.05; 0.006 – 0.31.
Note that in the 95% confidence interval as calculated using this transformation has two
desirable properties; it is symmetrical when viewed on the logit scale and second, the
calculated 95% confidence interval remains within the range 0 to 1 and cannot extend
outside zero or ‗1‘. Note that any proportion ‗p‘ close to 0 (as in this example) or 1 will
have a small standard error whereas the log odds of ‗p‘ will have a large standard error.
-90-
Chapter 4: Materials & methods
measures of effect size
(odds ratio, risk ratio and risk difference)
100
event rate in
group 2 (%)
80
60
40
20
0
0
20
40
60
80
event rate group 1 (%)
100
Fig 4.1 Illustration of three different potential representations of study effect size as
could be represented within a L‘abbe plot. These are where the study effect size is
represented as;
a risk difference (RD; e.g. equivalent to a 20% excess; wide solid line),
a risk ratio (RR; e.g. equal to 2; dashed line) and
an odds ratio (OR; e.g. equal to 2; dotted curve).
Note that the risk difference (RD) implies an assumption that the effect is constant
over the range of event rates in group 1 whereas the odds ratio (OR) and risk
difference (RD) imply that the effect is relative. See page 91 for calculations.
The line of equivalence (narrow solid line) represents all of the following effect
sizes; a risk difference equivalent to a 0% excess, a risk ratio of 1 and an odds ratio
of 1.
Chapter 4: Materials & methods
4.3. Three phases of statistical methods
The analysis has three phases corresponding to use of statistical methods of
increasing sophistication. These various methods used in the three phases each have
strengths, limitations and optimal applications to the research questions of the thesis
(see summary in table 4.1). The more recent methods were not necessarily superior to
previous approaches but together these methods collectively allow a ―triangulation‖ of
the data, whereby the results by these different methods can be compared and
contrasted.
Table 4.1 provides a summary of the analytic methods used in this thesis.
-91-
-92-
Chapter 4: Materials & methods
Table 4.1 Summary of analytic methods used in this thesis.
Method
Simple methods
‗vote count‘
Tally
Robust
(median; IQR)
Used in
thesis
phase *
NA‡
1-3
1
Advantage
Disadvantages or
limitations
simplicity
simplicity
simplistic
large study bias
robust to outliers
non-parametric
Fixed effect methods [681, 683, 687; EOID 1996]
Mantel
1-3
traditional method
fixed effects
Haenszel †
& enables;
Forrest plots
L‘abbe plots
Metaregression †
Exact logistic
regression
2
3
basic regression
assumptions
robust to sparse &
unbalanced data
Examples;
T = Table
F = figure
T 2.2.1/.2
T 5.1.3/.4; 5.2.4
T 2.4.2
F 5.2.9 (p 144)
T 5.1.1; 5.2.1/.2
F 5.2.1/.2/.4/.5/.6
F 5.2.3/.7/.9
only study level
covariates
estimation not
included
T 5.1.5
T 5.2.5
Methods applicable to diagnostic tests [261-264, 311-312; AP&LM 2011]
ROC / SROC
2
diagnostic tests
fixed effects
T 5.1.1
HSROC
3
diagnostic tests
Computational
T 2.2.3; 5.1.1/.5
random effects
complexity
F 5.1.1/.2/.3/.4/.5
Random, mixed and multi-level methods [313, 314, 692-694, 700-707]
Der Simonian
2-3
quantification of
Computational
T 5.2.2/.3
& Laird
heterogeneity
complexity
GEE §
3
robust to clustered
Computational
T 5.2.5
data
complexity
Foot notes;
* Phase 1 = 1994-2000; phase 2 = 2000-2009; phase 3 = 2009-2012 as described in
section 4.3
‡ NA – not applicable as this method not used in the thesis. ‗Vote count‘ as used in
narrative reviews is a count of the number of positive studies
† Also adaptable to a random effects model.
§GEE = Generalized estimating equations
Chapter 4: Materials & methods
-93-
Phase 1 (up to 2000)
The Mantel-Haenszel (MH) method is the simplest technique to aggregate
proportion data arising as paired proportions. The MH method dates from the 1950‘s
and is computationally simple. The method is described in detail in an appendix to
JCM ‘94.
Paired proportions - Meta-analysis. For proportion data arising as paired
proportions (e.g. treatment and control group mortality proportions, endotoxin positive
and endotoxin negative group mortality proportions), the data within the pair of
proportions is usually reduced to an effect size. To exemplify using the pair of
proportions 4/20 (20%) and 4/40 (10%), this effect size would be derived as;
the difference between the proportions (risk difference; RD = 10% = 20%-10%);
the ratio between the proportions (risk ratio; RR = 2 = 20%/10%)]; or
the ratio of the odds (odds ratio; OR = 2.25 = (4/16) / (4/36)). Note however that if
either proportion in the pair is 0% or 100%, the RR and the OR will be
indeterminate and use of the continuity correction (e.g. 1 or 0.5) is required to
derive the effect size.
Several factors determine the choice among the effect measures (risk difference,
RD; risk ratio, RR or odds ratio, OR) [695, 696]. The OR has useful mathematical
properties including symmetry (See figure 4.1). Symmetry is the property where, for
example, the study OR calculated for patients dying is the inverse of the OR for
patients not dying. The RR lacks symmetry but has a more intuitive interpretation to
general readers which the OR lacks. The OR approximates the RR for OR<2. The RD
has the most intuitive interpretation but using the RD effect measure requires an
assumption that the size of the effect is constant over conditions of different underlying
risk. In this thesis the OR is the effect measure used most often as this avoids the
constant effect assumption. However, in 5.2 the RD is used (alongside with the OR) to
facilitate a description of the proporties of the data here.
Meta-analysis is a two or three stage process. The first stage requires the
derivation of an appropriate statistic (e.g. RR, OR, RD) from the paired proportions for
each respective study in a set of studies. These summary statistics can readily be
-94-
Chapter 4: Materials & methods
derived within the STATA® statistical software using the command ‗metan‘ [683].
Meta-analysis is applicable to summary statistics other than the OR, RR and RD. For
example, meta-analysis can be applied to continuous data effect estimates (using the
STATA® command ‗metan‘ with continuous data options), can be applied to effect
estimates and standard errors entered as pre-calculated from each study using the
STATA® command ‗metan‘ with pre-calculated effects estimates options (see phase 3
below) and can even be applied to p-values (using the STATA® command ‗metap‘, not
used in this thesis) or even to logits (see below).
The second stage consists of deriving a weighted average of these statistics
across the set of studies. Various weighting methods can be used, but, in essence, the
optimal weight for each study is the relative amount of information that each study
provides toward the overall estimate. Most commonly the estimate of study precision,
which is simply the inverse of the variance for that study (i.e. 1/standard error), is used
as the weighting for each of the individual studies.
The third stage is the representation of the information within one of the several
available graphical outputs from a meta-analysis. These outputs allow a visual
comparison of both the data from each study to other studies and also in relation to the
overall summary. There are three common graphical outputs and all three are used in
this thesis. These are as follows; a forest plot, a meta-regression and the L‘abbe plot.
The choice of graphical presentation of the study results is of particular interest [681]
given that the reconciliation of conflicting results among the studies is one of the
research questions of this thesis.
A forest plot is a vertical stacked array with the studies listed vertically with the
effect sizes and 95% confidence intervals.
There are two types of meta-regression. The first is one in which the studies are
stratified into sub-categories with sub-group analyses provided. This is usually done as
a vertically stacked array. Alternatively, the relationship of study effect to some study
measure of interest may be studied as a correlation. This relationship can be
investigated with the studies within a regression plot of study effect size (y-axis) versus
a continuous variable of interest in the regression (e.g. drug dose; x-axis). In this array,
the studies are ordered along a bi-variate plot.
The third graphical output for meta-analysis to RCT‘s is the L‘abbe plot [681].
Chapter 4: Materials & methods
-95-
In a L‘abbe plot, the effect size is not displayed. Instead, each study is represented as a
point in a scatter plot of event rate in the intervention group (or case group; y-axis)
versus event rate in the control group (x-axis). The graphical output that is homologous
to the L‘abbe plot from a meta-analysis of CSDT‘s is a SROC which is a scatter plot in
which each study is represented as a point of sensitivity (true positive rate) and false
positive rate (1-specificity) for each study (further discussed in APLM ’11). The main
advantage of a graphical output within a meta-analysis is that the heterogeneity
amongst individual study results and within the overall result can be readily appreciated
visually within the graphical output in addition to a formal estimation by specific
statistical techniques.
Phase 2 (2000 - 2009)
There are three main limitations of the MH method. It is applicable only to
binary (i.e. 2x2 tabular) data. The second is that the MH method is based on the fixed
effects model. The fixed effect is less optimal where the assumption of a common (i.e.
‗fixed‘) effect is less tenable as is usually the case with larger (e.g. > 10 studies) data
sets. The third limitation is that being a fixed effect method, the MH method provides
only a significance test for heterogeneity but not a measure of this.
From 2000, more computationally intensive methods extending beyond the
three limitations above became more readily available, such as the Der Simonian Laird method. Also, methods based on the ROC became available which were
specifically adapted to deal with the meta-analysis of diagnostic tests.
Correlated proportions. While the most common applications of meta-analysis
in the medical literature are in relation to summarizing the effects size in randomized
controlled trials (RCT‘s), this is not the only application. The meta-analysis technique
is applicable beyond RCT‘s with applications to observational studies in both the
medical and non-medical literature. The conventional meta-analysis technique as
applied to RCT‘s has one major limitation with respect to paired proportion data of the
type analysed in this thesis.
A limting issue in the meta-analysis of paired proportion data as commonly
undertaken is that the derivation of an effect size (i.e. risk ratio, odds ratio or risk
difference) for each pair of proportions is a form of data reduction in that it fails to use
-96-
Chapter 4: Materials & methods
Table 4.2 Meta-analysis: fixed effect versus random effects methods.
Fixed effect model
Random effects model
Exemplary method
Mantel-Haenszel
(MH)
Der Simonian - Laird
premise
ES common to all
studies
Random distribution of
ES amongst studies
homogeneity of ESi
heterogeneity of ESi
common ES = 0
mean ES = 0
grand mean ES calculated as
weighted mean across
all study ES
weighted mean across
all study ES
source of variation in study ESi
only within study (Vi)
within study (Vi) &
between study (τ2)
=Vi
=Vi + τ2
inverse of variance
inverse of variance
(=1/Vi)
(=1/(Vi + τ2))
large studies> small
studies
more equal
Presumption of homogeneity in
ESi
null hypothesis
variance for each study
weight for each study
relative weights for large
versus small studies
Foot notes;
ESi = Effect size for study(i)
Vi = Variance for study(i)
all the available data in the numerators and denominators. This data reduction is less of
an issue in the meta-analysis of a series of RCT‘s as the effect size (being OR, RR, RD)
could reasonably be expected to have little variation between studies. In this respect the
fixed effects method (Table 4.2) may be justifiable with this expectation that the effect
size is uniform [682]. This assumption will be most tenable under trials conducted
under standardized conditions and patient populations.
The consequence of data reduction to an effect size is most apparent in the
meta-analysis undertaken for the meta-analysis of diagnostic tests or prognosis studies
for two reasons. Firstly, the standardized conditions and patient populations is not a
tenable assumption.
Also, with the meta-analysis of diagnostic tests, the overall sensitivity and
Chapter 4: Materials & methods
-97-
specificity of a diagnostic test are each of interest to potential users of the diagnostic
test under evaluation and summation as an odds ratio, even a diagnostic odds ratio, fails
to convey these two pieces of information. In this context, fixed effects methods are
seldom justifiable and random effects methods are more appropriate [682]. Additional
methods to aggregate proportion data have become available which avoid the datareduction that occurs with meta-analysis (see below).
Methods specific for correlated proportions. For clinical studies of diagnostic
tests (CSDT‘s), there are two proportions of particular interest, sensitivity and
specificity. However, among a series of studies, these two proportions will be
negatively correlated. That is, one will be expected to increase as the other decreases
across studies that have applied different diagnostic breakpoints. Analytical techniques
to overcome this correlation have emerged to enable the evaluation of clinical studies of
diagnostic tests (CSDT‘s). For the evaluation of diagnostic tests by meta-analysis, the
specialized STATA® command ‗metandi‘ is available.
There are several similarities and differences between the applications of metaanalysis to RCT‘s versus CSDT‘s (discussed further in APLM 2011). The most striking
similarity is that for both RCT‘s and CSDT‘s, the original data is in the binary (i.e. 2 x
2) format. The most crucial difference is that in the case of CSDT‘s the binary data
represents correlated proportions (sensitivity and specificity) versus the case of RCT‘s
where the proportions are not correlated. ‗Metandi‘ fits a two-level mixed logistic
regression model, with independent binomial distributions for the true positives and
true negatives conditional on the sensitivity and specificity in each study, and a
bivariate normal model for the logit transforms of sensitivity and specificity between
studies (as in EJCMID ‘10). This is in contrast to the ‗metan‘ command in the analysis
of RCT‘s, for which there is no such correlation within the data. A summation as a
summary effect size as, for example a summary OR, with associated 95% confidence
interval and measures of heterogeneity, is justified for RCT‘s but not for CSDT‘s.
-98-
Chapter 4: Materials & methods
Phase 3 (since 2009)
Since 2000, methods have become available for the analysis of multiple
proportions arising from clustered data for which independence is not a tenable
assumption. These methods are computationally complex and intensive.
Multiple proportion data and random effects. In the context of deriving
inferences for multiple proportions, the following considerations (additional to i – iv
above – p 87) arise as to the ‗best‘ approach and method to;
(v)
estimate either the average proportion or the expected proportion for a
‗typical‘ study,
(vi)
allow for denominators of varying size (e.g. denominators of 20, 200 and
2000) among the multiple proportions (= unbalanced),
(vii) estimate the 95% confidence interval for the average proportion,
(viii) estimate the heterogeneity among the observed proportions,
(ix)
optimally compare the multiple proportions to each other,
(x)
compare studies over which the underlying event rate varies [697-699],
(xi) allow for non-independence (i.e. cluster effects and intra-cluster correlation)
in the comparison of proportions from multiple clusters.
The last listed consideration (xi) arises in the case of proportion data involving
patients within clustered settings which is particularly relevant for an infectious or
potentially contagious disease as is the case with the data presented in this thesis [700].
As a result of contagion and clustering, observations aarising from within a cluster can
not be assumed to be independent, a crucial assumption for statistical inference.
There are at least six approaches in the analysis of multiple proportions which
variably address these various issues. Five of these methods are used at various points
in this thesis; simplistic tallies, the beta-binomial model (not used in this thesis),
logistic regression, exact logistic regression (ELR), random effects within mixed
models, and the method of generalized estimating equations (GEE).
Simplistic tallies The three simplest methods in common use to derive an
average proportion are as; (1) a simple tally (summation) for the numerator and
denominator across all available studies (as was done by Elin – see Table2.2.1/.2/.3), or
(2) an arithmetic mean of all proportions, or (3) the median proportion (as was done in
[287]). For example, for the proportions 1/20, 10/200 and 200/2000, the summation
would be 211/2220 (≈0.1) whereas the arithmetic mean would be ((100+100+200) ÷
Chapter 4: Materials & methods
-99-
3)/2000 (≈0.067). Note that the summation is biased toward the largest denominator
(200/2000) and the arithmetic mean is biased toward the smaller denominators (1/20
and 10/200). The median is a more robust estimate as it is less biased to outlier data.
However, none of these simple methods addresses the other data issues and questions
of interest (i-xi) identified above.
The beta-binomial model is based on the beta distribution. It is a useful
distribution for modelling proportion data because the beta- distribution is confined
within the range between 0 and 1. It has two additional advantages for modelling
proportions. It is a method that is able to accommodate overdispersion in the data. It is
a flexible distribution in that the shape of the distribution is determined by the two
parameters of the distribution; ‗a‘ and ‘b‘. The mean of the distribution is a/(a+b) and
the variance is ab/((a+b)2(a+b+1)). The beta-binomial model is not used in this thesis.
The beta-binomial model is more commonly used for ecological modelling.
Logistic regression Logistic regression is a method of regression analysis
applicable for where the response variable is binary (i.e. 0 or 1). This method of
analysis utilizes the logit transformation (described above). The logit transformation
overcomes problems that would occur through undertaking a regression on
untransformed data, in particular the problem of fitted proportions that fall outside the
range of 0 to 1. The main advantage of logistic regression is that it can be estimated
using conventional parametric statistical techniques, in particular, maximum likelihood.
An additional advantage of logistic regression is that overdispersion (described below)
can be accommodated. Moreover, meta-analytic techniques can be applied to logit
transformed proportion data to derive estimates of mean and 95% confidence intervals
to enable a visual comparison of multiple proportions on the logistic scale.
Exact methods Exact logistic regression (ELR) is a computationally intensive
method of logistic regression as it uses permutation methods in contrast to the use of
asymptotic estimates and likelihood theory [701-705]. Exact logistic regression is an
extension of Fisher‘s exact test and is ideally suited for data sets containing proportions
that are sparse or unbalanced. ELR has two limitations. It is computationally intensive
and hence is not a feasible method for the following types of data sets; data sets that are
very large, have more than four or five explanatory variables or require tests for
interaction. Secondly, ELR is more suited for hypothesis testing (i.e. generating exact p
values) rather than for parameter estimation (i.e. deriving confidence intervals).
-100-
Chapter 4: Materials & methods
Mixed models The challenge of analysing multiple proportion data as found in
this thesis is best illustrated by considering the impact of additional data becoming
available. Consider that we have a data set with mortality proportion data coming from
three clusters; cluster A - 1/20, cluster B - 10/200 and cluster C - 200/2000. If the
research budget allowed us to collect data for an additional 1000 patients, how should
we do that optimally? Would this be with 1000 extra patients in cluster A, in cluster B,
in cluster C, or in a new cluster D, or in a series of new clusters D, E, F and G each
with 250 patients? The answer to this additional data dilemma is dependent on how
similar the mortality proportions are across different centres, i.e. intra-cluster
correlation [700, 706]. If the mortality proportion were similar in all clusters, the
additional data could just as optimally come from any one of the clusters. However,
that is not usually the case as the usual expectation is that that the mortality risk for the
patients found in one cluster would be more similar to each other than to the mortality
risk for patients found in another cluster.
The intra-cluster correlation coefficient (ICC) is the measure of similarity
within, versus between, clusters. It is the proportion of total variance that is ‗explained‘
by within cluster variation. The converse measure, being the of lack of similarity
between clusters, is overdispersion. This is when the total variance is greater than that
‗explained‘ by the cluster variance. Note that the total variance is not observable but is
derived from the cluster level estimates. This leads to the concept of mixed models.
There are two mixed models approaches for the analysis of clustered binary
data; methods leading to cluster specific (i.e. in this thesis the individual study) versus
methods leading to population averaged, also termed marginal (i.e. in this thesis the
study groups from within a strata) estimates. Both methods are used in this thesis. The
two methods lead to different levels of inference. The random effects modelling
approach provides for a level of inference at the level of a given study. In the random
effects models, the effect in each cluster is explicitly modelled by including cluster
specific random intercepts that are drawn from a probability distribution that uses
model-based variance estimates. These models are typically estimated using maximum
likelihood resulting in the random effects being ‗integrated out‘. These models are
particularly relevant when cluster specific (i.e. at the level of the individual study)
effects are of interest. In this thesis, this level of inference is of interest in relation to
finding individual studies that gave results that can be viewed as statistically atypical.
Chapter 4: Materials & methods
-101-
The two-level mixed logistic regression model described above for the analysis of
diagnostic tests is a special case example of this technique.
The population averaged modelling approach provides for a level of inference
at an ‗average‘ or ‗typical‘ study. This is of interest in relation to comparing groups of
studies that share a characteristic of interest moreso than in the individual studies
themselves. Methods leading to population averages include both the random effects
modelling approach as described above and alternately the method of Generalized
estimating equations (GEE). The difference between the two approaches is based on
how the between-cluster variation is estimated and incorporated into the calculation of
the standard errors. The random effects modelling approach estimates the betweencluster variation directly whereas the GEE method estimates its counterpart, the withincluster similarity of the residuals, using this in turn to estimate the standard errors.
Generalized estimating equations The method of GEE is a population
averaged approach to estimation. GEE fits a population averaged model (i.e. the level
of inference is at the level of the whole population rather than the individual cluster),
using an iteratively re-weighting algorithm and a robust method for the standard error
to estimate the parameters [700, 707]. While GEE is a method of individual level
regression modelling it differs from other regression modelling methods in how the
strength of correlation between observations within the cluster is estimated (as
‗working correlations‘). As a result, there are three conceptual differences between
methods based on random effects modelling versus GEE;
random effects modelling uses the between-cluster variability as the basis for
estimation whereas for the GEE method the counterpart, being the within-cluster
similarity, is used as the basis for estimation.
Secondly, for GEE the population effects are estimated by an algorithm (the
‗Estimating Equations‘ within ‗GEE‘). As this algorithm is not based on any
probability distribution it produces robust variances that can be described as
‗model-agnostic‘.
Thirdly, the GEE estimating method does not require that the data be independent.
This property is highly desirable in the analysis of data derived from clusters of
patients with infectious (i.e. potentially transmissible) diseases.
The robust variance comes under various names and is also known as the
Huber/White/sandwich estimate of variance. Moreover, these robust variances are
consistent even when the model is mis-specified and in this way, the population
estimates of the averages are robust [707].
-102-
Chapter 4: Materials & methods
Note that with both the random effects and population averaged methods, the
estimation of the model parameters does not occur as a single step calculation but
proceeds by iteration until convergence is achieved. Hence both techniques are
relatively computationally intensive. The methods have existed for over a decade but
they have become more feasible using statistical software on the desk-top computer in
the past ten years. Moreover, the methods are more relevant to the analysis of large
rather than small datasets (e.g. datasets with less than twenty groups).
-103-
5.
Results and discussion
This chapter presents the analyses of the data derived from the studies derived
from the literature search (figure 5.0.1). These studies are catalogued in Table 7.1/.2.
This thesis is a compilation of published works. This chapter provides an update of the
analysis presented in summary form. Additional details are contained in the compiled
publications (appended at the end of this thesis).
Figure 5.0.1 Flow chart of literature search and study accrual
Electronic search for eligible studies (n=208)
Search terms:
 endotoxemia
 Limulus assay
Hand search for additional eligible studies (n = 91)
Exclusion criteria (n=192)
Animal studies (n = 23)
Studies of patient groups other than suspected sepsis; e.g. post
gasto-intestinal endoscopic procedures (n = 19)
Studies of non-bacteremic infections (n=52)
Data insufficient or incorrect format (n = 67)
Studies with fewer than 10 patients (n=6)
Duplicate publication (n = 3)
Supplementary studies – see below (n= 12)
Included studies [1-95]
(n = 95);
G-limulus ‗4 group‘ (n=41)
C-limulus ‗4 group‘ (n=46)
Not stated ‗4 group‘ (n=8)
Studies with additional data in relation to
mortality data in ‗4 group‘ format (n=36)
GN bacteremia type listings (n=26)
Supplementary studies
[96-107]
(n = 12);
non-limulus ‗4 group‘ [96101]
limulus/non-limulus ‗2
group‘ [96-107]
Anti-endotoxin [88, 94]
-104-
Chapter 5: Results & discussion
These results are presented as follows here and in the compiled publications as
they bear on the detection of endotoxemia;
as a diagnostic test (section 5.1, JCM ‘94, AP&LM ‘00, EJCMID ’10)
and as a test of prognosis (section 5.2, JCM ‘95, EIHD ’99, JER ’03, CC ‘12)
5.1. Endotoxemia detection as a diagnostic test
The number of studies in the analysis as included in three prior publications
(JCM ‘94, AP&LM ‘00, EJCMID ‘10) together with an update for this thesis are
summarized in Table 5.1.1. Five of the publications [24, 54, 58, 65, 94] provided two
discrete studies each. The update prepared for this thesis includes 29 additional studies
including four additional non-English language studies. Forty-one studies used the Glimulus and 47 studies used the C-limulus assay. Eight small studies were unusable for
this part of the analysis as they had more than one zero entry within the studies‘ 2x2
table.
The listed assay sensitivity to the internal endotoxin standard in the studies
varied by more than a 100 fold being typically in the range 0.1 – 10 ng/ml for the Glimulus assays versus 0.001 – 0.1 ng / ml for the C-limulus based assays (see Table
7.1). The prevalence of GN bacteremia was similar in the studies using G-limulus and
C-limulus (24.1%; 16.7 – 31.5 versus 24.5 %; 20.5 – 28.5) versions of the assay.
Fifteen studies were limited to pediatric patients, and eight studies were limited to
oncology or neutropenic patients. There were 11 studies of sepsis patients, all of these
studies had been conducted within an intensive care unit setting and all post date the
use of sepsis as a clinical syndrome defined by consensus in 1991. All 11 had used the
C-limulus assay.
Chapter 5: Results & discussion
-105-
Table 5.1.1 Endotoxemia diagnostic studies: Summary results.
JCM ‗94
AP&LM
‗00
EJCMID
2010
Thesis
2012
Number of publications
43
54
55
80
Number of studies (N)
45
56
58
88
1970-1994
1970-1997
1970-2004
1970-2005
2539
4134
4681
5641
615 (24%)
856 (21%)
883 (19%)
1480 (26%)
Mantel
SROC
HSROC
HSROC
no
yes
yes
Details of meta-analysis
Publication years
Number of patients
Number of GN bacteremias (%)
Analytic method
Haenszel
Weighting for study size
Summary measure
All studies
G-limulus studies
Number of studies (N)
C-limulus studies
Number of studies (N)
Sub analysis
- defined by
G-limulus studies
Number of studies (N)
C-limulus studies
Number of studies (N)
yes
DOR;
Q*;
95% CI a
95% CI
b
DOR;
DOR;
95% CI a
95% CI a
8.3;
5.9 – 11.8
(29)
6.3;
4.4 – 9.1
(16)
0.71;
0.64 – 0.77
(28)
0.66;
0.61 – 0.72
(28)
10.7;
5.2 – 22
(25)
5.4;
3.2-9
(33)
8.4;
4.5 – 15.7
(41)
4.7;
3.0-7.2
(47)
Study GN in
range 13-40%
Excluding
small studies c
Excluding
small studies c
Excluding small
studies c
20.1;
12.5 – 32.4
(13)
0.77;
0.68 – 0.81
(20)
12.5;
6.1-26
(18)
10.0;
4.9-20.3
(25)
3.6;
2.3 – 5.6
(9)
0.65;
0.58 – 0.72
(22)
6.2;
3.6-11
(24)
5.5;
3.4-8.9
(33)
Foot notes; Abbreviations; DOR = Diagnostic Odds Ratio, SROC = summary receiver
operator characteristic; HSROC Hierarchical summary receiver operator characteristic
a.
DOR = 1.0 indicates diagnostic value no greater than chance
DOR > 1 indicates increasing diagnostic value with increasing DOR > 1
b.
Q* = 0.5 indicates diagnostic value no greater than chance
Q* >0.5 indicates increasing diagnostic value up to 1 (maximum)
c.
Small studies defined as < 25 patients per study
5.1.1. Concordance with detection of GN bacteremia
Limulus assay concordance with GN bacteremia
Individual studies. The concordance between the detection of endotoxemia and
-106-
Chapter 5: Results & discussion
1
Sensitivity
.8
.6
.4
.2
0
1
.8
.6
.4
Specificity
.2
0
Note: Y axis should read ‘sensitivity’ (= true positive rate)
X axis is intentionally backwards (as specificity = 1- false positive rate)
Fig 5.1.1. HSROC Plot for the detection of endotoxaemia versus GN bacteraemia
together with the fitted SROC curve and the bivariate summary estimate (solid
square) together with the corresponding 95% confidence ellipse (inner broken line)
and 95% prediction ellipse (outer dotted line). The summary DOR (―■‖) is 5.8 (4.0
– 8.3). The symbol size for each study is proportional to the study size (n = 88).
GN bacteremia among all 88 studies is illustrated using the HSROC method in figure
5.1.1. In the figure (figure 5.1.1) there is evidence suggestive of publication bias in that
a high proportion of small studies (12 of the 19 studies that had <25 patients) had
reported 100% sensitivity for the detection of endotoxemia using the limulus assay in
Chapter 5: Results & discussion
-107-
association with the detection of GN bacteremia.
Summary value. The concordance between the detection of endotoxemia and
GN bacteremia was expressed in three of the four meta-analytic summaries as a
summary diagnostic odds ratio (DOR) value. Most recently the DOR was derived using
the HSROC method. The DOR was lower in the C-limulus studies versus the studies
using the earlier G-limulus version of the assay (Table 5.1.1). This finding is apparent
in the most recent analysis including all 88 studies. In AP&LM ‘00 the Q* value was
used as the summary value and this represents the point on the ROC summary curve
where sensitivity = specificity. The overall findings are the same in analyses in which
the summary value is expressed as either a Q* value (APLM ‘00) or as a DOR (JCM
94, EJCMID 2010) as well as in the updated summary derived for this thesis.
A further analysis for all 88 studies using the HSROC methods to derive mean
sensitivity and specificity is presented in Table 5.1.2. This table details the comparison
of the relative performance of the C-limulus versus the G-limulus version of the assay
for subgroups of studies that are both large and have a higher quality of study method
in relation to the following;
overall. The inferior performance of the C-limulus versus the G-limulus versions of
the assay remains apparent among subgroups of studies selected on the basis of
being both large and being in the strata of higher study method quality (QS>2).
patient categories. The comparison of diagnostic experience in the various sub
groups of patient categories indicates low specificity and DOR amongst studies
undertaken in an ICU setting whereas the studies undertaken in unrestricted settings
(adults) displayed low sensitivity and intermediate DOR (Table 5.1.2). Studies
undertaken in pediatric or oncology settings (analysed as a combined category),
showed overall highest specificity and DOR. Note however, all of the studies
undertaken in ICU settings had used the CLAL version of the assay.
method of plasma pre-treatment. There was no direct comparison of the various
methods of plasma pre-treatment in any single study. However, among C-limulus
assay studies, the DOR is similar among the sub-group of studies that used dilution
and heat treatment as a method of plasma pre-treatment. Conversely, among Glimulus assay studies, the sub-group of studies that used chloroform as the plasma
-108-
Chapter 5: Results & discussion
pre-treatment method performed less well.
year of publication. There was no discernable trend in DOR over time, using the
year of publication, for either version of the assay (data not shown). Also, for both
the C-limulus and the G-limulus versions of limulus assay the DOR appeared to be
unrelated to the listed sensitivity to the internal endotoxin standard of the LAL
assay as used in each study among studies (see figure 4 of EJCMID ‘10).
Table 5.1.2 Summary results for HSROC analysis (updated 2012)
N*
DOR
Sensitivity
Specificity
All
88
5.8; 4.1 – 8.2
0.73; 0.65 – 0.79
0.68; 0.60 – 0.76
Large & QS>2 †
40
4.7; 3.1 – 7.0
0.65; 0.54 – 0.75
0.72; 0.60 – 0.81
Large, QS>2 & unrestricted
13
5.3; 2.5 – 11.1
0.57; 0.41 – 0.71
0.80; 0.60 – 0.92
8
9.9; 3.6 – 27.2
0.58; 0.28 – 0.83
0.88; 0.59 – 0.97
16
2.2; 1.4 – 3.7
0.66; 0.49 – 0.80
0.53; 0.39 – 0.66
41
8.4; 4.5 – 15.7
0.76; 0.60 – 0.87
0.72; 0.59 – 0.83
13
10.1; 3.6 – 28.5
0.71; 0.32 – 0.93
0.80; 0.51 – 0.94
5
2.0; 0.3 – 12.6
0.47; 0.37 – 0.57
0.70; 0.31 – 0.92
47
4.7; 3.0 – 7.2
0.71; 0.63 – 0.79
0.65; 0.54 – 0.75
27
3.8; 2.4 – 5.7
0.65; 0.55 – 0.74
0.67; 0.66 – 0.77
Large, QS>2 & D&H†
16
3.4; 2.1 – 5.4
0.70; 0.61 – 0.80
0.59; 0.43 – 0.72
Sepsis, QS>2
11
2.2; 1.3 – 3.7
0.67; 0.52 – 0.79
0.52; 0.41 – 0.63
(adult)
Large, QS>2 & pediatric or
oncology
Large, QS>2 & ICU setting
G-limulus studies
All
Large & QS>2
†
Large, QS>2 & chloroform †
C-limulus studies
All
Large & QS>2
†
Abbreviations; * N = number of studies; DOR, diagnostic odds ratio
† Studies with more than 24 patients studied & quality score > 2 of 4
* chloroform and D&H are chloroform and dilution and heat treatment methods for the pretreatment of plasma samples, respectively.
Chapter 5: Results & discussion
-109-
5.1.2. Impact of GN bacteremia species type
The association of endotoxemia with GN bacteremia was examined for specific
bacterial types of GN bacteremia from studies for which this data was available. There
were 833 GN bacteremic patients from 57 studies, of which 15 of these studies (265
bacteremias) were restricted to a specified type of GN bacteremia (restricted studies;
Table 5.1.3). The remaining 42 studies were from studies that were unrestricted to any
specific GN bacteria (unrestricted studies; Table 5.1.4). Note that 135 of these
bacteremic patients came as a personal communication from a single study (Opal [65]).
Restricted studies. (Table 5.1.3) There were 15 studies that were restricted to
one of the four specified GN bacteremia types. Additional data for the same four GN
bacteremia types and Haemophilus influenzae were abstracted from six unrestricted
studies. The proportion with detectable endotoxemia was lowest for studies that
examined Salmonella typhi bacteremia (30%) versus Neisseria meningitidis (75%) and
Haemophilus influenzae (100%).
Table 5.1.3 Bacterial species of bacteremia and concordance with
endotoxemia: studies restricted to specified GN bacteria.
Number limulus positive (n)
/
Number of patients with GN
bacteremia (N)
Studies restricted to GN bacteria
species listed below
Studies
n
/N
%
(N)
Salmonella typhi
4
17
57
30
Haemophilus influenzae
6
10
10
100
Neisseria meningitidis
11
138
183
75
Yersiniae pestis
2
5
7
67
Burkholderia pseudomallei
1
15
15
100
Foot notes; derived from 15 studies plus small numbers from six
unrestricted studies [2, 20, 37, 38, 59, 72]
Only 5 of these studies had > 9 patients with GN bacteremia.
-110-
Chapter 5: Results & discussion
Table 5.1.4 Concordance of bacteremia with endotoxemia by bacterial
species: studies unrestricted to specific GN bacteria
Number limulus positive (n)
/
Number of patients with GN bacteremia (N)
All studies
(N=42)
Studies unrestricted to GN bacteria
All studies
(excluding Opal [65])
n
/N
%
n
/N
%
135
239
56
102
200
51
94
187
50
67
154
44
Klebsiella species
35
81
43
Enterobacter species
14
39
36
52
80
65
5
16
31
18
26
69
types
Enterobacteriaceae
Escherichia coli
Enterobacteriaceae other than E coli*
Non-Enterobacteriaceae
Pseudomonas species
76
111
68
Anaerobic bacteria†
Mixed GN bacteria
48
58
83
Foot notes; derived from 42 studies (as listed in JCM ‗94 + 4 additional studies).
Includes Serratia species, Proteus species and other unspecified Enterobacteriaceae
† Includes; Fusobacterium species, Bacteroides species).
Chapter 5: Results & discussion
-111-
Unrestricted studies. (Table 5.1.4) There were 42 studies that were unrestricted to
specified GN bacteremias.
As part of the update prepared for this thesis, the concordance of GN bacteremia
with endotoxemia among the 42 studies that were unrestricted to specified GN
bacteremias was re-examined by three separate statistical methods. The first method
(method 1) is a simple tally (Table 5.1.4). The second method (method 2) uses the
HSROC method (as used to prepare figure 5.1.1) in which the proportions of true
positives is compared with false positives among patients of each study who were blood
culture negative (as for figure 5.1.1) (figure 5.1.2/.3/.4). The third method (method 3)
is a meta-regression of logit transformed proportions of true positives (i.e. GN
bacteremias concordant with endotoxemia) in each study versus the G- and C- versions
of the limulus assay as a categorical variable. The three different methods give different
analytic weights to studies of different size with more even weighting among studies
having different size for methods two and three.
(method 1) In a simple tally (Table 5.1.3 & Table 5.1.4), the percentages are up to
9 percentage points higher than as found in an earlier tally (Table 3 of JCM’94) due
to the high proportion (84%) of endotoxemia positive GN bacteremias overall in the
largest study (Opal [65]). After excluding the bacteremias from the study by Opal
[65], the proportion with detectable endotoxemia was lowest for bacteremias with
Klebsiella (40%) and Enterobacter (36%) species, highest for Pseudomonas species (63%)
and intermediate for Escherichia coli (50%).
-112-
Chapter 5: Results & discussion
Note: Y axis should read ‘sensitivity’ (= true positive rate)
X axis is intentionally backwards (as specificity = 1- false positive
rate
E coli
1
Sensitivity
.8
.6
.4
.2
0
1
.8
.6
.4
Specificity
.2
0
Fig 5.1.2 Plot of sensitivity and specificity for all studies (n = 34) for the detection
of endotoxaemia versus E coli bacteraemia together with the fitted SROC curve and
the bivariate summary estimate (solid square) together with the corresponding 95%
confidence ellipse (inner broken line) and 95% prediction ellipse (outer dotted line).
The summary DOR (―■‖) is 3.6 (1.8 – 7.0) (Table 5.1.5). The symbol size for each
study is proportional to the study size.
Chapter 5: Results & discussion
-113-
non E coli Enterobacteriaceae
1
Sensitivity
.8
.6
.4
.2
0
1
.8
.6
.4
Specificity
.2
0
Note: Y axis should read ‘sensitivity’ (= true positive rate)
X axis is intentionally backwards (as specificity = 1- false positive rate
Fig 5.1.3 Plot of sensitivity and specificity for all studies (n = 29) for the detection
of endotoxaemia versus non-E coli Enterobacteriaceae bacteraemia together with the
fitted SROC curve and the bivariate summary estimate (solid square) together with
the corresponding 95% confidence ellipse (inner broken line) and 95% prediction
ellipse (outer dotted line). The summary DOR (―■‖) is 3.2 (1.5 – 6.8) (Table 5.1.5).
The symbol size for each study is proportional to the study size.
(method 2) By HSROC (figure 5.1.2, figure 5.1.3, figure 5.1.4). In comparing the
different types of GN bacteremia, the proportion of true positives is up to 20
-114-
Chapter 5: Results & discussion
percentage points higher for Pseudomonas aeruginosa versus Non-E. coli
Enterobacteriaceae among studies stratified by type of LAL assay. Note that the
summary DOR and sensitivity found in the HSROC analysis here are each lower
versus that found previously with all GN bacteremias included (i.e. compare Table
5.1.2 & figure 5.1.1 versus Table 5.1.5 & figure 5.1.2/.3/.4).
(method 3) By meta-regression (Table 5.1.5). The median listed assay sensitivity
among studies that used the CLAL versus GLAL assay was 0.01 ng/ml versus 3
ng/ml. Despite this 300-fold difference in listed assay sensitivity, the difference in
the proportion of true positives between the G- versus C- versions of the limulus
assay used was only 1 percentage point in the case of E. coli. In contrast, there is a
14 percentage points in the proportion of true positives between the G- versus Cversions of the limulus in the case of non-E. coli Enterobacteriaceae and
Pseudomonas aeruginosa. Note that for these latter two GN bacteremia types, the
higher sensitivity in the case of the C- version versus the G- versus of the limulus
assay demonstrated here is again contrary to what was apparent in the analysis with
all GN bacteremias included (Table 5.1.2).
The proportion of true positives (sensitivity) for the various categories of GN
bacteremias is dependent on the method of aggregation. The proportions as
presented in Table 5.1.5 are less than that derived using the simple tally method
(Table 5.1.4). For both the HSROC and the meta-regression methods there is a
more even weighting of the different sized studies than is the case for a simple tally.
The simple tally method gives most weight to the largest study. The influence of the
exclusion of the largest study in the tally is demonstrated Table 5.1.4.
Note the disparity between the sensitivity with respect to the three categories of GN
bacteremia (Table 5.1.5) which is up to 20 percentage points lower than the
sensitivity with respect to GN bacteremia overall (Table 5.1.2). The reason for this
disparity could reflect inaccuracies in how GN bacteremias were recorded in
individual studies together with the greater weighting given to smaller studies by
the HSROC method of aggregation. Given these discrepancies, the sensitivities
given in Table 5.1.5 are considered most representative of a ‗typical‘ study.
Chapter 5: Results & discussion
-115-
Note: Y axis should read ‘sensitivity’ (= true positive rate)
X axis is intentionally backwards (as specificity = 1- false positive rate
Pseudomonas aeruginosa
1
Sensitivity
.8
.6
.4
.2
0
1
.8
.6
.4
Specificity
.2
0
Fig 5.1.4 Plot of sensitivity and specificity for all studies (n = 27) for the detection
of endotoxaemia versus non P aeruginosa bacteraemia together with the fitted
SROC curve and the bivariate summary estimate (solid square) together with the
corresponding 95% confidence ellipse (inner broken line) and 95% prediction
ellipse (outer dotted line). The summary DOR (―■‖) is 6.5 (2.8 – 15.4) (Table
5.1.5). The symbol size for each study is proportional to the study size
-116-
Chapter 5: Results & discussion
Table 5.1.5 Summary results for HSROC (= method 2) and meta-regression
analysis (= method 3)
N*
DOR
Sensitivity
Specificity
All
37
3.6; 1.8 – 7.0
0.55; 0.41 – 0.68
0.75; 0.61 – 0.85
Large studies only
24
3.4; 1.6 – 6.9
0.54; 0.39 – 0.68
0.74; 0.60 – 0.85
Analysis by HSROC
E coli (see Fig 5.1.2)
other Enterobactereiaceae (see Fig 5.1.3)
All
29
3.2; 1.5 – 6.8
0.55; 0.36 – 0.72
0.73; 0.59 – 0.83
Large studies only
22
2.8; 1.4 – 5.7
0.48; 0.30 – 0.66
0.75; 0.61 – 0.86
Pseudomonas aeruginosa (see Fig 5.1.4)
All
27
6.5; 2.8 – 15.4
0.69; 0.52 – 0.81
0.75; 0.61 – 0.85
Large studies only
20
5.3; 2.3 – 12.0
0.68; 0.51 – 0.81
0.72; 0.58 – 0.82
Analysis by meta-regression versus assay type (C versus G)
E coli
GLAL
37
0.53; 0.44 – 0.61
CLAL
24
0.54; 0.46 – 0.62
GLAL
26
0.39; 0.30 – 0.48
CLAL
22
0.53; 0.44 – 0.62
GLAL
24
0.52; 0.43 – 0.62
CLAL
21
0.66; 0.57 – 0.74
other Enterobactereiaceae
Pseudomonas aeruginosa
Abbreviations; * N = number of studies; DOR, diagnostic odds ratio
GLAL – median assay sensitivity 3 ng/ml
CLAL – median assay sensitivity 0.01 ng/ml
Chapter 5: Results & discussion
-117-
5.1.3. Study limitations and literature reconciliation
Several surprising observations are noted among the findings for the detection
of endotoxemia as a diagnostic test,
1. (Section 5.2.1). The proportion of patients with GN bacteremia for which
endotoxemia is undetectable is >25% in the studies overall (Table 5.1.2), >35%
after excluding the results of small studies, up to 50% when methods of aggregation
that are more evenly weighted for studies of differing sizes (Table 5.1.5) and up to
60% when individual GN bacteremias were tallied (Table 5.1.4). The smallest
(n<25) and largest [65] studies appear to give atypical results, a finding that would
not have been apparent without using meta-analysis.
2. (Section 5.2.1). For GN bacteremias overall, the proportion with detectable
endotoxemia (i.e. sensitivity = true positives) differs by < 6 percentage points in
studies that had used the more sensitive C-limulus assay versus the original Glimulus assay (as derived using HSROC summaries; Table 5.1.2).
3. (Section 5.2.2). This lack of difference between assay versions also applies in a
comparison limited to E. coli GN bacteremias (Table 5.1.5).
4. (Section 5.2.2). By contrast, in comparisons limited to either non- E. colienterobacteriaceae or Pseudomonas, the proportion with detectable endotoxemia
(i.e. sensitivity = true positives) was up to 14 percentage points higher in studies
that used the C-limulus versus the G-limulus assay (Table 5.1.5).
5. (Section 5.2.2). The estimate for the rates of detection for different bacteremia types
which is most representative of a ‗typical‘ study within the overall literature
experience is that provided by HSROC analysis (Fig 5.1.2/.3/.4). The proportion of
GN bacteremias with detectable endotoxemia (i.e. true positives) is 68% (51-81%)
for bacteremias with Pseudomonas (Fig 5.1.4), 54% (39-68%) for E. coli (Fig
5.1.2) and 48% (30-66%) for the non- E. coli-enterobacteriaceae; Fig 5.1.3) (Table
5.1.5).
6. (Section 5.2.2). For GN bacteremias among studies restricted to bacteremias of
specific bacteremia types, only the tally method is able to be applied in the
derivation of a summary of the rate with associated endotoxemia. With these
bacteremias the rates of detection of endotoxemia is lowest (30%) amongst studies
-118-
Chapter 5: Results & discussion
that were restricted to bacteremias with S typhi and highest among studies that were
restricted to bacteremias with N. meningitidis (75%) or B. pseudomallei (100%)
(Table 5.1.3).
There are a number of limitations to this analysis. The size of the studies in
relation to the number of patients per study varies considerably. The smaller studies
showed atypical results versus larger studies (Figure 5.1.1). Among the smallest
studies (n<25), there is an excess with a sensitivity >90%. This is consistent with a
selective publication bias as it is plausible that a small study finding low sensitivity
would be less likely to be published than would a large study with this finding.
However, the conclusions of the HSROC analysis are more robust to the inclusion or
exclusion of studies of different sizes (Table 5.1.1, Table 5.1.2, Table 5.1.5).
A second limitation of this analysis is that the influence of several factors of
potential interest, for example, patient groups of different ages, underlying disease,
source of infection, or type and timing of antibiotic treatment in relation to blood
sampling has not been specifically examined. The information on these factors was not
available in many of the studies. In any case, post-hoc analysis of sub-groups in a metaanalysis is potentially misleading as the likelihood of significant findings increases by
chance alone with the number of subgroups examined.
There was a wide range of patient groups among the studies and the studies
varied with respect to indicators of quality of study design methods such as clear
specification of patient inclusion criteria and reporting of GN bacteremia subtypes. The
sub-group of studies limited to sepsis syndrome patients was examined as these studies
were generally of higher quality (Table 5.1.2). Additional sensitivity analyses (data not
shown) with the exclusion of studies of pediatric patients, the exclusion of smaller
studies and also a comparison of the diagnostic odds ratio over time (data not shown)
were done in an attempt to further address these issues. These sensitivity analyses did
not alter the conclusions (data not shown).
The selection of studies included here has been published over a period of
nearly 40 years. Blood culture methods used to detect bacteremia may have improved
over this period of time. However, the prevalence of GN bacteremia was similar
among the more recently published C-limulus studies versus the G-limulus studies.
Chapter 5: Results & discussion
-119-
The literature has been searched and extracted by a single author. While the
abstracted data and the lists of studies included and excluded are provided here (Table
7.1/.2) and in previous versions of this analysis [JCM ‘94, AP&LM ‘00,
EJCMID ’10], it always remains possible that studies have been overlooked or their
extraction contestable. In this respect, additional small [2, 17, 74] and non-English
language [34, 92] studies and the largest study (Opal [65]) became available in
preparing this thesis. The addition of these newly found studies did not substantially
change the findings of the earlier published analyses (Table 5.1.1).
There would need to be a substantial number of studies that had been
overlooked or extracted imprecisely to alter the conclusions in this analysis. In
particular, for the C-limulus assay to have an improved utility to the G-limulus assay
there would need to be a substantial number of studies with a high DOR, exceeding for
example, those seen with the studies of procalcitonin [294, 295] (Table 2.2.2 and 2.5.1),
to negate the findings here. Such studies would be unlikely to have remained
unpublished or difficult to find.
The proportion of endotoxemia positive bacteremias for specific types of GN
bacteremia differs dependent on whether this proportion is derived as a simple tally
(Tables 5.1.3 & 5.1.4) or by HSROC meta-regression methods (Table 5.1.5). Of these
methods, the HSROC method is considered to best represent the findings of a ‗typical‘
study. A further limitation is that the number of samples evaluated per patient in each
study is not known and cannot be evaluated.
To further seek an overview and reconciliation of the entire literature analysed
here, the studies representative of specific patient groups are indicated by a key as
coded in Table 5.1.6 and presented on the summary HSROC plot in Figure 5.1.5 with
the large studies coded for identification. Both the generally low specificity (< 60%)
among several of the studies undertaken in an ICU setting [i1 – i9] and the generally
low sensitivity (< 80%) among several of the studies undertaken with unrestricted adult
patients [a0 – a11] can be visually appreciated. Also shown for reference purposes are
two supplementary list studies [@; 100, 101] which used non-limulus assays with study
data available in the same ‗4 groups‘ format as for the other 88 studies that had used the
limulus assay.
-120-
Chapter 5: Results & discussion
Table 5.1.6 Details of selected (QS>2) studies (& codes for figure 5.1.5)
code Author
year
ref
N
FP
%
TP
%
DOR
Ec En
(n)
Ps Tot
100
18
22
42
133
93
356
52
759
28
40
50
93
12
69
33
36
57
80
44
58
67
100
6
75
73
20
93
84
50
2.1
1.8
0.7
0.6
1.3
5.1
0.5
7.4
1.4
1.2
9 5
2 2
0 1
3 11
2 4
16 5
23 27
11 3
39 33
8 1
2
2
2
1
2
0
5
4
31
1
19
6
3
16
12
31
58
18
135
10
27
48
25
40
218
83
30
117
473
448
27
20
6
31
69
10
1
10
35
3
19
24
67
75
75
73
55
6
50
43
32
50
66
5.5
37.5
5.7
1.1
10.4
4.3
7.4
1.4
13.4
4.2
4.8
8 1
8 0
10 2
8 1
7 16
4 3
9 0
18 28
9 4
17 15
0 4
0
2
0
0
7
1
1
6
0
2
1
12
12
12
11
30
14
10
63
15
40
5
52
50
35
33
70
11
0.6
2.4
1.8
4.6
27.6
2
90
7
92
90
9.9
9.4
29
24.4
GN bacteremias
Intensive care unit
i1
i2
i3
i4
i5
i6
i7
i8
i9
i0
Danner
Dofferhoff
Fink
Garcia Curiel
Goldie
Guidet
Ketchum
Massignon
Opal
Strutz
1991
1992
1984
1979
1995
1994
1997
1996
2000
1999
Unrestricted adult patients
a0
a1
a2
a3
a4
a5
a6
a7
a8
a9
a11
Byl
Clumeck
G-bourboulis
Lau
Levin
Martinez-G
Prins
Stumacher
Van Deventer
VanLangevelde
Watzke
2002
1977
1999
1996
1972
1973
1995
1973
1988
2000
1987
Oncology & liver transplant
o1
o2
o3
o4
Bion
Elin
Engervall
Yoshida
Hynninen
1994
1975
1997
1994
1995
133
24
136
123
62
19
22
33
0
2004
1974
1985
1988
41
83
43
26
39
0
19
19
0
0
2
NS
NS
NS
1 1
1 11
13 10
3
13
3
6
30
26
1
9
3
1
3
12
2
1
8
28
5
9
Pediatric
p1
p2
p3
p4
Ahmed
Feldman
Cooperstock
Shenep
0
4
0
2
Restricted
t
Adinolfi
1987
21
29
50
2.2
mg
Brandtzaeg
1989
42
0
69
32
ml
Simpson
2000
30
67
100
16
Foot notes;
Only shown are studies with QS>2, N>20 and GN bacteremia types available.
Note, for figure 5.1.1, specificity = 1-FP
Chapter 5: Results & discussion
1
ml
p3
p4
i9l
@
a1
mg
a0a11
a2
i6
o3
.6
a4
a6 a9
.4
i5
a3
i2
@
i1
t
i3
i8
p1
.8
Sensitivity
-121-
i0
i9h
a7
o1o2
a8
.2
i7
o4a5
p2
i4
0
1
.8
.6
.4
Specificity
.2
0
Note: Y axis should read ‘sensitivity’ (= true positive rate)
X axis is intentionally backwards (as specificity = 1- false positive rate
Fig 5.1.5 Plot of sensitivity and specificity for all studies (as in Figure 5.1.1) with
study codes as indicated in Table 5.1.6. Symbol size for each study is proportional
to the study size. Hollow ―○‖symbols (n=90) are studies using the limulus assay,
and ―@‖ symbols are 2 non-limulus ‗4 group‘ supplementary studies [100, 101].
The fitted SROC curve and the bivariate summary estimate (solid square) together
with the corresponding 95% confidence ellipse (inner broken line) and 95%
prediction ellipse (outer doted line). The summary DOR (―■‖) is 5.8 (4.0 – 8.3).
Note the asymmetrical distribution of studies in an ICU setting (i1, i2, i3, i5, i6, i8,
i9l, & i0) with most studies falling on right hand side (i.e. specificity <60%) of
figure versus studies in other settings.
-122-
Chapter 5: Results & discussion
There are three possible explanations for why the proportion of endotoxemia
positive (i.e. labelled here as false positive) patients among patients without a GN
bacteremia is generally higher among studies undertaken in an ICU setting (as noted in
Figure 5.1.5). Firstly, all of the studies undertaken in ICU settings had used the CLAL
version of the assay. Second, it is likely that the patient population outside of an ICU
setting is at overall lower risk of having endotoxemia and that a lower proportion of
false positives could be expected. Thirdly, it is possible that the types of non GN
bacteremic infections giving rise to endotoxemia (i.e. labelled here as false positive)
may differ for the different patient populations. This explanation has not been explored
here. However, a difference in the distributions of types of infections among the patient
groups with GN bacteremia is to be explored in sections 5.2.2 & 5.2.3.
Overall, the findings presented within section 5.1 imply that the type of GN
bacteremia is at least as important a determinant toward the detection of concurrent
endotoxemia as is the sensitivity of the assay method used. The different results with
the three different methods of aggregation reflect the different relative weighting
provided to the smaller versus larger studies. The tally (first) method, as had been used
in the first attempt (JCM ‘94) to summarize the relationship between endotoxemia
detection and GN bacteremia type is not optimal as it provides disproportionate
weighting to the largest studies.
Chapter 5: Results & discussion
-123-
5.2. Endotoxemia detection as a prognostic test
The findings for the detection of endotoxemia as a test of prognosis are
presented here. The patient outcome of interest is mortality. As indicated within Figure
5.0.1 there are two types of studies with outcome data examined here, ‗four group‘
studies and ‗two group‘ studies as follows;
‘four group’ studies, being studies in which the patient outcome (mortality) was
reported for patients categorized into one of four groups;
o group 1; both endotoxemia and GN bacteremia detected
o group 2; GN bacteremia detected in isolation
o group 3; endotoxemia detected in isolation
o group 4; neither detected (reference group).
‘two group’ studies (amongst the supplementary studies as indicated within Figure
5.0.1) are studies using either limulus or non-limulus assays for endotoxemia
detection in which the patient mortality was reported for patients categorized into
one of two groups;
1. endotoxemia detected (equivalent to group 1 & 3 combined)
2. endotoxemia not detected (equivalent to group 2 & 4 combined)
(reference group).
The studies of primary interest are those identified in the literature search
(Figure 5.0.1) as ‗four group‘ studies. Note that the data within ‗four group‘ studies can
also be extracted in a ‗two group‘ format but not vice versa. For this reason, several
‗two group‘ and ‗four group‘ studies undertaken in an ICU setting are to be compared
as a preliminary analysis to place the ‗four group‘ studies within the wider literature
experience. Hence, the analysis proceeded as follows.
1. All studies undertaken in an ICU setting (‗four group‘ and ‗two group‘) were
extracted for analysis in the ‗two group‘ data format (see 5.2.1).
2. All ‗four group‘ studies were extracted for analysis in the ‗two group‘ data
format to enable a comparison of those undertaken in ICU settings versus
settings other than ICU (see 5.2.1).
3. An analysis of the outcome data was undertaken for only the ‗four group‘
-124-
Chapter 5: Results & discussion
studies extracted and analysed in the ‗four group‘ data format (see 5.2.2).
4. An analysis of patient outcome in relation to both endotoxemia detection
(equivalent to group one versus group two from the ‗four group‘ studies) and in
relation to the GN bacteremia isolate type for those patients for which this data
were available (see 5.2.3).
5. Studies with potential outlier outcome data were examined (see 5.2.4).
The numbers of studies, publications and patients for the analysis of all studies
analysed as ‗two group‘ studies are summarized in Table 5.2.1 with selected studies
highlighted in Table 5.2.2. The numbers of studies, publications and patients for the
analysis of all ‗four group‘ studies are summarized as included in the three publications
(JCM ‘95, JER ’03, CC ‘12) in Table 5.2.3. In preparing the most recent update,
seventeen additional ‘four group’ studies have been located including two in an ICU
setting [65, 74] with new data obtained by personal communication for three [17, 21,
65]. The studies for which outcome data was available are indicated by ψ under column
QS in Tables 7.1/.2.
Characteristics of the ‘two group’ studies There are six ‗two group’ studies
undertaken in an ICU setting with outcome data available (indicated by ψ under
column QS in the summary Tables 7.2.2) (Table 5.2.1). This includes supplementary
list studies that had used the LAL assay but for which outcome data was available only
for the ‗two group‘ format [102], studies that had used assays other than the limulus
assay to detected endotoxemia [101, 104-107] and two studies in which patients had
received anti-endotoxin antibody treatments [88, 94].
Characteristics of the ‘four group’ studies Thirty-five studies were included
from 32 publications of which 13 studies were supplemented with data provided by
personal communication (Table 5.2.3). The survival outcome was reported for a total
of 3235 patients among these 35 studies (indicated by ψ under column QS in Tables
7.1) of which 432 (13%), 272 (8%), 1091 (34%) and 1440 (44%) were in groups 1 to 4,
respectively. Patient inclusion for 9 of the studies was based on various criteria for
sepsis in adult patients in an ICU setting. Twenty-six of the 35 studies were published
within the 1980‘s and 1990‘s. Studies were excluded if any of the following applied;
mortality proportion was not obtainable from group 4 (21 studies), the patients had
received anti-endotoxin antibody treatments (2 studies), no patients in the study had GN
Chapter 5: Results & discussion
-125-
bacteremia (5 studies) or the data was not extractable into a 2x2x2 table format. The
largest study (Opal [65]) provided data in relation to endotoxemia detection at two
breakpoints which have been used here to derive mortality proportions stratified for
endotoxemia detection at high and low levels from this study.
5.2.1. Analysis of ‘4’ & ‘2’ group studies as ‘two group studies’
Table 5.2.1 Analyses of all studies with data extracted in ‗two group‘ (endotoexemia
detected versus not detected) format
Type of study
‗2 group‘ studies
‗4 group‘ studies
6
32 *
1993 – 2002
1972 - 2000
1210
3377
MH method
MH method
Yes
Yes
1.6; 1.2 – 2.0
(6)
3%
1.4; 1.2 – 1.7
(14)
0%
Details of meta-analysis
Number of publications
Publication years
Number of patients
Meta-analysis method
Weighting for study size?
Sub-group analysis; ICU setting
OR; 95% CI (n)
heterogeneity (I2)
unrestricted adult
OR; 95% CI (n)
heterogeneity (I2)
2.8; 2.1 – 3.9
(11)
77%
oncology
OR; 95% CI (n)
heterogeneity (I2)
5.9; 3.0 – 11.4
(3)
0%
pediatric
OR; 95% CI (n)
heterogeneity (I2)
Abbreviations; OR = Odds Ratio, MH = Mantel Haenzsel
2.8; 1.0 – 7.6
(4)
8%
* Note: studies not included here are;
restricted studies (5 studies)
studies with ‗double zero entries for which the OR is indeterminant (11 studies)
studies in which all patients received HA-1A [88, 94].
-126-
Chapter 5: Results & discussion
Forest plots (OR & RD)
The association with mortality for endotoxemia detection versus non-detection
amongst all 49 ‗four‘ and ‗two‘ group studies is presented as an odds ratio (OR; Table
5.2.1, Figure 5.2.1) and as a risk difference (RD; Figure 5.2.2). Note that the OR is
indeterminate for studies with double zero entries (e.g. Cooperstock ‘85; 0/16 versus
0/27), hence there are fewer studies appearing in the OR versus the RD forest plots. The
summary OR over all studies was 1.7 (1.4 – 1.9) (Figure 5.2.1, Table 5.2.1) and the
risk difference over all studies was 9.3% (7.2 – 11.4; Figure 5.2.2). However, both the
overall summary OR and overall summary RD were each associated with significant
heterogeneity (OR; I2 = 43%) (RD; I2 = 50%). This heterogeneity is visually apparent
in the non-overlap in summary 95% confidence intervals associated with the summary
values for patient groups in non-ICU versus in ICU settings as displayed in both the OR
(Figure 5.2.1) and RD (Figure 5.2.2) forest plots.
In both the OR (Figure 5.2.1) and the RD (Figure 5.2.2) forest plots, the
association with mortality for endotoxemia detection versus non-detection among the
individual studies undertaken in non-ICU settings is very heterogeneous. For example,
the range in RD amongst all studies extends from a more than 40% excess to a more
than 10% deficit. The spread of RD is reflected in the heterogeneity statistic (I2) which
is greater than 60% amongst the sub-group of studies of unrestricted adults (I2=77%),
the sub-group of oncology patients (I2=67%) and the sub-group of restricted studies
(I2=75%) (Figure 5.2.2).
By contrast, in both the OR (Figure 5.2.1) and the RD (Figure 5.2.2) forest
plots, the association with mortality for endotoxemia detection versus non-detection
among the studies undertaken in an ICU setting is minimal and the heterogeneity
statistic (I2) is less than 10% both in the case of the two group (0%) and the four group
studies (0%) conducted in an ICU (Table 5.2.1; Figure 5.2.1).
Chapter 5: Results & discussion
-127-
2&4 group studies odds ratio
Author
ICU4
Wortel '92 HA-1A
Goldie '95
Strutz
Opal-lo
Fugger '90
Danner '91
Billard '94
Ketchum '97
Opal-high
Dofferhoff '92
Guidet '94
Wortel '92 placebo
Watzke '87
Van Dissel '93 HA-1A
Subtotal (I-squared = 0.0%, p = 0.964)
.
ICU2
Casey '93 LAL
Venet '00 LAL
Marshall '02 EAA
Yaguchi '12 EAA
Kollef '97' SRE
Valenza '09 EAA
Subtotal (I-squared = 3.2%, p = 0.396)
.
Unrestricted
Bion '94 all
Stumacher '73
Lau '96
Bailey '76
Byl '01
Foulis '82
van Langevelde '00
Levin '72 all
G-bourboulis '99
Martinez-G '73
Das 73
Berger '95
Prins '95
Lumsden '89
Subtotal (I-squared = 68.8%, p = 0.000)
.
Oncology
Hynninen '95
Yoshida '93
Engervall '97
Subtotal (I-squared = 0.0%, p = 0.597)
.
pediatric
Togari '83
Ahmed '04
Feldman '74
Shenep '88
Cooperstock '85
Casey '92
Klein '88
Subtotal (I-squared = 8.0%, p = 0.353)
.
restricted
Butler '76
Simpson '00
Suyasa '95
Brandtzaeg '89
Magliulo '76
Magliulo '76
Subtotal (I-squared = 0.0%, p = 0.680)
.
Overall (I-squared = 42.5%, p = 0.002)
.1
.3
etx -; less survival
1
3
10
30
50
etx+; less survival
Fig 5.2.1 Forest plot (OR). Odds ratios for mortality for endotoxemia detected
(Groups 1 & 3 combined) versus endotoxemia not detected (groups 2 & 4
combined) presented as study specific and summary risk difference (and 95%
confidence intervals) derived from all 49 studies with studies sorted into those
conducted in an ICU versus other settings. Overall summary OR indicated by
vertical dashed line. Arrowheads indicate 95% confidence intervals that extend out
of range. ICU2 refers to ‗two group‘ supplementary studies undertaken in an ICU
setting. All other studies are ‗4 group‘ studies. Studies for which the OR is
indeterminate do not appear in this figure.
-128-
Chapter 5: Results & discussion
There are several notable OR and RD observations for individual studies in
Figures 5.2.1 & 5.2.2. The studies that used non-limulus assays such as the NCL assay
[100, 101, 105-107] and the SRE assay [104] have similar OR and RD to that observed
among the studies that had used the LAL assay. The two groups that had received HA1A anti-endotoxin antibodies (indicated by HA-1A next to author name in Figures
5.2.1/.2 [88, 94]) appear at opposite ends of the list of ICU studies ranked either by OR
and RD. However as the 95% confidence intervals for both studies overlap the common
result neither of these two studies can be regarded as being an outlier observation by
statistical criteria (see below).
Chapter 5: Results & discussion
-129-
2&4 group studies Risk Difference
Author
ICU4
Wortel '92 HA-1A
Goldie '95
Strutz
Opal-lo
Fugger '90
Danner '91
Dofferhoff '92
Ketchum '97
Opal-high
Billard '94
Wortel '92 placebo
Guidet '94
Watzke '87
Van Dissel '93 HA-1A
Subtotal (I-squared = 0.0%, p = 0.937)
.
ICU2
Casey '93 LAL
Venet '00 LAL
Marshall '02 EAA
Yaguchi '12 EAA
Valenza '09 EAA
Kollef '97' SRE
Subtotal (I-squared = 0.0%, p = 0.904)
.
Unrestricted
Bion '94 all
Stumacher '73
Lumsden '89
Prins '95
Berger '95
Lau '96
Bailey '76
van Langevelde '00
Byl '01
G-bourboulis '99
Foulis '82
Levin '72 all
Martinez-G '73
Das 73
Subtotal (I-squared = 77.4%, p = 0.000)
.
Oncology
Hynninen '95
Engervall '97
Yoshida '93
Subtotal (I-squared = 66.8%, p = 0.049)
.
pediatric
Togari '83
Casey '92
Cooperstock '85
Klein '88
Ahmed '04
Shenep '88
Feldman '74
Subtotal (I-squared = 14.1%, p = 0.322)
.
restricted
Magliulo '76
Magliulo '76
Butler '76
Suyasa '95
Simpson '00
Brandtzaeg '89
Subtotal (I-squared = 59.9%, p = 0.029)
.
Overall (I-squared = 47.9%, p = 0.000)
-.3
-.2
-.1
etx -; less survival
0
.1
.2
.3
.4
.5
etx+; less survival
Fig 5.2.2 Forest plot (RD). Risk difference for mortality for endotoxemia detected
(Groups 1 & 3 combined) versus endotoxemia not detected (groups 2 & 4
combined) presented as study specific and summary risk difference (and 95%
confidence intervals) derived from all 49 studies with studies sorted into those
conducted in an ICU versus other settings. Overall summary OR indicated by
vertical dashed line. Arrowheads indicate 95% confidence intervals that extend out
of range. ICU2 refers to ‗two group‘ supplementary studies undertaken in an ICU
setting. All other studies are ‗4 group‘ studies. See Fig 5.2.3 for corresponding
L‘abbe plot.
-130-
Chapter 5: Results & discussion
mortality (%)
endotoxemia detected versus not detected
70
60
50
40
a5
o3
mg
30
p4
20
i6
a4
ii1
i5
ml
a11
i7
ii3
i9h
ii2 i1 i9l
ii5
p1i2
a0
o2 a9 ii4
a2
a7
10
a3
0
0
10
20
30
40
50
60
Mortality group 2 & 4 combined
Note: Y axis should read ‘Mortality in groups 1 & 3 combined’
Fig 5.2.3 L‘abbe plot of mortality for groups with endotoxemia detected (Groups 1
& 3 combined) versus groups with endotoxemia not detected (Groups 2 & 4
combined) (see Table 5.2.2 for study codes). The line is x=y and represents the line
of equivalence shown for visual reference purposes. Note that there are eight studies
with 0% mortality in all groups.
Chapter 5: Results & discussion
-131-
Table 5.2.2
Details of selected (QS>2) ‗two‘- & ‗four‘-group studies (& codes for fig 5.2.3)
code Author
year
ref
N
Mortality proportions
ETX +
%
ETX - %
RD
%
Two-group studies (all are Intensive Care Unit setting)
ii1
ii2
ii3
ii4
ii5
ii6
Casey
Kollef *
Venet
Marshall *
Yaguchi*
Valenza*
1993
1997
2000
2004
2012
2009
102
104
91
106
101
107
97
140
106
857
210
102
11/22
8/29
25/61
45/262
28/106
2/17
50
28
41
17
26
12
36/75
17/114
21/55
70/595
18/104
0/102
48
15
38
12
17
0
2
13
3
5
6
12
12/57
1/8
20/41
22/49
71/237
17/30
21
13
49
45
30
26
25
47
57
7
8
-4
16
10
3
10
-1
16
Four-group studies (data extracted in 2 group format)
i1
i2
i5
i6
i7
i9l
i9h
i0
Intensive care unit setting
Danner
1991
Dofferhoff
1992
Goldie
1995
Guidet
1994
Ketchum
1997
Opal – low
2000
Opal - high
2000
Strutz
1999
Wortel
1992
21
23
37
38
47
65
65
74
94
100
18
133
93
356
455
432
12/43
2/10
41/92
27/44
47/119
28
90/306
105/304
6/13
41
8/11
28
20
45
61
39
29
35
46
73
27
25
40
218
117
448
27
2/11
1/7
1/28
21/36
8/46
16/100
5/12
18
14
4
58
17
16
42
1/16
0/18
0/12
31/182
19/71
16/348
3/15
6
0
0
17
27
5
20
12
14
4
41
-9
12
22
39/149
32/128
7/15
a0
a2
a3
a4
a7
a9
a11
Unrestricted adult patients
Byl
2002
G-bourboulis
1999
Lau
1996
Levin
1972
Stumacher
1973
VanLangevelde
2000
Watzke
1987
o2
o3
Oncology & liver transplant
8
Bion
1994
26
Engervall
1997
95
Yoshida
1994
52
24
136
1/32
1/6
25/56
3
17
45
4/20
0/18
9/80
20
0
11
-17
17
33
p1
p4
Pediatric
Ahmed
Shenep
41
26
4/21
3/12
19
25
2/20
0/14
10
0
9
25
2004
1988
17
36
52
54
75
89
92
2
72
Restricted
1
t
Adinolfi
1987
21
0/9
0
0/12
0
0
11
mg
Brandtzaeg
1989
42
9/24
38
1/18
3 35
73
ml
Simpson
2000
9/15
60
1/5
20 40
30
Note * Non-LAL assays; Marshall 2002 (ii4), neutrophil chemoluminesence assay; Kollef
(ii2), SRE assay.
Kollef 1997 [104 (ii2)] reported the LAL assay results was not predictive of mortality (data
not available)
-132-
Chapter 5: Results & discussion
L’abbe plot
The L‘abbe plot of mortality proportions in the patient groups with endotoxemia
detected versus in patient groups with endotoxemia not detected for each of the 49
studies is presented in Figure 5.2.3. Selected studies have been labelled for
identification (see Table 5.2.2 for codes). Both the forest plot (Figure 5.2.2) and the
L‘abbe plot (Figure 5.2.3) illustrate that the difference in mortality proportion for
patients with endotoxemia detected versus not detected was in the range 0 to 20% for
33 studies (see also column headed RD in Table 5.2.2). The mortality difference
exceeded 20% for only ten studies and was negative (<0%; i.e. mortality excess among
patients without endotoxemia detected) for six studies.
There are eight studies in which the mortality proportion was 0% in both the
groups of patients with endotoxemia detected and with endotoxemia not detected. As a
result there are eight studies in the scatter plot which coincide at the origin.
Note in Figure 5.2.3 that the mortality rates among ICU studies is typically
higher (i.e. >15%) versus in the non-ICU studies (<15%), as expected. This dichotomy
is further explored in section 5.2.2.
There are 18 studies which have a RD either >20% (n=10) or ≤0% (n=8). Of
these 18 studies, four studies (a4, a9, mg, o3, ii6) are noted here as being statistical high
outliers in either Figure 5.2.1 or 5.2.2 in that the studies 95% confidence intervals
associated with either the RD or the OR fail to include the common result in the forest
plots. Likewise three studies (i5, a7, Bion) are noted here as being statistical low
outliers. These 18 studies are to be discussed in section 5.2.4.
Chapter 5: Results & discussion
-133-
5.2.2. Analysis of ‘four group’ studies
Table 5.2.3 Summary results for analyses of prognostic studies
(four group studies only)
JCM ‗95
JER ‗03
CC ‗12
11
15
32
1972 – 1994
1972 - 2000
1972 - 2009
738
1937
3235
MH method
SROC
D-L
Yes
Yes
Yes
OR;
95% CI a
OR;
95% CI a
OR;
95% CI a
11
7.1;
4.2 – 12.0
2.0;
1.0 – 3.8
2.3;
1.3 – 4.0
16
3.6;
2.1 – 6.3
2.2;
0.9 – 5.8
2.0;
0.8 – 4.8
35
3.1;
2.0 – 4.8
1.3;
0.9 – 1.9
1.5;
1.2 – 1.8
Details of meta-analysis
Number of publications
Publication years
Number of patients
Analytic method
Weighting for study size
Summary measure
All studies
Number of studies
Group 1 v group 4
Group 2 v group 4
Group 3 v group 4
ICU setting
ICU setting
7
9
Number of studies
2.5;
1.5;
Group 1 v group 4
1.6 – 3.9
1.01 – 2.1
1.7;
1.2;
Group 2 v group 4
1.2 – 2.6
0.74 – 2.0
1.4;
1.4;
Group 3 v group 4
0.8 – 2.5
1.09 – 1.8
Foot notes; Abbreviations; OR = Odds Ratio, MH = Mantel Haenzsel, D-L =
Dersimonian and Laird
Note: sub-groups are
group 1; endotoxemia and GN bacteremia detected
group 2; only GN bacteremia detected
group 3; only endotoxemia detected
group 4; neither endotoxemia or GN bacteremia detected (reference group)
Sub analysis -
-134-
Chapter 5: Results & discussion
Forest plots
The data extracted from the studies used in the analysis presented here is as
summarized within Table 1 of CC ’12. As in section 5.2.1, the analysis here examines
both the OR and RD meta-analysis. There is no substantive difference between the
analyses using either the RD or the OR. The RD analysis is presented in the figures
here (Figure 5.2.4/.5/.6) and that using the OR is presented in the figures in the
compiled publication (Figures 2/3/4 of CC ’12).
For the 32 ‗four-group‘ studies, the summary odds ratio (OR) for death for
group 1, group 2 and group 3 each respectively versus group 4, are summarized in
Table 5.2.3. The RD forest plots are presented in Figure 5.2.4 (group 1 versus group 4),
Figure 5.2.5 (group 2 versus group 4) and Figure 5.2.6 (group 3 versus group 4) and
the L‘abbe plots are presented in Figure 5.2.7. The appearance of the forest plots differ
to those displayed in CC ’12 because the categories here are labelled ―icu‖ and ―nonicu‖ and the studies have been sorted and ranked in order of study RD.
The species types of GN bacteremia isolates were identified for 31 of the 35
studies. Among the mono-microbial GN bacteremias, there were 174 (26%), 134 (22%),
74 (12%), and 94 (15%) bacteremias with E. coli; Enterobacteriaceae other than E. coli
(e.g. Klebsiella species, Enterobacter species); P. aeruginosa; and N. meningitidis,
respectively. After excluding studies restricted to specified infections, there were 497
GN bacteremias for which the species type was known. Among these 497 GN
bacteremias, there was an uneven distribution of E. coli versus P. aeruginosa identified
among the GN bacteremias of group 1 versus group 2; E. coli was less common in
group 1 than group 2 (92 of 303, 30% versus 82 of 194, 42%, p = 0.007; chi-square
test). By contrast, P. aeruginosa was more common in group 1 than group 2 (53 of 303,
17% versus 23 of 194, 12%, p = 0.09; chi-square test). This reciprocal mal-distribution
was also apparent among the 9 studies of adults with sepsis in an ICU setting (data not
shown).
Chapter 5: Results & discussion
-135-
group1 versus group4; risk difference
author
icu
Strutz
Opal (low)
Goldie
Opal (high)
Guidet
Danner
Bates
Wortel (placebo)
Billard
Dofferhoff
Subtotal (I-squared = 0.0%, p = 0.698)
.
non-icu
Bion
Lumsden
Casey
Hynninen
Berger
Magliulo & Adinolfi (Typhoid)
Magliulo (Salmonella)
Cooperstock
Lau
Butler & Butler (plague)
Stumacher
Suyasa (Typhoid)
Shenep
Byl
Giamarellos
Brandtzaeg(meningococcal)
Van Langevelde
Ahmed
Brandtzaeg(meningococcal)
Engervall
Levin
Yoshida
Bailey
Foulis
Subtotal (I-squared = 66.5%, p = 0.000)
.
Overall (I-squared = 58.5%, p = 0.000)
NOTE: Weights are from random effects analysis
-.4
-.2
0
.2
.4
.6
.8
group 4 mortality higher <=> group 1 mortality higher
Fig 5.2.4 Forest plot (RD). Risk difference for mortality for endotoxemia and GN
bacteremia co-detected (Group 1) versus neither detected (Group 4) presented as
study specific and summary risk difference (and 95% confidence intervals) derived
with studies sorted into those conducted in an ICU versus other settings. Overall
summary OR indicated by vertical dashed line. Arrowheads indicate 95%
confidence intervals that extend out of range. See Fig 5.2.7a for corresponding
L‘abbe plot.
Pooled RD for ICU studies is 7.4% (-0.4 – 15.1)
Pooled RD for non-ICU studies is 18.7% (7.7 – 29.8)
Pooled RD for all studies is 16.1% (8.1 – 24.1)
-136-
Chapter 5: Results & discussion
group2 versus group4; risk difference
author
icu
Danner
Dofferhoff
Guidet
Bates
Opal (low)
Goldie
Strutz
Wortel (placebo)
Subtotal (I-squared = 29.9%, p = 0.189)
.
non-icu
Bion
Butler & Butler (plague)
Brandtzaeg(meningococcal)
Brandtzaeg(meningococcal)
Klein
Lau
Giamarellos
Byl
Shenep
Magliulo & Adinolfi (Typhoid)
Engervall
Hynninen
Van Langevelde
Cooperstock
Yoshida
Suyasa (Typhoid)
Levin
Stumacher
Ahmed
Subtotal (I-squared = 31.3%, p = 0.095)
.
Overall (I-squared = 28.4%, p = 0.086)
NOTE: Weights are from random effects analysis
-.4
-.2
0
.2
.4
.6
.8
group 4 mortality higher <=> group 2 mortality higher
Fig 5.2.5 Forest plot (RD). Risk difference for mortality for GN bacteremia
detected alone (Group 2) versus neither detected (Group 4) presented as study
specific and summary risk difference (and 95% confidence intervals) derived with
studies sorted into those conducted in an ICU versus other settings. Arrowheads
indicate 95% confidence intervals that extend out of range. Overall summary OR
indicated by vertical dashed line. See Fig 5.2.7b for corresponding L‘abbe plot.
Pooled RD for ICU studies is -0.3 (-12.8 - 12.2)
Pooled RD for non-ICU studies is -2.1% (-8.6 – 4.4)
Pooled RD for all studies is -1.8% (-7.4 – 3.7)
Chapter 5: Results & discussion
-137-
group3 versus group4
author
icu
Billard
Dofferhoff
Goldie
Danner
Opal (low)
Bates
Opal (high)
Wortel (placebo)
Strutz
Guidet
Subtotal (I-squared = 0.0%, p = 0.569)
.
non-icu
Bion
Butler & Butler (plague)
Casey
Lumsden
Klein
Giamarellos
Lau
Berger
Byl
Bailey
Magliulo & Adinolfi (Typhoid)
Magliulo (Salmonella)
Engervall
Cooperstock
Stumacher
Ahmed
Van Langevelde
Foulis
Brandtzaeg(meningococcal)
Yoshida
Shenep
Levin
Subtotal (I-squared = 16.5%, p = 0.241)
.
Overall (I-squared = 14.0%, p = 0.243)
NOTE: Weights are from random effects analysis
-.4
-.2
0
.2
.4
.6
group 4 mortality higher <=> group 3 mortality higher
Fig 5.2.6 Forest plot (RD). Risk difference for mortality for endotoxemia detected
alone (Group 3) versus neither detected (Group 4) presented as study specific and
summary risk difference (and 95% confidence intervals) derived with studies sorted
into those conducted in an ICU versus other settings. Overall summary OR indicated
by vertical dashed line. Arrowheads indicate 95% confidence intervals that extend
out of range. See Fig 5.2.7c for corresponding L‘abbe plot.
Pooled RD for ICU studies is 6.8% (1.7 – 11.8)
Pooled RD for non-ICU studies is -0.7% (-5.8 – 4.4)
Pooled RD for all studies is 2.3% (-1.6 – 6.1)
-138-
Chapter 5: Results & discussion
Figure 5.2.7a
group1 versus group4
‗Percent 100
mortality 90
in group 1
Percent mortality group 1
80
70
60
50
40
30
20
10
0
0
5
10
15
20
25
30
35
Percent mortality group 4
40
45
50
40
45
50
40
45
50
group2 versus group4
Figure 5.2.7b
Percent mortality group 2
‗Percent
mortality
in group 2‘
100
90
80
70
60
50
40
30
20
10
0
0
10
15
20
25
30
35
Percent mortality group 4
group3 versus group4
Figure 5.2.7c
‗Percent
mortality
in group 3‘
Percent mortality group 3
5
100
90
80
70
60
50
40
30
20
10
0
0
5
10
15
20
25
30
35
Percent mortality group 4
Fig 5.2.7. L’Abbé plots of study specific mortality rates. Each figure shows
mortality rates for studies undertaken in an ICU (triangles) or non-ICU (circles) setting
with symbols proportional to group size with the line of equivalence (y = x; dashed
line) shown for visual reference purposes.
Shown are Groups 1 (Endotoxemia & GN bacteremia co-detected; Figure 5.2.7a - top),
Groups 2 (GN bacteremia alone; Figure 5.2.7b - middle), and Groups 3 (Endotoxemia
alone; Figure 5.2.7c - bottom) each versus Groups 4 (neither detected).
Chapter 5: Results & discussion
-139-
The summary odds ratio (OR) for death for Group 1, Group 2 and Group 3 each
respectively versus Group 4, are presented in Table 5.2.3 for all studies and for the
studies undertaken in an ICU setting as a sub-analysis. The summary RD and I2 are
presented in the legends of figures 5.2.4/.5/.6. With only the nine ICU studies
considered, the summary odds ratios were all either not significant or borderline (OR <
2). These findings are also apparent amongst the corresponding summary RD being less
than 10% (figures 5.2.4/.5/.6).
With the 26 studies undertaken outside of an ICU setting considered, the
summary odds ratio for Groups 1 (co-detection of endotoxemia with GN bacteremia)
versus Group 4 was 6.9 (4.4-11.0) whereas the summary odds ratio (OR) for Groups 2
(GN bacteremia alone) and groups 3 (endotoxemia alone) versus Group 4 (neither)
were not significant or borderline (OR < 2) (data shown in Table 2 of CC ‗12). These
findings are also apparent amongst the corresponding overall summary RD being less
than 10% (Figures 5.2.5/.6) with the exception of Group 1 versus Group 4 (Figures
5.2.4).
The heterogeneity amongst the studies was most apparent in the analysis of
Groups 1 (co-detection of endotoxemia with GN bacteremia) versus Group 4 from all
35 studies (OR: I2=42%; as shown in Table 2 of CC ‘12) and particularly so for studies
undertaken outside of an ICU setting (RD I2=67%; as shown within Figure 5.2.4).
Otherwise, for all other comparisons; Group 2 versus Group 4 and Group 3 versus
Group 4, the heterogeneity amongst the studies was minimal whether calculated as OR
(I2=0%; as shown in Table 2 of CC ‗12) or RD (I2<35%; as shown within Figures 5.2.4,
5.2.5, & 5.2.6).
In comparing the prior meta-analyses (JCM ‘95, JER) versus the update for the
thesis, there has been a decline in the OR‘s for death for Group 1, Group 2 and Group 3
each versus Group 4 (Table 5.2.3). This decline is partly attributable to the most
recently included study which also is the largest study (Opal [65]).
L’abbe plots
The L‘abbe plots (Fig 5.2.7. a, b, c) presented here are identical to the L‘abbe
plots in CC ‘12 (see figure 5 of CC ‗12).
-140-
Chapter 5: Results & discussion
The group 4 mortality proportion exceeded 15% for all studies in an ICU setting
whereas all but one of the studies in a non- ICU setting had a Group 4 mortality
proportion <15% (Fig 5.2.7.a, b, c). This stratification in mortality porpotions for ICU
versus non-ICU studies is as would be expected. Both the overall dispersion in the
mortality proportions and the dispersion away from the line of identity is most apparent
in the plot of group 1 versus group 4 (Fig 5.2.7 a) in that for 17 of the 34 studies the
mortality proportion in group 1 was >20 percentage points higher in group 1 versus
group 4. By contrast, there were only 8 studies for which the mortality proportion in
either group 2 (Fig 5.2.7 b) or group 3 (Fig 5.2.7 c) differed by more than 20
percentage points versus the mortality in group 4. The greater overall dispersion is also
apparent by contrasting the RD as displayed in the forest plots (i.e. Fig 5.2.4 versus Fig
5.2.5 & fig 5.2.6).
5.2.3. Impact of GN bacteremia species type
The association of patient outcome with endotoxemia detection and with
specific GN bacteremia types in the categories detailed below was examined for those
studies for which this data was available. All of these observations have been obtained
from groups 1 and 2 of the ‗four group‘ studies. Twenty-seven studies published over
three decades were identified. These report the results for a total of 545 GN bacteremic
patients of whom 163 (29.9%) had a fatal outcome. Among the mono-microbial GN
bacteremias, there were 476 bacteremias with outcome known in the following broad
categories of GN bacteremia types;
E. coli; 144 (25%)
Enterobacteriaceae other than E. coli (e.g. Klebsiella species,
Enterobacter species); 97 (17%)
P. aeruginosa; 59 (10%),
N. meningitidis; 138 (24%)
Burkholderia pseudomallei ; 15 (3%)
Yersinia pestis ; 7(1%)
Salmonella typhi; 36 (6%)
Other (including anaerobic); 70 (12%)
Chapter 5: Results & discussion
-141-
The GN bacteremias were classified as follows; bacteremias arising from
studies which included GN bacteremias of any type (unrestricted studies; 17 studies;
Table 5.2.4) and studies which were restricted to GN bacteremias of a specified type
(e.g. N. meningitidis; restricted studies ; 10 studies; Table 5.2.5). Note that there were
an additional five studies [8, 54, 75, 77, 92] with 114 GN bacteremias from studies for
which the GN bacteremia types are known but the individual patient outcomes are not
known. These five studies are not analysed here but are considered further in section
5.2.4.
The summary analysis is presented with four separate methods of summation;
tally, GEE and ELR (Table 5.2.5) and Mantel-Haenzsel method (Figure 5.2.8).
The summary odds ratios for the categories of GN bacteremias are summarized
in Table 5.2.5 and presented as a forest plot in figure 5.2.8. Sparse data from 9 nonICU studies was collapsed into two broad categories to enable their representation
within figure 5.2.8. These two categories are adult non-ICU studies and a combined
category of oncology and pediatric studies. The largest study [65] provided data in
relation to endotoxemia detection at two breakpoints and this has been used to stratify
the mortality proportions from this study within figure 5.2.8. There was no significant
heterogeneity associated with any of the summary odds ratios (figure 5.2.8). The I2
associated with each of these summary OR‘s was <10%, indicating minimal
heterogeneity.
Unrestricted studies: Among the 17 studies that were unrestricted to type of
GN bacteremia, there were 360 patients with 116 (32%) fatal outcomes (Table 5.2.4).
There were eight studies limited to adult patients in ICU [21, 23, 37, 38, 46, 59, 65, 74],
and eight non-ICU based studies including three studies of febrile oncology patients [26,
42, 95] and two studies limited to a pediatric age group [2, 72]. E. coli was the most
common GN bacteremia overall. The mortality proportion was lowest for E. coli (144
with 34 deaths; 24%), versus Enterobacteriaceae other than E. coli (97 with 33 deaths;
34%). Likewise, among GN bacteremias summed across only the eight ICU studies, the
mortality rates for bacteremias with E. coli, Enterobacteriaceae other than E. coli, and
Pseudomonas species were 23%, 35%, and 47%, respectively.
-142-
Chapter 5: Results & discussion
Table 5.2.4: Mortality proportions for patients with gram negative
bacteremia: unrestricted studies only
Numbers of patients with fatal outcome / Number tested
Assay a Census b
Study [reference]
Bates 1998 [46] d
Danner 1991 [21] d
Dofferhoff [23]
1992
Goldie [37] 1995 d, e
Guidet [38] 1994 d, f
Maury [59] 1998 d, g
Strutz [74] 1999
Opal [65] 1999
(Low; 20-660
pg/ml) d, h
Opal [65] 1999
(High; >660 pg/ml)
E coli
non- E coli
Enterobacteriaceae
nonEnterobacteriaceae c
ng/ml
0.005
0.01
day
28
14
Etx+
0/3
1/5
Etx6/20
0/4
Etx+
4/6
1/2
Etx4/14
0/3
Etx+
1/2
1/2
0.005
21
0/1
0/1
1/1
0/1
1/2
0.002
0.005
0.005
0.01
30
28
30
NS
0/1
3/9
0/8
1/4
1/1
3/7
2/3
4/6
2/4
0/1
1/1
1/2
0/1
2/3
0.02
28
2/15
0.66
28
5/18
3/4
0/1
1/1
0/1
2/12
2/6
Etx2/5
4/8
4/15
1/6
7/16
3/7
20/50
6/25
17/36
6/15
d, h
Sub-total
Non-ICU (Adult)
4 studies i
Non-ICU (oncology
& pediatric)
5 studies j
Total
12/64 15/43
3/13
2/19
1/2
0/4
0/3
0/4
2/4
0/1
5/10
1/6
5/14
0/6
26/62
7/35
22/53
8/25
17/81 17/63
a.
Refers to endotoxin assay breakpoint for endotoxemia detection
b.
Refers to survival census day
c.
Non- Enterobacteriaceae includes the following bacteremias; Pseudomonas species, Acinetobacter species,
and Bacteroides species.
d.
Data for these studies [21, 37, 38, 46, 65] provided by personal communication.
e.
Goldie [37] – H. influenzae bacteremia (n=1) from this study not included
f.
Guidet [38] – H. influenzae bacteremia (n=1) from this study not included
g.
Maury [59] – H. influenzae bacteremia (n=1) from this study not included
h.
Data for this study Opal et al [65] stratified into two categories of endotoxemia detected; low (20-660
pg/ml) and high (>660 pg/ml) versus endotoxemia not detected (<20 pg/ml)
i.
Non–ICU (adult) represents Martinez-G et al 1973 [57], Byl et al, 2002 [17], Prins et al., 1995[67] and
Giamarellos et al 1999 [36] collapsed into one category
j.
Non-ICU (oncology & pediatric) represents Hynninen et al 1995 [42], Engervall et al 1997 [26], Yoshida
et al 1994 [95], Shenep et al 1988 [72] and Ahmed et al 1991 [2] collapsed into one category.
Chapter 5: Results & discussion
-143-
For these GN bacteremias from the unrestricted studies, only in the case of
Enterobacteriaceae other than E. coli, was the summary odds ratio significantly
elevated above 1, indicative of a worse outcome in endotoxemic patients (Table 5.2.5,
Figure 5.2.8). The increased mortality for this category is particularly apparent for
Klebsiella species (9 of 13 deaths in 8 studies if endotoxemia was detected versus 2 of
17 deaths in 6 studies if endotoxemia was not detected). For the E. coli GN bacteremias,
none of the study level odds ratios (Figure 5.2.8), or the summary odds ratio, were
significantly elevated above 1 (Table 5.2.5). Recalculation of the OR‘s including only
the eight studies limited to the ICU setting [21, 23, 37, 38, 46, 59, 65, 74] yielded
summary OR‘s similar to previously with none that were significantly different from 1.
The method used to derive the summary OR‘s in Table 5.2.5 (GEE versus ELR) or
Figure 5.2.8 (MH) made no substantive difference to the overall findings.
Enterobacteriaceae bacteremias and endotoxemia. The finding (Table 5.2.5,
Figure 5.2.8) that the risk of mortality was increased in association with the detection
versus non-detection of endotoxemia with GN bacteremias with Enterobacteriaceae
other than E. coli but not for E. coli was unexpected. The statistical significance of this
unexpected finding was tested for as an interaction term within a regression model.
This interaction term was found to be significant indicating an impact of endotoxemia
on prognosis additional to the impact of GN bacteremia type for E. coli versus non-E.
coli Enterobacteriaceae. This was apparent for both an analysis limited to the studies
undertaken in ICU (OR 3.8; 1.13 – 12.8) [21, 23, 37, 38, 46, 59, 65, 74] and also for an
analysis including all the unrestricted studies (OR 2.8; 1.01 - 8.1) [2, 17, 26, 36, 42, 57,
67, 95].
Restricted studies: Among the studies restricted to one of the four specified
GN bacteremias; Neisseria meningitidis (five studies) [11, 12, 35, 37, 38, 68, 72];
Yersinia pestis (two studies) [14, 15], melioidosis (one study) [73]; and Salmonella
typhi (three studies) [1, 56, 77], there were 196 bacteremic patients with 49 (25%) fatal
outcomes. For these GN bacteremias, only in the case of Neisseria meningitidis
bacteremia was the summary odds ratio for mortality significantly elevated (p < 0.0001)
in association with the detection of endotoxemia (Table 5.2.5).
-144-
Chapter 5: Results & discussion
Table 5.2.5: Mortality proportions and odds ratios in association with
endotoxemia detection: unrestricted and restricted studies
Dead / total number of
GN bacteremia species
Odds ratio; 95% CI
patients (%)
Number
of
studies
ICU studies
E coli a
Endotoxemia Endotoxemia
detected
not detected
GEE
ELR
15/43 (35) 0.54; 0.25 - 1.13 0.51; 0.17 - 1.5
8
12/64 (19)
non- E coli
Enterobacteriaceae a,b
8
20/50 (40)
6/25 (24)
2.7; 0.90 - 8.4 3.1; 0.82 - 13.7
Non-Enterobacteriaceae c
7
17/36 (47)
6/15 (40)
1.2; 0.35 - 4.1 1.06; 0.27 - 4.3
ICU & non-ICU studies
E coli a
15
17/81 (21)
17/63 (27)
0.78; 0.36 - 1.7 0.81; 0.28 - 2.3
non- E coli
Enterobacteriaceae a,b
14
26/62 (42)
7/35 (20)
3.7; 1.3 - 10.3
3.9; 1.1 - 16.7
Non-Enterobacteriaceae c
13
22/53 (42)
8/25 (32)
1.7; 0.65 - 4.9
1.4; 0.42 - 5.3
Neisseria meningitidis
5
37/106 (35)
0/32 (0)
26.0; 1.6 - 321
36.9; 7.7 - Inf
Yersinia pestis
2
1/5 (20)
0/2 (0)
Undefined
Undefined
Salmonella typhi
3
1/19 (5)
1/17 (6)
Undefined
0.89; 0.01-74.1
Burkolderia pseudomallei
1
9/15 (60)
1/5 (20)
Undefined
Undefined
Abbreviations: GEE = Generalized estimating equation method; ELR = Exact logistic regression
method
a. The impact of the detection of endotoxemia on the difference in morality proportions for E
coli versus non-E. coli Enterobacteriaceae bacteremias was tested as an interaction term
within a regression model using generalized estimating equations. This interaction term was
found to be significant (OR 3.8; 1.13 – 12.8) for an analysis limited to the studies undertaken
in ICU and also for an analysis including both ICU and non-ICU studies (OR 2.8; 1.01 - 8.1).
b. Non-E. coli Enterobacteriaceae includes the following bacteremias; Klebsiella species,
Enterobacter species, Proteus species, and Serratia species.
c. Non- Enterobacteriaceae includes the following bacteremias; Pseudomonas species,
Acinetobacter species, and Bacteroides species.
Chapter 5: Results & discussion
author
E coli
Goldie
Bates
Strutz
Opal lo
Guidet
Opal hi
Dofferhoff
Adult Non-ICU
Oncology & pediatric
Danner
Subtotal (I-squared = 0.0%, p = 0.733)
.
non-E coli Enterobacteriaceae
Goldie
Opal lo
Opal hi
Bates
Oncology & pediatric
Danner
Dofferhoff
Adult Non-ICU
Subtotal (I-squared = 0.0%, p = 0.810)
.
non-Enterobacteriaceae
Guidet
Adult Non-ICU
Opal hi
Opal lo
Bates
Goldie
Oncology & pediatric
Subtotal (I-squared = 0.0%, p = 0.779)
.
.003 .01 .03 .1 .3 1 3 10 30 100300
etx -; less survival etx +; less survival
Fig 5.2.8 Forest plot (OR). . Forest plot of mortality in relation to the detection of
endotoxemia (etx + versus etx -) for patients with GN bacteremia are presented as
study specific odds and summary odds ratios; ‗etx – less survival‘ and ‗etx + less
survival‘ indicates the ranges in OR‘s for which mortality is more common in
association with either endotoxemia non-detection or detection, respectively.
-145-
-146-
Chapter 5: Results & discussion
Comparison versus each benchmark IQ range. The mortality risk for GN
bacteremias of the three types usually seen in unrestricted studies as seen in the broader
literature since 1975 [287, 338-390] was summarized as an inter-quartile (IQ) range
(Table 2.4.2). The overall IQ range was 19-43% (figures 2.12/.13/.14/.15). This survey
included opportunistic type GN bacteremias as reported for various hospital based
settings (e.g. ICU versus non-ICU) and studies of various sizes.
The mortality proportions for patients with versus without endotoxemia in
association with GN bacteremia in studies unrestricted to specific GN bacteremia types
are mostly within each respective range and the overall IQ range (Figure 5.2.9). By
contrast, the mortality proportions for patients with GN bacteremias in studies restricted
to specific GN bacteremia types such as with Burkholderia (melioidosis – endotoxemia
detected) and N. meningitidis (endotoxemia not detected) are partly outside this range
and for S. typhi (Typhoid – both endotoxemia detected and endotoxemia not detected)
are wholly outside this range.
Escherichia coli
+ -
non-E coli
Enterobactereiaceae
-
non-Enterobactereiaceae
Neisseria meningitidis
+
- +
+
-
Salmonella typhi
- +
Burkolderia pseudomallei
-
1
3
5
10
20 30
Mortality (%)
+
50
70
Fig 5.2.9 Comparison versus IQ range. Plot of mortality proportions in relation to
the detection of endotoxemia (etx + versus etx -) (Table 5.2.5) for various types of
GN bacteremia versus each respective IQ range (represented by horizontal lines)
and also versus GN bacteremia overall (represented by vertical lines) as derived
from studies in the literature (Table 2.4.2). Note logit scale.
Chapter 5: Results & discussion
-147-
5.2.4. Sudy limitations and literature reconciliation
These are the findings from this analysis of the detection of endotoxemia as a
test of prognosis;
1. (section 5.2.1). The mortality RD associated with the detection versus nondetection of endotoxemia (without regard to the co-detection of GN bacteremia)
is similar amongst the ‗two group‘ versus the ‗four group‘ studies for studies
conducted in an ICU setting. Among 20 such studies only two statistical outlier
studies were identified; one high (Valenza; ii6 [107]) and one low (Goldie; i5
[37]) (Table 5.2.1, Fig 5.2.1). The following is used as a criterion to define a
statistical outlier study; a study‘s 95% confidence interval fails to include the
common summary result in either the OR (Figure 5.2.1) or RD (Figure 5.2.2)
forest plots.
2. (section 5.2.1). By contrast, for studies conducted outside an ICU setting, the
mortality RD associated with the detection versus non-detection of endotoxemia
(without regard to the co-detection of GN bacteremia) is more heterogeneous
versus that for studies conducted in an ICU setting with more variability both
within and between study categories. Among 30 such studies eight statistical
outlier studies were identified; three low (Bion, a7, Cooperstock) and five high
(Brandtzaeg [mg], Das [22], Levin [a4], Yoshida [o3], van Langevelde [a9])
(Fig 5.2.1 & Fig 5.2.2).
3. For the largest study (i9, [65]), the 135 GN bacteremic patients were able to be
classified into sub-groups with high (>660 pg/ml), low (>25 pg/ml) and nondetectable (<25 pg/ml) levels of endotoxemia. There was no significant
evidence for a dose response effect in any of the three categories of GN
bacteremia species type studied in that the mortality odds ratios were similar at
high and low levels of endotoxemia (fig 5.2.3).
4. (section 5.2.2). Summary results (Table 5.2.3). The co-detection of
endotoxemia with GN bacteremia (Group 1) is most predictive of increased
mortality risk (OR 6.9; 4.4-11.0; RD 19%; 11-27%) versus the detection of
neither (Group 4; Figure 5.2.4/.5/.6 and Figure 2/3/4 & Table 2 of CC ‘12).
However, this is most apparent amongst the sub-group of studies undertaken
outside an ICU setting and is a finding which is associated with marked
-148-
Chapter 5: Results & discussion
heterogeneity (OR I2 35%; RD I2 67%). By contrast, the detection of either
endotoxemia or GN bacteremia in isolation (Group 2 and Group 3; Figure
5.2.4/.5/.6 and Fig 2/3/4 & Table 2 of CC ‘12) versus the detection of neither
(group 4) is associated with borderline elevation of risk (OR <2.0; RD <10%)
and, moreover, is associated in each case with heterogeneity that is no more
than moderate (OR & RD I2<35%). This lack of heterogeneity is surprising
given the diversity of patient groups, underlying risk, and clinical settings in
these studies that were conducted and published over a period exceeding three
decades.
5. L‘Abbé plots (Fig 5.2.7). The underlying patient risk, as reflected in the group 4
mortality proportion, was higher for studies in an ICU setting versus studies in
other settings (Fig 5.2.7a-c), as might be expected. However, the additional
mortality associated the detection of either or both of endotoxemia or GN
bacteremia was less apparent in the studies of patients at high versus low
underlying risk (Table 5.2.3).
6. (section 5.2.3). The prognostic value of endotoxemia detection versus nondetection in association with GN bacteremia varies across different species
types of GN bacteremia. This variability across species types, which is in
contrast to the lack of heterogeneity across studies, is visually apparent in the
forest plot (Fig 5.2.8).
7. The detection of endotoxemia in association with Neisseria meningitidis
bacteremias has the strongest prognostic value (Table 5.2.5). With the GEE
method, the width of the 95% confidence interval (CI) is determined by the
number of clusters together with the similarity of their results (intra-cluster
correlation). Hence with this method, clusters that are few (e.g. N. meningitidis),
or small or which have disparate results will result in 95% CI‘s that are wide
versus clusters that are numerous, large and which have similar results.
8. Surprisingly, the detection of endotoxemia in association with E. coli
bacteremia, which amongst the unrestricted GN bacteremia category has the
most number of patients, has no prognostic value (Fig 5.2.8, Table 5.2.5) in any
setting. This finding is in stark contrast to the association of endotoxemia, GN
bacteremia and outcome overall (i.e. group 1 versus group 4; Table 5.2.1). As a
Chapter 5: Results & discussion
-149-
sensitivity analysis, it was found that it would require at least three studies each
having at least 40 patients with E. coli bacteremias with an OR for mortality of
> 10 in association with the detection versus non-detection of endotoxemia to
change the conclusions found here (data not shown). It is unlikely that such
studies would be unpublished or easily overlooked.
9. This unexpected finding for E. coli bacteremias versus GN bacteremia overall is
to be explored as a basis for the conflicting findings of the individual studies
(Tables 5.2.6 & figure 5.2.9).
10. Note that the mortality proportions for patients bacteremic with E. coli, and
Pseudomonas species found in the studies analysed in Table 5.2.5 are mostly
within each respective IQ range as derived from the broader literature (Table
2.4.2) whether or not endotoxemia is detected.
11. There is a significant interaction effect between the detection of endotoxemia
and the type of Enterobacteriaceae (E. coli versus non-E. coli) bacteremia on the
prognosis (Table 5.2.5).
12. The prevalence and species types of GN bacteremias varied among individual
studies and groups (Table 5.1.6). However the numbers in individual studies are
small typically being less than 20 GN bacteremias in total.
13. Of the patients included within this analysis of prognosis (Figure 5.2.4, Figure
5.2.5, Figure 5.2.6), 20% had GN bacteremia detected (Group 1 and Group 2)
of which 23% were E. coli whereas only 12% were P. aeruginosa. However,
there are significant reciprocal differences in the frequency of P. aeruginosa
versus E. coli among the bacteremias of Groups 1 and 2.
14. Outlier studies. The following is used as criteria to define a statistical outlier
study; a study‘s 95% confidence interval fails to include the common summary
result in either the OR (Figure 5.2.1) or RD (Figure 5.2.2) forest plots. Using
this as a criterion to define a statistical outlier reveals 10 studies with outlier
results in figures 5.2.1, & 5.2.2. Using this criterion there are five studies which
are high outlier and have either an RD of >30% (a4, mg, o3, Das) or an OR > 3
(a4, mg, o3, a9, Das). Likewise, there are five studies which are low outliers
with an RD of <0% [Bion, a7, p3, Magliulo] or an OR <1 (Bion, a7, i5). There
are additional studies with a RD >20% or <0% (see also Figures 5.2.3/.9) but
-150-
Chapter 5: Results & discussion
which do not meet the statistical outlier criterion listed above. These are termed
potentially outlier studies. Possible reasons that might account for why the
results of these studies are outlier are discussed below.
15. The relevance of endotoxemia detection toward prognosis amongst patients with
GN bacteremia within the studies analyzed here [1-107] is contrasted against
each IQ range that was derived from other studies in the broader literature
experience since 1975 [287, 338-360] to enable three inferences to be drawn.
16. Firstly, amongst the unrestricted studies the mortality proportions for patient
with versus without endotoxemia detected are mostly within the IQ range for
each respective GN type (Figure 5.2.9) with the exception of non-E. coli
Enterobactereiaceae. Hence for the GN bacteremia types as classified here, the
impact of endotoxemia detection toward patient prognosis for any of the three
GN bacteremia types (Table 5.2.5) cannot be clearly discerned versus the
impact of the many other factors contributing toward the range in mortality
proportions among these GN bacteremia types in the broader literature
experience.
17. The impact of endotoxemia detection versus non- detection on mortality
proportions is most evident for non-E. coli Enterobactereiaceae bacteremias.
18. Thirdly, the experience amongst the restricted studies is not representative of
that amongst the unrestricted studies. In this regard, the mortality proportions
for patient with versus without endotoxemia detected are either partly above
(Neisseria meningitidis), partly below (Burkholderia pseudomallei) or wholly
outside (S. typhi) this benchmark IQ range (Table 5.2.5).
Overall, the analysis presented in section 5.2 implies that the study setting being
ICU versus non-ICU, and the study prevalence of GN bacteremia type may be
unrecognized confounders of the relationship between endotoxemia and outcome. This
warrants further consideration toward the interpretation of the studies included here.
However, there are a number of limitations to these findings.
This analysis is unable to identify the mechanism for any increased mortality
risk. Many relevant patient-specific details such as age and co-morbidities for the
patients of the four groups in each study were not available. In this analysis, a variable
proportion of blood cultures that were classified as negative for GN bacteria would
Chapter 5: Results & discussion
-151-
have yielded gram-positive bacteremias or fungemias. The detection of endotoxemia in
association with blood culture isolates other than GN bacteremias is a commonly
reported finding [25, 54, 75]. In this respect, the prognostic impact of blood stream
infections other than GN bacteremias in relation to the co-detection with endotoxemia
has not been addressed here. Moreover, the origin of endotoxemia for patients in Group
3 is uncertain and the possibility of endotoxin originating from other sources, such as
gut barrier breakdown, as is presumed to occur for non-septic forms of shock, needs to
be considered [708].
It needs to be noted that endotoxemia [709] and GN bacteremia are each either
episodic or dynamic phenomena and the criteria for a positive detection of each will
have differed among the studies. The kinetics of bacteremia, endotoxemia and
pathophysiology may not be concordant. For example, in one clinical study [21] of 100
patients with sepsis in an ICU setting, the cumulative percent found to have
endotoxemia rose from 20% to 40 % between 0 and 24 hours after study entry.
Moreover, in many studies, the timings of antibiotic administration in relation to
collections of samples for blood cultures and endotoxemia assay were not known.
The additional mortality associated the co-detection of endotoxemia and GN
bacteremia (Group 1) was less apparent in the studies of patients at high (i.e. ICU
setting) versus low underlying risk (in Table 5.2.3 and Table 2 of CC ‘12). This
finding here at a group level of analysis resembles other recent findings at an individual
patient level of analysis among bacteremias of all species types occurring in an ICU
setting [301]. However, this inference requires caution for three reasons. Firstly, it
needs to be clarified as to whether the increased risk is absolute [340] or relative [301].
Secondly, it should be noted that the L‘Abbé plots are useful merely as simple
graphical methods to facilitate visual comparisons of the group mortality proportion
over the range of underlying risk as found in individual studies within a meta-analysis
[681]. The issues underlying the statistical testing for variation in either additional risk
or treatment effect in relation to underlying risk are not simple [681-683, 686-687]. In
particular, in using linear regression (which has not been done here) in conjunction with
L‘Abbe plots to explore heterogeneity over a range of underlying risk, regression to the
mean is an important consideration as has been noted among treatment studies of mild
hypertension [687].
-152-
Chapter 5: Results & discussion
Thirdly, there were insufficient studies that had used assays other than the
limulus assay, or patient groups that had received anti-endotoxin antibodies to include
these studies so as to enable a systematic study of these variables by meta-regression.
Only eight studies [1, 11-16, 56, 77] were limited to specific GN bacteremia species
types. Also, many relevant study level details, such as method of blood culture used and
antibiotic therapy protocols were not available. Anti-endotoxin antibodies are
detectable in patients with severe sepsis and septic shock and the kinetics of these
antibodies over time differs between survivors and non-survivors [710]. However, the
impact of these anti-endotoxin antibodies on both the detection of endotoxemia and
possibly also on patient outcome within the studies examined here is uncertain [711712]. A more detailed examination adjusting for relevant prognostic variables
contributing to underlying patient level risk and also the inter-relation between various
GN bacteria known to have differing lipid-A structures would require an individual
patient data meta-analysis.
Finally, it needs to be conceded that some of the findings in section 5.2 are
unstable to additional ‗missing‘ data whether published or otherwise. It is not
unprecedented that the finding of a large meta-analysis is overturned by unpublished
data and the controversial example of the meta-analyses of mortality related to
cefepime [713-718] was cited earlier (section 2.6). In this regard, estimates for the
amount of missing data that would be required to overturn these findings are provided.
Literature reconciliation With these caveats mentioned in relation to the
limitations of this analysis, the studies will now be examined with particular interest in
the outlier studies identified in Figures 5.2.1/.2/.3 in an attempt to achieve a
reconciliation of the literature. Figure 5.2.10 is Figure 5.2.3 reproduced with the
following modifications;
Studies that are small (< 20 patients or < 2 GN bacteremias) or with QS<2 have
been excluded due to concerns relating to publication bias and atypical DOR results
(as identified in 5.1.1),
Studies restricted to specified bacteremias have all been excluded except for one
study of typhoid ([77] see below).
The studies have been coded to indicate whether the number of E. coli bacteremias
as a proportion of the total number of patients is <3%, 3-5%, >5% or unknown (see
figure legend),
Studies for which the number of E. coli bacteremias is <5% have the studies‘
diagnostic odds ratio (DOR) as tabulated in Table 5.2.6 displayed,
Chapter 5: Results & discussion
-153-
Reference lines indicating risk differences of 0% and 20% are displayed in Table
5.2.9 to indicate potentially outlier studies,
Six supplementary list studies including four non-Limulus studies are included,
With these modifications, Figure 5.2.10 summarizes the influence of
underlying patient risk at a group level (mortality in Group 2 & 4 combined), the
prevalence of E. coli bacteremias as a proportion of the total number of patients, the
diagnostic odds ratio and the risk difference in each study. In addition the reference
lines in the figure help to identify potential outlier studies (i.e. by using the screening
definition of a potential outlier being RD<0% or RD >20%).
‗▲‘ non-limulus studies. Five studies that used the EAA (non-limulus) assay
method in an ICU setting (indicated by ‗▲‘ in Figure 5.2.10) are displayed and only
one is a statistical outlier (ii6; RD = 12%). This study is a high outlier (ii6 [107]) and
while it is classified having been undertaken in an ICU setting, all patients were postsurgical and the mortality risk in the endotoxemia negative group is 0% and the GN
bacteremia information is not available for this study.
‘two group’ studies. None of the six large ‗two group‘ studies in an ICU setting
(ii3 [91], ii5 [101], ii1 [102], ii2 [104] ii4 [106] ii6 [107]) found mortality differences
exceeding 15 percentage points between patient groups that were endotoxemia positive
versus negative. While one study had reported E. coli prevalence (accounting for two of
a total of only four GN bacteremias [106]), the GN bacteremia species types were not
stated in the other studies. A seventh large non-ICU study did not report mortality as an
outcome and could not be included here [86].
‘○’ / ‘+’ studies. There are 22 studies for which the number of E. coli
bacteremias as a proportion of the total number of patients is >5% (indicated by a ‗○‘)
or unknown (indicated by a ‗+‘) and only one of these is an outlier [Das, 22]. This
study, which is a high outlier, included both pediatric group and adult patients.
However, further information relating to patient selection and GN bacteremias is not
available for this study.
‗●‘ / ‗●‘ studies with RD<0. There are 12 studies for which the number of E.
coli bacteremias as a proportion of the total number of patients is <3% or 3-5%
(indicated by ‗●‘ or ‗●‘ closed symbols, respectively). There were three which have a
RD<0% of which one of these three is a small (n=10) Japanese study of neonates
receiving exchange transfusion as a treatment for sepsis [81, not shown in Figure 5.2.9].
-154-
Chapter 5: Results & discussion
The other two are studies which both have a low DOR (0.6, [Bion; 8] & 1.3 [Goldie;
37]). The low DOR in these two studies raises the possibility that assay malfunction
occurred and hence resulted in misclassification bias with the likelihood that the
patients with endotoxemia were incorrectly classified in these studies.
mortality (%)
endotoxemia detected versus not detected
80
70
60
10.4
50
4.3
4.6
40
4.8
1.3
30
24.4 20.5
20
1.8 4.2
9.9
10
.6
0
0
10
20
30
40
50
60
Mortality group 2 & 4 combined
Fig 5.2.10 Mortality L‘abbe plot for endotoxemia detected (Groups 1 & 3
combined) versus endotoxemia not detected (Groups 2 & 4 combined). Restricted
and small studies (number of patients; N< 20 or number of GN bacteremias<2) not
shown. Symbols are coded for calculated E coli bacteremias as % of N; black ‗●‘ if
<3%, gray ‗●‘ if 3-5%, open ‗○‘ if >5%, ‗+‘ if unknown and ‗▲‘ if non-limulus
assay used. The numbers refer to the DOR (studies with <5% E coli bacteremias;
see table 5.1.6). Two lines are shown for visual reference; the line of equivalence
(solid line) and that for a risk difference equivalent to a 20% excess (dashed line).
Note: Y axis should read ‘Mortality in groups 1 & 3 combined’
Chapter 5: Results & discussion
-155-
Table 5.2.6 Details of selected (QS>2 & N>20) studies (& codes for figure 5.2.3)
code Author
year
ref
DOR
N
Ec
%
RD
GNB
n
E coli bacteremias <3% (●)
0.6
4.6
9.9
1.3
10.4
4.8
20.5
136
41
133
218
27
21
0
1
1
2
7
0
0
0
1
2
2
3
0
0
-17
33
9
1987
1995
8
95
2
37
54
92
77
22
18
2
30
8
12
30
5
4
E coli bacteremias 3-5% (●)
o2
Engervall
1997
p4
Shenep
1988
a9
VanLangevelde
2000
a5
Martinez-G
1973
27
72
89
57
1.8
24.4
4.2
4.3
24
26
448
83
1
1
17
4
4
4
4
5
17
25
12
41
6
9
40
14
65
65
47
20
21
23
75
38
52
74
17
36
1.4
1.4
0.5
29
2.1
1.8
1.4
5.1
1.1
1.2
5.5
5.7
759
759
356
43
100
18
117
93
40
28
27
25
39
39
23
3
9
2
18
16
8
8
8
10
5
5
6
7
9
11
15
17
20
29
30
40
6
6
135
135
58
5
19
6
63
31
11
10
12
12
102
91
22
94
94
33
NS
NS
18.4
11.3
2.8
1.6
97
106
NS
NS
NS
NS
NS
NS
o3
p1
i5
a4
a11
Bion
Yoshida
Ahmed
Goldie
Levin
Watzke
Suyasa
1994
1994
2004
1995
1972
E coli bacteremias >5% (○)
i9
Opal (lo & hi)
2000
i9
Opal (lo & hi)
2000
i7
Ketchum
1997
p3
Cooperstock
1985
i1
Danner
1991
i2
Dofferhoff
1992
a7
Stumacher
1973
i6
Guidet
1994
a3
Lau
1996
i0
Strutz
1999
a0
a2
Byl
G-bourboulis
2002
1999
52
-4
41
10
0
7
8
-9
16
4
0
12
14
E coli bacteremias unknown (+)
ii1
ii3
Casey
Venet
Das
Wortel placebo
Wortel HA-1A
Fugger
1993
2000
1973
1992
1992
1990
non-limulus assay supplementary list
studies (▲)
ii2
1997 104 NS
Kollef *
ii4
Marshall *
2002 106 NS
ii5
Yaguchi*
2012 101 NS
ii6
Valenza*
2009 107 NS
Foot notes;
Only shown are studies with QS>2 and N>20.
54
41
41
24
2
3
63
16
-5
6
140
857
13
5
210
102
6
12
NS
NS
NS
NS
NS
NS
NS
4
20
NS
-156-
Chapter 5: Results & discussion
‗●‘ / ‗●‘ studies with RD>0. Of the remaining 9 of the 12 studies for which the
number of E. coli bacteremias as a proportion of the total number of patients is <3% or
3-5% (indicated by ‗●‘ or ‗●‘ closed symbols, respectively) all nine have a RD>9% and
these all have DOR‘s above 4. Three of these studies (ml – [73, not shown in Figure
5.2.9], Levin [54], Yoshida [95]) are also flagged as high outliers by statistical criteria.
Two studies (Magluilo [56], Copperstock; p3-[20]) are flagged as statistical low
outliers and each has a RD= 0 but also have mortalities of zero in both those with and
without endotoxemia detected. Both had a high prevalence of either Salmonella species
(100% Magliulo [56], not shown in Figure 5.2.9) or E. coli (9%; Copperstock; p3 [20])
isolates.
Restricted studies. Six other studies restricted to specified bacteremias are not
displayed in Figure 5.2.9 but four warrant specific mention. The studies of
meningococcemia (mg [11]) and melioidosis (ml [73]) both have RD>20%. One study
of typhoid (t [1]) has a RD of 0%. Another study of typhoid is displayed in Figure 5.2.9
(Suyasa, [77]) is classified under the category of restricted studies although 4 of the 21
typhoid patients of this study were found to also have a Pseudomonas bacteremia. This
study [77] has a RD of 18%, a DOR of 20.5.
Finally, the three original studies published in 1972-1975 which produced
conflicting conclusions and which inspired this thesis can be examined [25, 54, 75].
One does not provide either mortality or GN bacteremia data [Elin et al; 25] and hence
cannot be further assessed here. The second [a7; Stumacher; 75] has a high prevalence
of E. coli (15%) isolates together with a relatively low DOR (1.4) as possible
explanatory factors for being a statistical low outlier study. The third [a4; Levin; 54]
has a low prevalence of E. coli (3%) isolates together with a high DOR (10.4) as
possible explanatory factors for being a statistical high outlier study.
-157-
6. Summary and conclusions
6.1 Summary of findings
The research described here addresses three questions related to the detection of
endotoxemia in various clinical settings using the limulus (LAL) assay with respect to;
1. what is its diagnostic relevance versus GN bacteremia detection,
2. what is its prognostic relevance versus outcome (mortality), and
3. whether the disparate study results are reconcilable?
In the four decades following the inception of the limulus assay for
endotoxemia in 1972, the following developments bearing on the above three questions
have occurred;
There are over 300 studies using the limulus assay in various clinical
contexts including patients at risk for sepsis [1-107] and various other settings
[498- 673].
the chromogenic limulus assay superseded the less sensitive gelation based limulus assay after 1984,
the patient population of particular interest within this thesis is the
population within the ICU setting with sepsis which has been defined most
commonly using the 1991 consensus sepsis criteria (Sepsis Syndrome). However,
these definitions will continue to evolve and non-ICU populations are also of
interest.
the structure activity of endotoxin has become better understood with
recognition of the importance of a hexaacyl (6 chain) as opposed to non-hexaacyl
(4, 5 or 7 chain) structure of the lipid-A component of the lipopolysaccharide
molecule (endotoxin) in host recognition by the human immune system,
the statistical techniques of meta-analysis used in the derivation of
summary estimates have evolved. In addition, the derivation of estimates of
heterogeneity as a measure of the diversity of experience within the literature is
now possible. In the most recent phase of this study, various multi-level methods
became available to derive estimates of summary proportions that are optimal for
data for which the assumption of independence is not tenable, which is most
pertinent to the analysis of data in an infectious diseases setting.
1. What is the diagnostic relevance of the detection of endotoxemia
versus GN bacteremia detection?
1.1. Endotoxemia is not detectable in at least 25% and up to 60% of
patients with GN bacteremia.
1.2. The ‗typical‘ proportion of patients with GN bacteremia in
which endotoxemia is not detectable is dependent on the method of data
aggregation across the studies. The HSROC method is considered optimal to
derive estimates that are most representative of a ‗typical‘ study.
-158-
Chapter 6: Summary & conclusion
1.3. The choice of method of plasma pre-treatment prior to the
limulus assay may be important and the optimal method identified here is the
method of dilution and heating.
1.4. There is no direct comparison of the gelation versus the more
sensitive chromogenic limulus assay methods in the same patient population in
the literature. In the analysis here using GN bacteremia overall as the ‗gold
standard‘, the newer chromogenic assay is not a superior assay.
1.5. Surprisingly, there appears to be no relationship between the
diagnostic odds ratio and the listed sensitivity of the limulus assay to the
internal endotoxin standard as used within each study over a range of more
than 1000 fold (<.01 ng/ml to >10 ng/ml).
1.6. Using HSROC methods of analysis, the proportion of GN
bacteremias with detectable endotoxemia is 68% (95% CI: 51-81%) for
bacteremias with Pseudomonas, 54% (39-68%) for E. coli and 48% (30-66%)
for the non-E. coli-enterobacteriaceae. (Table 5.1.5).
1.7. With the three categories of specified GN bacteremia types there
appears to be a differential relationship between the detection of GN
bacteremia and the associated endotoxemia as follows; the chromogenic
version is more sensitive than the gelation-version for the detection of
endotoxemia in association with bacteremias with Pseudomonas and the non-E.
coli-enterobacteriaceae. (Table 5.1.5).
1.8. However, the 300 fold increase in sensitivity limit in the limulus
assay method (chromogenic- versus gelation-versions) is associated with an
increase in the sensitivity toward endotoxemia detection in association with
bacteremias with Pseudomonas and the non-E. coli-enterobacteriaceae by no
more than 14 percentage points.
1.9. The two versions (chromogenic- versus gelation-versions) are
equivalent in sensitivity for E. coli bacteremias (Table 5.1.5).
1.10. The specificity of the assay is dependent on the study setting
with less specificity among studies undertaken in an ICU setting (51%; 3865%) versus a pediatric or oncology setting (88%; 59-97%). The sensitivity is
not dependent on study setting. (Table 5.1.5).
1.11. Overall, the findings presented here imply that the type of GN
bacteremia is at least as important a determinant toward the detection of
concurrent endotoxemia as is the sensitivity of the limulus assay method used.
1.12. Surprisingly, the relationship between endotoxemia detection
and the type of GN bacteremia is not simply determined by the type of lipid-A
structure as originally predicted at the outset of this thesis.
1.13. For a ‗typical‘ study having 100 patients and 24 GN bacteremias,
overall the expectation would be that there would be 16 true positives
(concurrent endotoxemia and GN bacteremia detected), 8 false negatives (GN
bacteremia in isolation), 57 true negatives (neither endotoxemia nor GN
bacteremia detected), and 19 false positives (endotoxemia in isolation) (Table
5.1.2).
1.14. The expected number of true positives for a ‗typical‘ study
Chapter 6: Summary & conclusion
-159-
would need to be modified as follows; downward for E. coli and non-E. colienterobacteriaceae GN bacteremias. For 24 E. coli bacteremias there would be
an expectation of 13 true positives and 11 false negatives and with 24 non-E.
coli-enterobacteriaceae bacteremias there would be an expectation of 12 true
positives and 12 false negatives (Table 5.1.5).
1.15. For 24 non-enterobacteriaceae GN bacteremias the expected
number would be modified upwards; there would be 16 true positives and 8
false negatives expected.
2. What is the prognostic relevance of the detection of endotoxemia
versus outcome (mortality)?
2.1. The prognostic value of endotoxemia detection versus nondetection has been most commonly studied in the ICU setting. In the ICU
setting, endotoxemia detection has little (OR = 1.4; 1.2-1.7, RD 7%; 3-10%)
prognostic value in studies reporting either as ‗two group‘ (Figure 5.2.1) or
‗four group‘ studies (Figure 5.2.2).
2.2. Endotoxemia detection has been studied in several non-ICU
settings where it has been found to have prognostic value which is generally
higher but more variable than in ICU settings (Figure 5.2.1 & fig 5.2.2).
2.3. Amongst the ‗four group‘ studies, the co-detection of
endotoxemia with GN bacteremia (Group 1) is most predictive of increased
mortality risk versus the detection of neither (group 4; Figure 5.2.3). While
this is most apparent amongst the sub-group of studies undertaken outside an
ICU setting (OR 6.9; 4.4-11.0; RD 19%; 11-27%), this finding is associated
with at least moderate heterogeneity (OR I2 35%; RD I2 67%).
2.4. By contrast, the mortality risk in association with the detection
of either endotoxemia or GN bacteremia in isolation (Group 2 and Group 3;
Figure 5.2.4 & Figure 5.2.5) versus the detection of neither (Group 4) is
associated with borderline or non-significant elevation of risk (OR <2.0; RD
<10%). Moreover, this finding applies in both ICU and non-ICU studies and
the associated heterogeneity is no more than moderate whichever summary
effect size is used (OR & RD I2<35%).
2.5. The prognostic value of endotoxemia detection versus nondetection in association with GN bacteremia varies across different species
types of GN bacteremia (Figure 5.2.8, Table 5.2.5) as follows;
2.6. The detection of endotoxemia in association with bacteremias
with Neisseria meningitidis has the strongest prognostic value. Whilst this is
the largest category of GN bacteremia, these patients have mostly come from
two studies [11, 12] from a single center. The possibility that this infection is
epidemic and clonal in the region in which this center is located needs to be
considered. The limited number and lack of diversity of these clusters is
reflected in the wide 95% CI calculated by the GEE method.
2.7. Surprisingly, the detection of endotoxemia in association with E.
coli bacteremia, which is the second largest GN bacteremia category, has no
prognostic value in any setting. This experience is derived from eight ICU
studies. The experience among small numbers from an additional 6 non-ICU
-160-
Chapter 6: Summary & conclusion
studies is not significantly different. It is notable that this lack of prognostic
value with E. coli bacteremia is at variance with the experience for GN
bacteremia overall as this provides a basis for exploring the conflicting results
of individual studies.
2.8. There is a significant interaction effect between the detection of
endotoxemia and the type of Enterobacteriaceae (E. coli versus non-E. coli)
bacteremias on the prognosis.
2.9. This last finding is surprising given the expected commonality of
the hexa-acyl structure of lipid-A amongst all Enterobactereiaceae whether E.
coli or non-E. coli.
3. Are the disparate findings among the published studies reconcilable?
3.1. There is a large range in diagnostic odds ratio among the studies.
The results of the three original studies [25, 54, 75] published in 1972-1975
could not be considered have been outlier or atypical in the analysis of the 107
studies overall. (Figure 5.1.5).
3.2. Toward the detection of GN bacteremia, the sensitivity limit of
the limulus assay is not the only limiting factor and other factors bearing on the
diagnostic utility of the assay such as the setting, the plasma preparation
method and the types of GN bacteremias are at least as important.
3.3. For studies conducted in an ICU setting, the mortality RD
associated with the detection versus non-detection of endotoxemia (without
regard to the co-detection of GN bacteremia) is similar amongst studies whose
data is reported in the ‗two group‘ versus the ‗four group‘ format (Table 5.2.1).
3.4. By contrast, for studies conducted outside an ICU setting whose
data is analyzed in the ‗two group‘ format, the mortality RD associated with
the detection versus non-detection of endotoxemia (without regard to the codetection of GN bacteremia) is generally higher and more variable versus that
amongst studies conducted in an ICU setting (Table 5.2.1).
3.5. In regard to the interpretation of the limulus assay results in
relation to prognostic importance, the setting, the mix of GN bacteremias and
the DOR obtained all appear to be important determinants towards the findings
in each study (Section 5.2.2/.3/.4).
3.6. The seemingly contrary results reported in the literature could be
reconciled by the mix of GN bacteremia seen, with a higher proportion of E.
coli potentially able to account for the ‗negative‘ conclusions of some studies
in comparison to studies with a lower proportion of E. coli. (Figure 5.2.10).
3.7. The major contribution from this thesis is that the interpretation
of the detection of endotoxemia in a given patient in relation to the literature
experience as analysed within this thesis [1-107] and possibly even more
broadly [498-673] cannot be given without knowledge of three contextual
factors; whether the patient in question has a GN bacteremia, if so what type of
GN bacteremia and the background mortality risk of the population in which
the patient belongs.
Chapter 6: Summary & conclusion
-161-
6.2 Recapitulation of study limitations and strengths
Limitations. The limitations of the research described here have been
mentioned in 5.1.3 and 5.2.4. Several additional limitations are listed here;
1.
The data is observational; this limits any causal interpretation due to the
inherent confounding and bias that exists such as selection, sampling and publication
biases.
2.
The data has been abstracted for each meta-analysis by a single author.
However, this is not unprecedented in this area [231-232, 685]. These have each been
submitted for peer review and accepted for publication as meta-analyses and not as
systematic reviews.
3.
The data was incomplete for the majority of the 107 studies.
4.
The definitions of study populations were imprecise for many studies,
more so for those undertaken outside of an ICU setting.
5.
Some of the sub-groups of studies (e.g. pediatric) were small (< 8
studies) and this necessitated analysis as a composite sub-group (e.g. with oncology
studies).
6.
The exposure classification (endotoxemia positive versus negative)
varies from study to study due to variable endotoxemia detection thresholds.
7.
Some important study parameters, and in particular mortality census,
were mssing. Moreso for those studies undertaken outside an ICU setting.
8.
GN bacteremia has been used as a reference standard. The limitations
and strengths relating to this have been discussed in section 2.4
9.
With the exception of six ‗two group‘ supplementary studies,
endotoxemia detection has been defined with respect to activity in the limulus assay
and generalization of these findings beyond endotoxemia as defined by detection with
the limulus assay is cautioned.
10.
The influence of underlying patient risk as presented in this analysis
may be misleading. Underlying patient risk has been defined at the group level for each
study using the mortality rate of the reference group (Group 4; neither endotoxemia nor
GN bacteremia detected) rather than at the patient level (see below).
11.
There is considerable variability with respect to study sizes. As a
consequence of this there are slight differences in summary effect sizes depending on
the method of data aggregation used in the analysis.
12.
The data for the 107 studies is somewhat dated having been mostly
published more than ten years ago.
13.
The contributory effect of other important factors such as appropriate
antibiotic therapy could not be examined.
14.
That the origin of the endotoxemia is from the concomitant GN
bacteremia is a presumption.
15.
The most surprising finding here, that endotoxemia has no prognostic
relevance for mortality when found in the co-presence of E. coli bacteremias, cannot be
readily corroborated.
-162-
Chapter 6: Summary & conclusion
16.
There being only 51 non-Enterobactericeae GN bacteremias from seven
studies undertaken in an ICU setting (Table 5.2.5), there is insufficient data to derive a
definitive answer to the question of the effect of non-hexa-acyl lipid-A structure toward
clinical relevance of endotoxemia detection.
17.
There are limitations to the interpretations of the findings of a metaanalysis particularly with the cautions required for any post-hoc analysis.
Strengths. On the other hand, there are a number of strengths of the research
described here.
1.
A large number of studies in several categories have been found after
repeated searching of the literature.
2.
The data has been expanded by a call for data with additional data
obtained by personal communication.
3.
The abstracted data is appended to this thesis (section 7.1/.2). For each
of the eight compiled research publications appended to this thesis, the data as
abstracted from the original studies was available for scrutiny during the peer review
process prior to publication and remains accessible there. Moreover, the data has been
updated as new data has come to hand.
4.
Meta-analytic methods have been used to enable the quantitative
overview of a complex literature that could not otherwise be achieved. In particular, the
techniques used have enabled estimations of measures of variability which are of
particular interest.
5.
Recently developed meta-analytic methods have been used as these have
become available and these are optimally suited for;
variable endotoxemia detection thresholds between studies,
variable study size,
variable underlying patient risk (at the group level),
derivation of measures of heterogeneity,
graphical presentation of study results to enable a visual overview of the
extensive literature and identification of studies with atypical results,
enabling estimations of the number of studies that would need to be
unpublished or ‗missing‘ to overturn the findings,
data analysis issues arising from inclusion of study groups that are small,
unbalanced and clustered,
enabling comparisons with other meta-analytic summaries of diagnostic
and prognostic tests applicable to the study populations of interest as
reported in the literature (Table 2.3.2).
6.
The data as abstracted and summated by Elin has been re-analyzed to
enable the findings as reported by Elin to be compared with those obtained with the
more robust contemporary methods used here (Figure 2.8/.9/.10).
7.
The original conclusions are robust to re-analysis with new data.
Namely, the findings as published in JCM ’94 and JCM ’95 with 45 and 11 studies,
Chapter 6: Summary & conclusion
-163-
respectively are in each publication broadly the similar, albeit with attenuated OR‘s as
in the update for this thesis with 88 and 32 studies, respectively (Table 5.1.1 and Table
5.2.3).
8.
Two attempts have been made to reconcile the findings here with the
broader literature. Firstly, special effort has been made to compare the results of studies
with (e.g. the ‗four group‘ studies) versus without (e.g. the ‗two group‘ studies)
complete data (Table 5.1.2) and the findings are consistent. Second, the findings here
for mortality rates for the three main GN bacteremia types have been compared for
each of these GN bacteremias to the broader literature. These are mostly within the
inter-quartile range with the exception of non-E. coli Enterobactereiaceae (Figure
5.2.9).
9.
The main study questions, the effect of limulus assay version, and the
effect of hexa-acyl versus non-nexa-acyl lipid-A structure toward clinical relevance of
endotoxemia detection, can each be regarded as pre-specified in that the questions
arose independently of the data.
6.3 Future prospects
With the main findings of the research together with the limitations and
strengths of the analysis listed above, it is timely to address the implications of the
findings.
1. The diverse and conflicting literature can be reconciled if it is accepted that,
regardless of the sensitivity level of the endotoxemia detection assays that have
been used, not all GN bacteremias have detectable endotoxemia and conversely,
not all patients with endotoxemia will have GN bacteremia. The first
implication of this is that the detection of endotoxemia will not serve as a
practical rapid screening test for GN bacteremia (in contrast to the case for
fluids other than blood) (see Figures 2.7 & 2.8).
2. The second implication is that unrecognized confounders, namely underlying
patient risk and type of GN bacteremia, could account for the different findings
among different studies with respect to the prognostic relevance of endotoxemia
detection. Also, study size needs to be recognized as an important confounding
factor together with the compounding effect of publication bias. Variable
reporting as a result of confounding and study imprecision is not unusual in the
literature. However, only through the application of meta-analytic techniques as
used in this thesis can these influences be recognized.
3. The influence of underlying patient risk as presented in this analysis may be
-164-
Chapter 6: Summary & conclusion
misleading. As stated above, underlying patient risk has been defined for each
study using the mortality rate of the reference group (Group 4; neither
endotoxemia nor GN bacteremia detected, being generally the largest group of
each study) rather than specifically at the patient level. The underlying risk of
Group 4 is unlikely to be representative of the underlying patient risk for the
patients in other groups in each study, especially so in studies that were small or
had used imprecise patient inclusion criteria to define the study populations.
However, the pertinent inference from this analysis is that in studies with a high
background underlying patient risk (i.e. ICU settings) endotoxemia detection
has a prognostic value which is weak and adds little to that provided by GN
bacteremia detection.
4. As stated above, the data is observational and this limits any causal
interpretations, in particular whether endotoxemia has a causal influence in the
outcome of patients with suspected GN sepsis. However, the detection of
endotoxemia in a study undertaken in the ICU setting is associated with a
mortality risk difference of 10% (Figure 5.2.2 & 5.2.3) or an OR of 1.4 to 1.6
(Figure 5.2.1 & Table 5.2.1). Hence, the evidence accumulated here would
suggest that the prognostic inference associated with the detection of
endotoxemia (without regard to the detection of GN bacteremia) overall is
similar to that associated with the detection of GN bacteremia (without regard
to the detection of endotoxemia) as discussed in section 2.4.2.
5. The analysis here in relation to the prognostic inference associated with the
detection of endotoxemia raises the likelihood that this is unequal for different
GN bacteremias. Surprisingly, endotoxemia may have a strong prognostic
inference in patients with non- E. coli Enterobactericeae GN bacteremias versus
no prognostic inference in E. coli bacteremias (Table 5.2.5).
6. Moreover, the mix of GN bacteremias within individual studies and their
distribution between groups 1 and 2 may be an unrecognised confounder
leading to the conflicting results of studies of the prognostic value of
endotoxemia. In this regard, the relative rankings of the three common GN
bacteremia types in relation to both their relative frequency and mortality risk
(Table 2.4.1/.1/.2/.3/.4/.5) overall versus the atypical rankings found amongst
the smaller sized studies is of note (as is apparent in Figures 2.12/.13/.14/.15).
Chapter 6: Summary & conclusion
-165-
7. The effect of endotoxin derived from non-E. coli Enterobactericeae warrant
greater study in animal models and other pre-clinical studies as it has been
relatively under-studied in comparison to the extent of experimental study
devoted to the effect of endotoxin derived from E. coli.
8. There are too few studies [100-107] of the newer assays (SRE, EAA) for the
detection of endotoxemia to determine whether these perform any differently
from studies that used the limulus assay. However, the confounding influences
mentioned above will need to be considered in any comparison.
9. The results of randomized trials of novel anti-endotoxin therapies warrant a reexamination in the light of the findings here. In particular, the types of GN
bacteremia under study needs attention. However, mostly, the types of GN
bacteremias were either not stated (Tables 2.5.1&.3), or the studies were
restricted to meningococcemia primarily (Table 2.5.2). Few of the studies
found a significant reduction in mortality with the anti-endotoxemia treatment
(as discussed in section 2.5.2). However there are two studies (as indicated
within Table 2.5.3) in which a significant reduction in mortality with the antiendotoxemia treatment was noted and which also had an unusually high
proportion of non- E. coli Enterobactericeae versus E. coli and versus total
numbers of GN bacteremias (71 versus 87 versus 200 [350] and 30 versus 12
versus 49 [442]). Whilst this is a speculative conclusion derived from a post hoc
analysis, it is worthy of further examination in future randomized trials.
10. The research questions in this thesis have been limited to endotoxemia detection
with the limulus assay and gram-negative bacteremia with particular interest in
the patient group with sepsis. Newer techniques are emerging to assist progress
toward finding the pathogenesis of sepsis. In particular, newer methods of
detection of blood stream infection in the patient group with sepsis are
emerging. These techniques, such as MALDI-TOF (Matrix Assisted Laser
desorption-ionization Time Of Flight mass spectrometry fingerprinting) offer
not only rapid detection but also rapid identification [719]. Also of promise are
methods for the molecular profiling of both the bacterial and host molecules in
the diagnosis and investigation of sepsis. There are methodological challenges
associated with the evaluation of these molecular methods (discussed in section
2.3.2). To date, the results of cytokine profiling in sepsis patients have not
-166-
Chapter 6: Summary & conclusion
always been as expected. For example, evidence supporting the classic two
phase (inflammatory/ anti-inflammatory) model of sepsis has not yet been
found. One study suggested that the incremental cost in using these PCR
methods is justifiable only when the rate of inadequate initial anti-microbial
therapy is greater than 25% [720].
11. Finally, two recent studies [333, 720] are of interest in relation to research
question two of this thesis (endotoxemia versus outcome -mortality). These
studies address the predictive value of microbiological identification made
amongst culture negative sepsis patients using polymerase chain reaction (PCR)
methods that is additional to that obtained with culture detection. The smaller of
the two studies [333] examined blood culture versus PCR methods for 126
critically ill (criteria not specified) adult ICU patients. The finding of this first
study [333] was that the SOFA score was similar for the 64 PCR positive (8.2
(SD 3.6)) versus the 134 PCR negative (8.3 (SD 4.0)) patients, respectively.
12. The larger of the two studies [720] investigated the prognostic value of using a
broad (not limited to GN bacteremia) multiplex PCR method applied to 221
sepsis patients among which the 30-day non-survival was 59/221 (27%) overall.
The patients of this study could be divided into four groups corresponding to
those that were both PCR and blood culture positive (not limited to GN
bacteremias), those that were only blood culture positive, those that were only
PCR positive, and a fourth group that was positive for neither. Note that these
four groups are homologous to the four groups as detailed within section 5.2 of
this thesis. The finding of this study [720] was that the 30-day mortality was
15/33 (45%), 3/11 (27%), 9/40 (23%), and 10/37 (27%), for groups 1 to 4,
respectively. Note the risk differences between the percentages between groups
1 to 3 each versus group 4 (as the reference group) in this study are similar to
the risk differences reported in the footnotes to Figures 5.2.4/.5/.6. In particular,
for this study [720] the greatest risk difference (18%) versus group 4 is that
observed for group 1 and in this thesis, the risk difference was 16.1% (Figure
5.2.4).
-167-
7. Catalogue of studies and personal correspondence
Column headings and explanatory notes for 7.1 & 7.2:
1. First author and reference
2. LAL; G = Gelation version C = chromogenic version; SRE & EAA are non-limulus
assays for endotoxemia (see 2.2); R = rabbit assay.
3. ng/ml = sensitivity limit of assay to the internal endotoxin standard
4. Pl = method used to pre-treatment plasma prior to LAL assay
5. Set_ = Setting & _Pop = population;
A_ = Adult
o _fever/shock/sepsis = fever/shock/clinically suspected sepsis
o _ASCCM shock
o _GI = gastrointestinal conditions (e.g. pancreatitis, GI surgery, cirrhosis)
o _UTI = documented urinary tract infection
o _malaria = malaria
P_ = pediatric
o _CVS = cardiovascular surgery
o _maln = malnutrition
o _HUS = hemolytic uremic syndrome
o _NEC = necrotizing entero-colotis
O_ = oncology
ICU_ = Intensive care unit
o 2_shock = shock
R_ = studies restricted to;
_Ty = typhoid
_Mg = meningococcal disease
_Pg = plague
_S = Salmonellosis
_Ml = melioidosis
6. N = Total number of patients ( = TP + FN + FP + TN)
7. TP = True positives; GN bacteremia with concomitant bacteremia
8. FN = False negatives; GN bacteremia without concomitant bacteremia
9. FP = False positives; endotoxemia without concomitant GN bacteremia
10. TN = True negatives; Neither endotoxemia nor GN bacteremia
11. GNB% = Percent of patients with GN bacteremia (= (TP+FN)/T)
12. QS = quality score symbols code; ●/ψ/§ = 1; ○ = 0
1; study inclusion criteria specified
2; sampling strategy; i.e. blood culture with each endotoxemia sampling
3; outcomes detailed as patient survival (‗ψ‘ indicates data used in 5.2)
4; listing of GN bacteremia isolates given (‗§‘ indicates data used in 5.1.2, 5.2.3 or
5.2.4)
13. code = study code used in figures 5.1.5 & 5.2.3 to assist with literature reconciliation
14. notes
NE = non-English language study
PE = per-episode data
pc = Information received by personal communication
R = study restricted to specified GN bacteremias
D = duplicate study (only one study shown)
HA-1A = anti-endotoxin monoclonal antibody (HA-1A) given to all
-168-
Chapter 7: Catalogue of studies
7.1 Summary of included studies
First
author
Ref
LAL
ng/
ml
Pl
Pop
Set
N
TP
FN
FP
TN
GNB
QS
nb
%
Adinolfi
‗87
Ahmed
‗04
Bailey ‗76
Code
1
g
0.3
DH
Ty
R
21
7
7
2
5
67
●●ψ§
t-R
2
c
0
pH
GI
P
42
9
1
12
19
26
●○ψ§
p1
3
g
5
Ch
GI
A
24
1
0
12
11
4
Behre ‗92
Berger
‗95
Bigatello
‗87
Billard‘94
4
c
0.01
DH
shock
O
21
5
0
12
4
24
●○ψ§
●○○○
5
g
GI
A
32
1
0
20
11
3
●○ψ§
6
c
GI
A
21
1
1
18
1
10
○○○§
7
c
NS
NS
shock
ICU
18
6
0
4
8
33
●○ψ○
Bion ‗94
Bjorvatn
‗84
Bose ‗02
8
c
0.02
NS
9
0
10
c
NS
11
c
0
‗02
12
c
Buller ‗85
Butler
‗73
Butler
‗76
Butler ‗78
13
c
0
14
g
15
GI
ICU
52
1
1
31
19
4
●●ψ§
Mg
R
4
4
0
-
-
0
●●●§
R
NS
GI
A
20
0
0
17
3
0
●●●○
DH
Mg
R
42
24
11
0
7
83
●●ψ§
Mg
R
133
64
30
4
33
71
●●ψ§
R
Mg -R,
pc
Mg -R,
pc
DH
GI
A
56
12
1
4
39
23
●○○○
1
Ch
Pg
R
10
2
0
7
1
20
●●ψ§
R
g
5
Ch
Pg
R
10
3
2
0
5
50
●●ψ§
R
16
g
1
Ch
Ty
R
21
0
13
0
8
62
●●●§
Byl ‗01
17
g
0
sepsis
A
27
8
5
3
12
44
●●ψ§
R
a0 - pc
Casey ‗92
Clumeck
‗77
Coopersto
ck ‗85
Danner
‗91
Das ‗73
18
g
NS
CVS
P
24
1
0
15
8
4
●●ψ§
19
g
3
pH
sepsis
A
48
9
3
2
34
25
●●○§
a1 -NE
20
g
1
DH
sepsis
P
43
10
1
6
26
26
●○ψ§
p3
21
c
0.01
DH
shock
ICU
100
11
8
32
49
19
●●ψ§
i1 -pc
22
g
10
Ch
febrile
A
54
9
2
7
36
20
○○●○
23
c
0.01
NS
sepsis
ICU
18
4
2
6
6
33
●●ψ§
24
c
0.01
DH
153
7
12
28
106
12
○●○○
i2
PE
24
c
0.01
DH
sepsis
P
21
3
5
7
14
38
○●○○
PE
24
c
0.01
DH
sepsis
A
113
4
7
21
92
10
○●○○
PE
25
g
10
Ch
sepsis
A
237
8
15
40
174
10
●●○○
PE
Brandtzaeg
‗89
Brandtzaeg
Dofferhoff
‗92
Dolan ‗87
Dolan
(child) ‗87
Dolan
(adult) ‗87
Elin ‗75
Chapter 7: Catalogue of studies
-169-
7.1 Summary of included studies (continued)
First
author
Ref
LAL
ng/
ml
Pl
Pop
Set
N
TP
FN
FP
TN
GNB
QS
nb
%
Engervall
‗95
Engervall
‗97
Exley ‗92
Feldman
‗74
Fink ‗84
Fossard
‗74
Foulis ‗82
Fugger
‗90
Garcia
Curiel ‗79
Gardlund
‗95
Code
26
c
0.005
NS
fever
O
26
6
3
9
8
35
●●ψ§
pc, PE
27
c
0.005
NS
fever
O
24
2
4
4
14
25
●●ψ§
o2 - pc
28
c
0.03
DH
GI
P
8
2
0
5
1
25
●○●§
29
g
1
pH
30
g
31
g
0.1
32
0
33
sepsis
P
89
2
28
0
53
34
●●●§
febrile
ICU
22
3
0
14
1
14
○○○○
Ch
GI
A
25
2
0
22
1
8
●●○§
5
NS
GI
A
26
1
0
12
13
c
1
Ch
GI
ICU
24
5
0
17
2
21
●○●○
34
g
1
NS
shock
A
42
1
15
3
23
38
●●○§
i4 – NE
35
c
Mg
R
10
9
1
-
-
●○●§
R
‗99
Goldie
‗95
Guidet
‗94
Harris ‗84
Harthug
‗83
Hass ‗86
Hynninen
‗95
Jacob ‗77
36
c
0.01
DH
UTI
A
25
9
3
4
9
48
●●ψ§
a2 – pc
37
c
0
DH
sepsis
ICU
133
9
3
83
38
9
●●ψ§
i5 – pc
38
c
0.01
DH
sepsis
ICU
93
24
9
20
40
35
●●ψ§
i6 – pc
39
c
0.01
DH
fever
O
43
5
1
12
25
14
●○○○
PE
40
c
0.5
pH
Mg
R
5
2
0
1
2
40
○●●§
R
41
c
0.01
PCA
fever
P
36
3
0
8
25
8
●○○§
42
c
0.01
NS
sepsis
O
123
3
24
0
96
22
●●ψ§
43
g
1
DH
GI
A
24
1
0
3
20
4
●●○§
Jirillo ‗75
44
g
1
Ch
maln
P
10
1
1
4
4
20
Kelley ‗85
Kelsey
‗82
Ketchum
‗97
Kinoshita
‗84
Kiss ‗83
45
g
NS
Ch
GI
A
20
0
0
13
7
0
●○○§
●●○○
46
g
50
pH
sepsis
P
30
1
1
0
28
7
○●○§
47,
48
c
0.01
pH
sepsis
ICU
356
10
39
109
198
14
●●ψ§
49
g
NS
Ns
sepsis
A
19
9
0
7
3
47
●●○○
50
g
1
DH
GI
A
100
0
0
9
91
0
●○○○
Klein ‗88
51
g
5
Ch
maln
P
16
0
1
7
8
6
●●ψ§
G-bourboulis
p2
D
○○ψ○
NE
i7 - pc
-170-
Chapter 7: Catalogue of studies
7.1 Summary of included studies (continued)
First
author
Ref
LAL
ng/
ml
Pl
Pop
Set
N
TP
FN
FP
TN
GNB
Lau ‗96
52
c
10
DH
GI
A
40
8
3
20
9
Levin ‗70
Levin (all)
‗72
Levin
53
g
5
Ch
sepsis
A
98
10
5
7
54
g
5
Ch
sepsis
A
218
17
14
(sBP<100)
54
g
5
Ch
shock
A
61
14
54
g
5
Ch
sepsis
A
157
55
c
0.02
DH
GI
A
56
g
1
pH
Sal
56
g
1
pH
57
g
5
58
c
QS
Code
28
●●ψ§
nb
a3 - pc
76
15
●●ψ§
19
168
14
●●ψ§
a4
3
6
38
28
●●ψ○
a4
6
11
10
130
11
●●ψ○
a4
21
1
0
13
7
5
●●ψ§
R
25
4
0
9
12
16
●●ψ§
R
Ty
R
14
8
4
1
1
86
●●ψ§
R
pH
sepsis
A
83
1
15
1
66
19
●●●§
a5
5
DH
sepsis
A
52
14
1
21
16
29
●●○§
i8 – pc
29
●●○§
i8 -pc
pc
%
‗72
Levin
(sBP>100)
‗72
Lumsden
‗89
Magliulo
‗76
Magliulo
‗76
MartinezG ‗73
Massignon
‗96
Massignon
58
c
5
DH
sepsis
A
52
12
3
21
16
59
0
0.05
DH
sepsis
ICU
17
17
0
-
-
60
g
0.1
Ch
fever
A
31
15
0
16
0
48
●●○§
61
c
NS
ICU
O
28
1
2
8
17
11
●●○§
‗93
Ng ‗95
Oberle
‗74
Opal ‗99
(all)
Opal ‗99
(lo)
Opal ‗99
(hi)
Pearson
‗85
Prins ‗95
62
g
0.04
NS
Ty
R
22
0
16
0
6
73
●○○§
63
g
NS
DH
sepsis
ICU
56
34
4
5
13
68
●●○○
64
g
0.5
Ch
maln
P
23
3
0
6
14
13
●●○§
65
c
DH
sepsis
ICU
759
114
21
496
128
18
●●ψ§
i9 – pc
65
c
.02
DH
sepsis
ICU
759
114
21
496
128
18
●●ψ§
i9lo – pc
65
c
.66
DH
sepsis
ICU
759
63
72
241
383
18
●●ψ§
i9hi – pc
66
c
0.1
DH
sepsis
A
45
9
1
2
32
22
●○○○
67
c
0.04
NS
UTI
A
30
5
5
2
18
33
●●●§
Prins ‗98
68
0
Mg
R
31
30
1
-
-
Prytz ‗76
Scheifele
‗81
Scheifele
‗85
Shenep
‗88
Simpson
‗00
69
g
10
DH
GI
A
52
0
2
17
33
4
●●●§
●●○○
70
g
10
DH
sepsis
P
63
5
3
10
45
13
●●○○
71
g
0.2
DH
NEC
P
47
3
2
20
22
11
●○○§
72
g
0.01
DH
sepsis
P
26
9
1
3
13
38
●●ψ§
73
c
0.002
DH
Ml
R
30
15
0
10
5
50
●●ψ§
Strutz ‗99
Stumache
r ‗73
Suda ‗95
74
c
0.01
DH
Sepsis
ICU
28
5
5
8
10
36
●●ψ§
ml - R,
pc
i0
75
g
0.5
Ch
sepsis
A
139
28
37
26
48
47
●●ψ§
a7
76
c
0.01
PCA
sepsis
P
7
2
0
5
0
29
●●○○
‗96
Maury ‗98
McCartney
‗83
McCartney
‗87
McGladdery
●●●§
R
pc
R, pc
p4
Chapter 7: Catalogue of studies
-171-
7.1 Summary of included studies (continued)
First
author
Ref
LAL
ng/
ml
Pl
Pop
Set
N
TP
FN
FP
TN
GNB
QS
nb
%
Suyasa
‗95
Tanaka
‗87
Tarao ‗77
Thomas
‗84
Togari ‗83
Trinchet
‗83
Uriu ‗02
Usawattan
akul ‗79
Usawattan
akul ‗85
Van
Deventer
‗87
Van
Deventer
‗88
Van
Dissel
(HA-1A)
‗93
van
Langevelde
‗00
Van
Wieringen
‗76
Venet ‗00
Watzke
‗87
Wig ‗98
Wortel
(all) ‗92
Wortel
(placebo)
‗92
Wortel
(HA-1A)
‗92
Yoshida
‗93
Code
77
c
0.01
PCA
Ty
R
21
4
0
5
12
0
●○ψ§
78
g
75
PCA
UTI
A
17
1
0
13
3
6
●○○§
79
g
5
Ch
GI
A
41
0
0
23
18
0
●●●○
80
c
0.01
DH
sepsis
A
400
14
9
12
365
6
●●○○
81
g
0.5
pH
shock
P
10
1
0
7
2
10
●●●§
82
g
1000
DH
GI
A
48
1
0
13
34
2
●○○§
83
g
0.01
NS
shock
A
24
8
7
2
7
63
●●●§
84
g
50
Ch
fever
A
58
34
6
2
16
69
●○○○
85
g
50
Ch
malaria
A
43
9
4
7
23
30
●○○○
86
c
0.01
DH
fever
A
76
15
23
0
38
50
●●○○
87
c
0.01
DH
fever
A
473
17
36
14
406
11
●●○§
a8
88
c
NS
NS
sepsis
ICU
14
4
0
7
3
29
●○ψ§
HA1A
89
c
0.01
NS
fever
A
448
24
24
76
324
11
●●ψ§
pc
90
g
0.5
Ch
HUS
P
35
6
0
1
28
17
●○○§
91
c
0.01
DH
sepsis
ICU
116
19
4
42
51
20
●●●○
92
c
0.01
DH
fever
ICU
27
8
4
4
11
44
●●ψ§
93
c
NS
NS
GI
A
14
0
0
10
4
0
●●●○
94
c
0.01
DH
sepsis
ICU
82
18
14
9
41
39
●●ψ○
94
c
0.01
DH
sepsis
ICU
41
8
5
3
25
0
●●ψ○
94
c
0.01
DH
sepsis
ICU
41
10
9
6
16
0
●●ψ○
HA1A
95
c
0.003
NS
fever
O
136
21
9
35
71
22
●●ψ§
o3 - pc
R
NE
a11 NE
-172-
Chapter 7: Catalogue of studies
7.2 Summary of supplementary studies
7.2.1 ‘4 group’ studies using non-LAL assays
First
author
Ref
ng/
ml
LAL
Pl
Pop
Set
N
TP
FN
FP
TN
GNB
QS
Beller ‗73
Douglas
‗63
McGill
‗70
Porter ‗64
Dholakia
‗11
Yaguchi
‗12
7.2.2
First
author
96
R
abort
A
12
3
0
4
5
25
●●●●
97
R
sepsis
A
16
2
1
8
5
19
●○●●
98
R
sepsis
A
41
1
7
10
26
20
●●●○
99
R
shock
A
5
9
24
100
EAA
septic
P
100
3
2
52
43
5
●○●●
@ - pc
101
EAA
sepsis
ICU
210
16
4
90
100
10
●○ϕ●
ii5, @
10
●●●●
‘2 group’ studies using LAL or non-LAL assays
Ref
LAL
ng/
ml
Pl
Pop
Set
N
Mortality proportions
ETX+
n/N
Casey ‘93
McGill
‗67
Kollef
‗97
Dholakia
‗11
Marshall ‘
02
Marshall ‘
04
Valenza
‗09
Venet ‗00
Yaguchi
‗12
Code
nb
%
GNB
QS
Code
nb
ii1
ETXn/N
%
sepsis
ICU
97
11
22
36
75
16
●○ϕ○
sepsis
A
23
6
10
10
13
NS
●●●●
SRE
sepsis
ICU
140
8
29
17
114
NS
●●ϕ○
100
EAA
sepsis
P
100
7
55
3
45
5
●○●●
105
EAA
sepsis
ICU
74
8
31
13
43
●○ϕ○
106
EAA
sepsis
ICU
857
70
595
45
262
●○ϕ●
ii4
107
EAA
GI
ICU
102
2
34
1
68
NS
●○ϕ○
ii6
91
c
sepsis
ICU
106
26
61
21
55
20
●○ϕ○
ii3
101
EAA
sepsis
ICU
210
28
106
18
104
9.5
●●ϕ●
ii5
102
C
103
R
104
10
0.01
DH
ii2
-173-
8. References
1.
Adinolfi LE, Utili R, Gaeta GB, Perna P, Ruggiero G: Presence of endotoxemia and its relationship
to liver dysfunction in patients with typhoid fever. Infection 1987, 15(5):359-362.
2.
Ahmed T, Azam MA, Armed N, Jamil KM, Hassan F, Ogura N, Tamura H, Yokochi T: Detection of
endotoxin in sera from children hospitalized for treatment of diarrhea in Bangladesh. J Endotoxin
Res 2004, 10(4):223-228.
3.
Bailey ME: Endotoxin, bile salts and renal function in obstructive jaundice. Br J Surg 1976,
63(10):774-778.
4.
Behre G, Schedel I, Nentwig B, Wormann B, Essink M, Hiddemann W: Endotoxin concentration in
neutropenic patients with suspected gram-negative sepsis: correlation with clinical outcome and
determination of anti-endotoxin core antibodies during therapy with polyclonal immunoglobulin Menriched immunoglobulins. Antimicrob Agents Chemother 1992, 36(10):2139-2146.
5.
Berger D, Boelke E, Stanescu A, Buttenschoen K, Vasilescu C, Seidelmann M, Beger HG:
Endotoxemia and mediator release during colonoscopy. Endoscopy 1995, 27(9):671-675.
6.
Bigatello LM, Broitman SA, Fattori L, Di Paoli M, Pontello M, Bevilacqua G, Nespoli A:
Endotoxemia, encephalopathy, and mortality in cirrhotic patients. Am J Gastroenterol 1987,
82(1):11-15.
7.
Billard J-l, Berthier-Berrada S, Page Y, et al. (1994) Endotoxemia in human septic shock: relation to
gastric intra-mucosal pH. Crit Care Med 22: A113
8.
Bion JF, Badger I, Crosby HA, Hutchings P, Kong KL, Baker J, Hutton P, McMaster P, Buckels JA,
Elliott TS: Selective decontamination of the digestive tract reduces gram-negative pulmonary
colonization but not systemic endotoxemia in patients undergoing elective liver transplantation. Crit
Care Med 1994, 22(1):40-49.
9.
Bjorvatn B, Bjertnaes L, Fadnes HO, Flaegstad T, Gutteberg TJ, Kristiansen BE, Pape J, Rekvig OP,
Osterud B, Aanderud L: Meningococcal septicaemia treated with combined plasmapheresis and
leucapheresis or with blood exchange. Br Med J (Clin Res Ed) 1984, 288(6415):439-441.
10. Bose SM, Verma GR, Mazumdar A, Giridhar M, Ganguly NK: Significance of serum endotoxin and
antiendotoxin antibody levels in predicting the severity of acute pancreatitis. Surg Today 2002,
32(7):602-607.
11. Brandtzaeg P, Kierulf P, Gaustad P, Skulberg A, Bruun JN, Halvorsen S, Sorensen E: Plasma
endotoxin as a predictor of multiple organ failure and death in systemic meningococcal disease. J
Infect Dis 1989, 159(2):195-204.
12. Brandtzaeg P, Hogasen K, Kierulf P, Mollnes TE: The excessive complement activation in fulminant
meningococcal septicemia is predominantly caused by alternative pathway activation. J Infect Dis
1996, 173(3):647-655.
13. Buller HR, ten Cate JW, Sturk A, Thomas LL: Validity of the endotoxin assay in post surgical
patients. Prog Clin Biol Res 1985, 189:405-417.
14. Butler T, Levin J, Do Quang C, Walker RI: Bubonic plague: detection of endotoxemia with the
limulus test. Ann Intern Med 1973, 79(5):642-646.
15. Butler T, Levin J, Linh NN, Chau DM, Adickman M, Arnold K: Yersinia pestis infection in Vietnam.
II. Quantiative blood cultures and detection of endotoxin in the cerebrospinal fluid of patients with
meningitis. J Infect Dis 1976, 133(5):493-499.
16. Butler T, Bell WR, Levin J, Linh NN, Arnold K: Typhoid fever. Studies of blood coagulation,
bacteremia, and endotoxemia. Arch Intern Med 1978, 138(3):407-410.
17. Byl B, Clevenbergh P, Kentos A, Jacobs F, Marchant A, Vincent JL, Thys JP: Ceftazidime- and
imipenem-induced endotoxin release during treatment of gram-negative infections. Eur J Clin
Microbiol Infect Dis 2001, 20(11):804-807.
-174-
Chapter 8: References
18. Casey WF, Hauser GJ, Hannallah RS, Midgley FM, Khan WN: Circulating endotoxin and tumor
necrosis factor during pediatric cardiac surgery. Crit Care Med 1992, 20(8):1090-1096.
19. Clumeck N, Lauwers S, Kahn A, Mommens M, Butzler JP: [Contribution of the "limulus test" to the
diagnosis of endotoxemias and meningitis due to gram negative bacteria]. Nouv Presse Med 1976,
6(17):1451-1454.
20. Cooperstock M, Riegle L: Plasma Limulus gelation assay in infants and children: correlation with
gram negative bacterial infection and evidence for "intestinal endotoxemia". Prog Clin Biol Res
1985, 189:329-345.
21. Danner RL, Elin RJ, Hosseini JM, Wesley RA, Reilly JM, Parillo JE: Endotoxemia in human septic
shock. Chest 1991, 99(1):169-175.
22. Das J, Schwartz AA, Folkman J: Clearance of endotoxin by platelets: role in increasing the accuracy
of the Limulus gelation test and in combating experimental endotoxemia. Surgery 1973, 74(2):235240.
23. Dofferhoff AS, Bom VJ, de Vries-Hospers HG, van Ingen J, vd Meer J, Hazenberg BP, Mulder PO,
Weits J: Patterns of cytokines, plasma endotoxin, plasminogen activator inhibitor, and acute-phase
proteins during the treatment of severe sepsis in humans. Crit Care Med 1992, 20(2):185-192.
24. Dolan SA, Riegle L, Berzofsky R, Cooperstock M: Clinical evaluation of the plasma chromogenic
Limulus assay. Prog Clin Biol Res 1987, 231:405-416.
25. Elin RJ, Robinson RA, Levine AS, Wolff SM: Lack of clinical usefulness of the limulus test in the
diagnosis of endotoxemia. N Engl J Med 1975, 293(11):521-524.
26. Engervall P, Granstrom M, Andersson B, Bjorkholm M: Monitoring of endotoxin, interleukin-6 and
C-reactive protein serum concentrations in neutropenic patients with fever. Eur J Haematol 1995,
54(4):226-234.
27. Engervall P, Granstrom M, Andersson B, Kalin M, Bjorkholm M: Endotoxemia in febrile patients
with hematological malignancies. Relationship of type of bacteremia, clinical findings and serum
cytokine pattern. Infection 1997, 25(1):2-7.
28. Exley AR, Leese T, Holliday MP, Swann RA, Cohen J: Endotoxaemia and serum tumour necrosis
factor as prognostic markers in severe acute pancreatitis. Gut 1992, 33(8):1126-1128.
29. Feldman S, Pearson TA: The Limulus test and gram-negative bacillary sepsis. Am J Dis Child 1974,
128(2):172-174.
30. Fink PC, Grunert JH: Endotoxemia in intensive care patients: a longitudinal study with the limulus
amebocyte lysate test. Klin Wochenschr 1984, 62(20):986-991.
31. Fossard DP, Kakkar VV, Elsey PA: Assessment of limulus test for detecting endotoxaemia. Br Med
J 1974, 2(5917):465-468.
32. Foulis AK, Murray WR, Galloway D, McCartney AC, Lang E, Veitch J, Whaley K: Endotoxaemia
and complement activation in acute pancreatitis in man. Gut 1982, 23(8):656-661.
33. Fugger R, Hamilton G, Rogy M, Herbst F, Kwasny W, Schemper M, Schultz F: Prognostic
significance of endotoxin determination in patients with severe intraabdominal infection. J Infect Dis
1990, 161(6):1314-1315.
34. Garcia Curiel A, Cantos de la Casa A, Martin Martin A, Aleman Aleman C: [Determination of
endotoxemia by means of the Limulus test in patients with septicemia (author's transl)]. Med Clin
(Barc) 1979, 72(9):375-377.
35. Gardlund B, Sjolin J, Nilsson A, Roll M, Wickerts CJ, Wretlind B: Plasma levels of cytokines in
primary septic shock in humans: correlation with disease severity. J Infect Dis 1995, 172(1):296-301.
36. Giamarellou-Bourboulis EJ, Perdios J, Lelekis M, Eoconomou E, Tsouroulas P, Giamarellou H:
Impact of cefuroxime administration on endotoxin (LPS) and tumour necrosis factor-alpha
(TNFalpha) blood levels in patients suffering from acute pyelonephritis: a preliminary report. Int J
Antimicrob Agents 1999, 11(2):115-119.
Chapter 8: References
-175-
37. Goldie AS, Fearon KC, Ross JA, Barclay GR, Jackson RE, Grant IS, Ramsay G, Blyth AS, Howie
JC: Natural cytokine antagonists and endogenous antiendotoxin core antibodies in sepsis syndrome.
The Sepsis Intervention Group. Jama 1995, 274(2):172-177.
38. Guidet B, Barakett V, Vassal T, Petit JC, Offenstadt G: Endotoxemia and bacteremia in patients with
sepsis syndrome in the intensive care unit. Chest 1994, 106(4):1194-1201.
39. Harris RI, Stone PC, Evans GR, Stuart J: Endotoxaemia as a cause of fever in immunosuppressed
patients. J Clin Pathol 1984, 37(4):467-470.
40. Harthug S, Bjorvatn B, Osterud B: Quantitation of endotoxin in blood from patients with
meningococcal disease using a limulus lysate test in combination with chromogenic substrate.
Infection 1983, 11(4):192-195.
41. Hass A, Rossberg MI, Hodes HL, Hyatt AC, Hodes DS: Endotoxin levels in immunocompromised
children with fever. J Pediatr 1986, 109(2):265-269.
42. Hynninen M, Valtonen M, Vaara M, Markkanen H, Kuusela P, Saxen H, Takkunen O: Plasma
endotoxin and cytokine levels in neutropenic and non-neutropenic bacteremic patients. Eur J Clin
Microbiol Infect Dis 1995, 14(12):1039-1045.
43. Jacob AI, Goldberg PK, Bloom N, Degenshein GA, Kozinn PJ: Endotoxin and bacteria in portal
blood. Gastroenterology 1977, 72(6):1268-1270.
44. Jirillo E, Pasquetto N, Marcuccio L, Monno R, De Rinaldis P, Fumarola D: Endotoxemia detected
by Limulus assay in severe malnourished children. Plasma effects on leucocyte migration:
preliminary investigations. G Batteriol Virol Immunol 1975, 68(7-12):174-178.
45. Kelley CJ, Ingoldby CJ, Blenkharn JI, Wood CB: Colonoscopy related endotoxemia. Surg Gynecol
Obstet 1985, 161(4):332-334.
46. Kelsey MC, Lipscomb AP, Mowles JM: Limulus amoebocyte lysate endotoxin test: an aid to the
diagnosis in the septic neonate? J Infect 1982, 4(1):69-72.
47. Ketchum PA, Parsonnet J, Stotts LS, et al. (1997) Utilization of a chromogenic Limulus amebocyte
lysate blood assay in a multi-center study of sepsis. J Endotoxin Res 4: 9-16
48. Bates DW, Parsonnet J, Ketchum PA, Miller EB, Novitsky TJ, Sands K, Hibberd PL, Graman PS,
Lanken PN, Schwartz JS et al: Limulus amebocyte lysate assay for detection of endotoxin in patients
with sepsis syndrome. AMCC Sepsis Project Working Group. Clin Infect Dis 1998, 27(3):582-591.
49. Kinoshita H, Hirohashi K, Igawa S, Nagata E, Sakai K: Cholangitis. World J Surg 1984, 8(6):963969.
50. Kiss A, Ferenci P, Graninger W, Pamperl H, Potzi R, Meryn S: Endotoxaemia following
colonoscopy. Endoscopy 1983, 15(1):24-26.
51. Klein K, Fuchs GJ, Kulapongs P, Mertz G, Suskind RM, Olson RE: Endotoxemia in protein-energy
malnutrition. J Pediatr Gastroenterol Nutr 1988, 7(2):225-228.
52. Lau JY, Chung SC, Leung JW, Ling TK, Yung MY, Li AK: Endoscopic drainage aborts
endotoxaemia in acute cholangitis. Br J Surg 1996, 83(2):181-184.
53. Levin J, Poore TE, Zauber NP, Oser RS: Detection of endotoxin in the blood of patients with sepsis
due to gran-negative bacteria. N Engl J Med 1970, 283(24):1313-1316.
54. Levin J, Poore TE, Young NS, Margolis S, Zauber NP, Townes AS, Bell WR: Gram-negative sepsis:
detection of endotoxemia with the limulus test. With studies of associated changes in blood
coagulation, serum lipids, and complement. Ann Intern Med 1972, 76(1):1-7.
55. Lumsden AB, Henderson JM, Alspaugh J: Endotoxemia during percutaneous manipulation of the
obstructed biliary tree. Am J Surg 1989, 158(1):21-24.
56. Magliulo E, Scevola D, Fumarola D, Vaccaro R, Bertotto A, Burberi S: Clinical experience in
detecting endotoxemia with the limulus test in typhoid fever and other Salmonella infections.
Infection 1976, 4(2):21-24.
57. Martinez LA, Quintiliani R, Tilton RC: Clinical experience on the detection of endotoxemia with the
limulus test. J Infect Dis 1973, 127(1):102-105.
-176-
Chapter 8: References
58. Massignon D, Lepape A, Debize G, Remillieux MF, De Pasquale V, Banssillon V, Coeur P:
Detection of gram-negative bacteraemia in early sepsis by a quantitative chromogenic and kinetic
endotoxin assay. The Study Group. Eur J Clin Invest 1996, 26(7):596-601.
59. Maury E, Barakett V, Blanchard H, Guitton C, Fitting C, Vassal T, Chauvin P, Guidet B, Offenstadt
G: Circulating endotoxin during initial antibiotic treatment of severe gram-negative bacteremic
infections. J Infect Dis 1998, 178(1):270-273.
60. McCartney AC, Banks JG, Clements GB, Sleigh JD, Tehrani M, Ledingham IM: Endotoxaemia in
septic shock: clinical and post mortem correlations. Intensive Care Med 1983, 9(3):117-122.
61. McCartney AC, Robertson MR, Piotrowicz BI, Lucie NP: Endotoxaemia, fever and clinical status in
immunosuppressed patients: a preliminary study. J Infect 1987, 15(3):201-206.
62. McGladdery SR, Larasati N, Silitonga N, Punjabi M, Lesmana S, Pulungsih, O‘Hanley P. 1993.
Acute inflammatory cytokine response in typhoid fever, abstr. 284. Abstracts of the 1993 IDSA
Annual meeting. Clin Infect Dis 1993 17: 578.
63. Ng KP, Bhanumathy M, Ong GSY, et al. (1995) Endotoxin tests in patients with sepsis. J Endotoxin
Res 2: 387-393
64. Oberle MW, Graham GG, Levin J: Detection of endotoxemia with the Limulus test: preliminary
studies in severely malnourished children. J Pediatr 1974, 85(4):570-573.
65. Opal SM, Scannon PJ, Vincent JL, White M, Carroll SF, Palardy JE, Parejo NA, Pribble JP, Lemke
JH: Relationship between plasma levels of lipopolysaccharide (LPS) and LPS-binding protein in
patients with severe sepsis and septic shock. J Infect Dis 1999, 180(5):1584-1589.
66. Pearson FC, Dubczak J, Weary M, Bruszer G, Donohue G: Detection of endotoxin in the plasma of
patients with gram-negative bacterial sepsis by the Limulus amoebocyte lysate assay. J Clin
Microbiol 1985, 21(6):865-868.
67. Prins JM, van Agtmael MA, Kuijper EJ, van Deventer SJ, Speelman P: Antibiotic-induced
endotoxin release in patients with gram-negative urosepsis: a double-blind study comparing
imipenem and ceftazidime. J Infect Dis 1995, 172(3):886-891.
68. Prins JM, Lauw FN, Derkx BH, Speelman P, Kuijper EJ, Dankert J, van Deventer SJ: Endotoxin
release and cytokine production in acute and chronic meningococcaemia. Clin Exp Immunol 1998,
114(2):215-219.
69. Prytz H, Holst-Christensen J, Korner B, Liehr H: Portal venous and systemic endotoxaemia in
patients without liver disease and systemic endotoxaemia in patients with cirrhosis. Scand J
Gastroenterol 1976, 11(8):857-863.
70. Scheifele DW, Melton P, Whitchelo V: Evaluation of the Limulus test for endotoxemia in neonates
with suspected sepsis. J Pediatr 1981, 98(6):899-903.
71. Scheifele DW, Olsen EM, Pendray MR: Endotoxinemia and thrombocytopenia during neonatal
necrotizing enterocolitis. Am J Clin Pathol 1985, 83(2):227-229.
72. Shenep JL, Flynn PM, Barrett FF, Stidham GL, Westenkirchner DF: Serial quantitation of
endotoxemia and bacteremia during therapy for gram-negative bacterial sepsis. J Infect Dis 1988,
157(3):565-568.
73. Simpson AJ, Opal SM, Angus BJ, Prins JM, Palardy JE, Parejo NA, Chaowagul W, White NJ:
Differential antibiotic-induced endotoxin release in severe melioidosis. J Infect Dis 2000,
181(3):1014-1019.
74. Strutz F, Heller G, Krasemann K, Krone B, Muller GA: Relationship of antibodies to endotoxin core
to mortality in medical patients with sepsis syndrome. Intensive Care Med 1999, 25(5):435-444.
75. Stumacher RJ, Kovnat MJ, McCabe WR: Limitations of the usefulness of the Limulus assay for
endotoxin. N Engl J Med 1973, 288(24):1261-1264.
76. Suda H, Moroi C, Inada K, Chida S, Fujiwara T, Yoshida M: Application of a new perchloric acid
treatment method to measure endotoxin by an endotoxin-specific chromogenic Limulus test in
neonatal septicemia. Acta Paediatr Jpn 1995, 37(5):579-581.
Chapter 8: References
-177-
77. Suyasa IG, Reka IG, Inada K, Suda H, Kojima M, Mushiaki K, Okamoto S, Yoshida M: Plasma
endotoxin in typhoid fever. Kobe J Med Sci 1995, 41(5):175-186.
78. Tanaka M, Matsumoto T, Kumazawa J, Ohsumi S: Endotoxemia and bacteremia in patients
following ultrasonic lithotripsy. Urol Res 1987, 15(1):45-47.
79. Tarao K, So K, Moroi T, Ikeuchi T, Suyama T: Detection of endotoxin in plasma and ascitic fluid of
patients with cirrhosis: its clinical significance. Gastroenterology 1977, 73(3):539-542.
80. Thomas LL, Sturk A, Buller HR, ten Cate JW, Spijker RE, ten Cate H: Comparative investigation of
a quantitative chromogenic endotoxin assay and blood cultures. Am J Clin Pathol 1984, 82(2):203206.
81. Togari H, Mikawa M, Iwanaga T, Matsumoto N, Kawase A, Hagisawa M, Ogino T, Goto R,
Watanabe I, Kito H et al: Endotoxin clearance by exchange blood transfusion in septic shock
neonates. Acta Paediatr Scand 1983, 72(1):87-91.
82. Trinchet JC, Hornstein M, Jupeau A, Beaugrand M, Hecht Y, Ferrier JP: [Do latent bacteremia and
endotoxinemia exist in cirrhotics? A study of 48 patients]. Gastroenterol Clin Biol 1983, 7(5):518522.
83. Uriu K, Osajima A, Hiroshige K, Watanabe H, Aibara K, Inada Y, Segawa K, Anai H, Takagi I, Ito
A et al: Endotoxin removal by direct hemoperfusion with an adsorbent column using polymyxin Bimmobilized fiber ameliorates systemic circulatory disturbance in patients with septic shock. Am J
Kidney Dis 2002, 39(5):937-947.
84. Usawattanakul W, Tharavanij S, Limsuwan A: Tachypleus lysate test for endotoxin in patients with
Gram negative bacterial infections. Southeast Asian J Trop Med Public Health 1979, 10(1):13-17.
85. Usawattanakul W, Tharavanij S, Warrell DA, Looareesuwan S, White NJ, Supavej S, Soikratoke S:
Factors contributing to the development of cerebral malaria. II. Endotoxin. Clin Exp Immunol 1985,
61(3):562-568.
86. van Deventer SJ, Buller HR, ten Cate JW, Sturk A, Pauw W: Endotoxaemia: an early predictor of
septicaemia in febrile patients. Lancet 1988, 1(8586):605-609.
87. van Deventer SJ, de Vries I, Statius van Eps LW, Pauw W, Buller HR, Sturk A, ten Cate JW:
Endotoxemia, bacteremia and urosepsis. Prog Clin Biol Res 1988, 272:213-224.
88. van Dissel JT, van Furth R, Compier BA, Feuth HD, Frolich M: Survival in selected patients with
gram-negative sepsis after adjunctive therapy with HA-1A. Lancet 1993, 341(8850):959-960.
89. van Langevelde P, Joop K, van Loon J, Frolich M, Groeneveld PH, Westendorp RG, van Dissel JT:
Endotoxin, cytokines, and procalcitonin in febrile patients admitted to the hospital: identification of
subjects at high risk of mortality. Clin Infect Dis 2000, 31(6):1343-1348.
90. van Wieringen PM, Monnens LA, Bakkeren JA: Hemolytic-uremic syndrome: absence of circulating
endotoxin. Pediatrics 1976, 58(4):561-563.
91. Venet C, Zeni F, Viallon A, Ross A, Pain P, Gery P, Page D, Vermesch R, Bertrand M, Rancon F et
al: Endotoxaemia in patients with severe sepsis or septic shock. Intensive Care Med 2000,
26(5):538-544.
92. Watzke H, Schwarz HP, Luger A, Stummvoll HK: [Clinical value of endotoxin determination in
infection. Comparison of the Limulus amebocyte lysate test with detection of bacterial pathogens].
Acta Med Austriaca 1987, 14(1):21-24.
93. Wig JD, Kochhar R, Ray JD, Krishna Rao DV, Gupta NM, Ganguly NK: Endotoxemia predicts
outcome in acute pancreatitis. J Clin Gastroenterol 1998, 26(2):121-124.
94. Wortel CH, von der Mohlen MA, van Deventer SJ, Sprung CL, Jastremski M, Lubbers MJ, Smith
CR, Allen IE, ten Cate JW: Effectiveness of a human monoclonal anti-endotoxin antibody (HA-1A)
in gram-negative sepsis: relationship to endotoxin and cytokine levels. J Infect Dis 1992,
166(6):1367-1374.
95. Yoshida M, Obayashi T, Tamura H, Tanaka S, Kawai T, Sakamoto S, Miura Y: Diagnostic and
prognostic significance of plasma endotoxin determination in febrile patients with haematological
malignancies. Eur J Cancer 1994, 30A(2):145-147.
-178-
Chapter 8: References
96. Beller FK, Douglas GWD. Thrombocytopenia indicating gram-negative infection and endotoxemia.
Obstet Gynec 1973; 4:521-524.
97. Douglas GW, Beller FK, Debrovner CH: The Demonstration of Endotoxin in the Circulating Blood
of Patients with Septic Abortion. Am J Obstet Gynecol 1963, 87:780-788.
98. McGill MW, Porter PJ, Kass EH: The use of a bioassay for endotoxin in clinical infections. J Infect
Dis 1970, 121(2):103-112.
99. Porter PJ, Spievack AR, Kass EH: Endotoxin-Like Activity of Serum from Patients with Severe
Localized Infections. N Engl J Med 1964, 271:445-447.
100. Dholakia S, Inwald D, Betts H, Nadel S: Endotoxemia in pediatric critical illness--a pilot study. Crit
Care, 15(3):R141.
101. Yaguchi A, Yuzawa J, Klein DJ, Takeda M, Harada T: Combining intermediate levels of the
Endotoxin Activity Assay (EAA) with other biomarkers in the assessment of patients with sepsis:
results of an observational study. Crit Care, 16(3):R88.
102. Casey LC, Balk RA, Bone RC: Plasma cytokine and endotoxin levels correlate with survival in
patients with the sepsis syndrome. Ann Intern Med 1993, 119(8):771-778.
103. McGill MW, Porter PJ, Vivaldi E, Kass EH: Use of a bioassay for endotoxin in clinical infections
and in experimental vasomotor collapse. Antimicrob Agents Chemother (Bethesda) 1967, 7:132-136.
104. Kollef MH, Eisenberg PR: A rapid qualitative assay to detect circulating endotoxin can predict the
development of multiorgan dysfunction. Chest 1997, 112(1):173-180.
105. Marshall JC, Walker PM, Foster DM, Harris D, Ribeiro M, Paice J, Romaschin AD, Derzko AN:
Measurement of endotoxin activity in critically ill patients using whole blood neutrophil dependent
chemiluminescence. Crit Care 2002, 6(4):342-348.
106. Marshall JC, Foster D, Vincent JL, Cook DJ, Cohen J, Dellinger RP, Opal S, Abraham E, Brett SJ,
Smith T et al: Diagnostic and prognostic implications of endotoxemia in critical illness: results of the
MEDIC study. J Infect Dis 2004, 190(3):527-534.
107. Valenza F, Fagnani L, Coppola S, Froio S, Sacconi F, Tedesco C, Maffioletti M, Pizzocri M, Salice
V, Ranzi ML et al: Prevalence of endotoxemia after surgery and its association with ICU length of
stay. Crit Care 2009, 13(3):R102.
108. Braude AI: Bacterial endotoxins. Sci Am 1964, 210(5):36-45.
109. Rietschel ET, Brade H: Bacterial endotoxins. Sci Am 1992, 267(2):54-61.
110. Raetz CR: Bacterial endotoxins: extraordinary lipids that activate eucaryotic signal transduction. J
Bacteriol 1993, 175(18):5745-5753.
111. Opal SM: The clinical relevance of endotoxin in human sepsis: a critical analysis. J Endotoxin Res
2002, 8(6):473-476.
112. Cohen J: The detection and interpretation of endotoxaemia. Intensive Care Med 2000, 26 Suppl
1:S51-56
113. Corriveau CC, Danner RL: Antiendotoxin therapies for septic shock. Infect Agents Dis 1993,
2(1):44-52.
114. Opal SM, Gluck T: Endotoxin as a drug target. Crit Care Med 2003, 31(1 Suppl):S57-64.
115. Corriveau CC, Danner RL: Endotoxin as a therapeutic target in septic shock. Infect Agents Dis 1993,
2(1):35-43
116. Morrison DC: Bacterial endotoxins and pathogenesis. Rev Infect Dis 1983, 5 Suppl 4:S733-747.
117. Morrison DC, Ulevitch RJ: The effects of bacterial endotoxins on host mediation systems. A review.
Am J Pathol 1978, 93(2):526-618.
118. Morrison DC, Ryan JL: Endotoxins and disease mechanisms. Annu Rev Med 1987, 38:417-432.
119. Pearson FC: A comparison of the pyrogenicity of environmental endotoxins and
lipopolysaccharides. Prog Clin Biol Res 1985, 189:251-265.
Chapter 8: References
-179-
120. Pearson FC, Weary ME, Sargent HE, Novitsky TJ, Lin H, Lindsay G, Berzofsky RN, Lane AL,
Wilson JD, Cooper JF et al: Comparison of several control standard endotoxins to the National
Reference Standard Endotoxin--an HIMA collaborative study. Appl Environ Microbiol 1985,
50(1):91-93
121. Wachtel RE, Tsuji K: Comparison of limulus amebocyte lysates and correlation with the United
States Pharmacopeial pyrogen test. Appl Environ Microbiol 1977, 33(6):1265-1269.
122. Suffredini AF, Hochstein HD, McMahon FG: Dose-related inflammatory effects of intravenous
endotoxin in humans: evaluation of a new clinical lot of Escherichia coli O:113 endotoxin. J Infect
Dis 1999, 179(5):1278-1282.
123. Danner RL, Natanson C, Elin RJ, Hosseini JM, Banks S, MacVittie TJ, Parrillo JE: Pseudomonas
aeruginosa compared with Escherichia coli produces less endotoxemia but more cardiovascular
dysfunction and mortality in a canine model of septic shock. Chest 1990, 98(6):1480-1487.
124. Hoffman WD, Pollack M, Banks SM, Koev LA, Solomon MA, Danner RL, Koles N, Guelde G,
Yatsiv I, Mouginis T et al: Distinct functional activities in canine septic shock of monoclonal
antibodies specific for the O polysaccharide and core regions of Escherichia coli lipopolysaccharide.
J Infect Dis 1994, 169(3):553-561.
125. Natanson C, Danner RL, Reilly JM, Doerfler ML, Hoffman WD, Akin GL, Hosseini JM, Banks SM,
Elin RJ, MacVittie TJ et al: Antibiotics versus cardiovascular support in a canine model of human
septic shock. Am J Physiol 1990, 259(5 Pt 2):H1440-1447.
126. Hochstein HD, Fitzgerald EA, McMahon FG, Vargas R: Properties of US standard endotoxin (EC5) in human male volunteers. J Endotoxin Res 1994, 1:52-56.
127. Elin RJ, Sandberg AL, Rosentreich DL: Comparison of the pyrogenicity, Limulus activity
mitogenicity and complement reactivity of several bacterial endotoxins and related compounds. J
Immunol 1976, 117(4):1238-1242.
128. Takada H, Kotani S, Tanaka S, Ogawa T, Takahashi I, Tsujimoto M, Komuro T, Shiba T, Kusumoto
S, Kusunose N et al: Structural requirements of lipid A species in activation of clotting enzymes
from the horseshoe crab, and the human complement cascade. Eur J Biochem 1988, 175(3):573-580.
129. Mueller M, Lindner B, Kusumoto S, Fukase K, Schromm AB, Seydel U: Aggregates are the
biologically active units of endotoxin. J Biol Chem 2004, 279(25):26307-26313.
130. Laude-Sharp M, Haeffner-Cavaillon N, Caroff M, Lantreibecq F, Pusineri C, Kazatchkine MD:
Dissociation between the interleukin 1-inducing capacity and Limulus reactivity of
lipopolysaccharides from gram-negative bacteria. Cytokine 1990, 2(4):253-258.
131. Gutsmann T, Howe J, Zahringer U, Garidel P, Schromm AB, Koch MH, Fujimoto Y, Fukase K,
Moriyon I, Martinez-de-Tejada G et al: Structural prerequisites for endotoxic activity in the Limulus
test as compared to cytokine production in mononuclear cells. Innate Immun, 16(1):39-47.
132. Dehus O, Hartung T, Hermann C: Endotoxin evaluation of eleven lipopolysaccharides by whole
blood assay does not always correlate with Limulus amebocyte lysate assay. J Endotoxin Res 2006,
12(3):171-180.
133. Devleeschouwer MJ, Cornil MF, Dony J: Studies on the sensitivity and specificity of the Limulus
amebocyte lysate test and rabbit pyrogen assays. Appl Environ Microbiol 1985, 50(6):1509-1511.
134. Brandenburg K, Howe J, Gutsman T, Garidel P: The expression of endotoxic activity in the
Limulus test as compared to cytokine production in immune cells. Curr Med Chem 2009,
16(21):2653-2660.
135. Dixon DR, Darveau RP: Lipopolysaccharide heterogeneity: innate host responses to bacterial
modification of lipid a structure. J Dent Res 2005, 84(7):584-595.
136. Amura CR, Silverstein R, Morrison DC: Mechanisms involved in the pathogenesis of sepsis are not
necessarily reflected by in vitro cell activation studies. Infect Immun 1998, 66(11):5372-5378.
137. Tanamoto K, Zahringer U, McKenzie GR, Galanos C, Rietschel ET, Luderitz O, Kusumoto S, Shiba
T: Biological activities of synthetic lipid A analogs: pyrogenicity, lethal toxicity, anticomplement
activity, and induction of gelation of Limulus amoebocyte lysate. Infect Immun 1984, 44(2):421-426.
-180-
Chapter 8: References
138. Luchi M, Morrison DC: Comparable endotoxic properties of lipopolysaccharides are manifest in
diverse clinical isolates of gram-negative bacteria. Infect Immun 2000, 68(4):1899-1904.
139. Kirikae T, Kirikae F, Saito S, Tominaga K, Tamura H, Uemura Y, Yokochi T, Nakano M:
Biological characterization of endotoxins released from antibiotic-treated Pseudomonas aeruginosa
and Escherichia coli. Antimicrob Agents Chemother 1998, 42(5):1015-1021.
140. Hitchcock PJ, Leive L, Makela PH, Rietschel ET, Strittmatter W, Morrison DC: Lipopolysaccharide
nomenclature--past, present, and future. J Bacteriol 1986, 166(3):699-705.
141. Kotani S, Takada H, Tsujimoto M, Ogawa T, Takahashi I, Ikeda T, Otsuka K, Shimauchi H, Kasai
N, Mashimo J et al: Synthetic lipid A with endotoxic and related biological activities comparable to
those of a natural lipid A from an Escherichia coli re-mutant. Infect Immun 1985, 49(1):225-237.
142. Trent MS, Stead CM, Tran AX, Hankins JV: Diversity of endotoxin and its impact on pathogenesis.
J Endotoxin Res 2006, 12(4):205-223.
143. Raetz CR, Reynolds CM, Trent MS, Bishop RE: Lipid A modification systems in gram-negative
bacteria. Annu Rev Biochem 2007, 76:295-329.
144. Michie HR, Manogue KR, Spriggs DR, Revhaug A, O'Dwyer S, Dinarello CA, Cerami A, Wolff SM,
Wilmore DW: Detection of circulating tumor necrosis factor after endotoxin administration. N Engl
J Med 1988, 318(23):1481-1486.
145. Tesh VL, Vukajlovich SW, Morrison DC: Endotoxin interactions with serum proteins relationship to
biological activity. Prog Clin Biol Res 1988, 272:47-62.
146. Beller FK, Debrovner CH, Douglas GW: Potentiation of the Lethal Effect of Endotoxin by
Heterologous Plasma. J Exp Med 1963, 118:245-256.
147. Andra J, Gutsmann T, Muller M, Schromm AB: Interactions between lipid A and serum proteins.
Adv Exp Med Biol 2009, 667:39-51.
148. Cross AS, et al., Pre-treatment with recombinant murine tumour necrosis factor and murine interleukin 1 protects mice from lethal infection. J Exp Med 169;2021 (1989)
149. Eden CS, Shahin R, Briles D: Host resistance to mucosal gram-negative infection. Susceptibility of
lipopolysaccharide nonresponder mice. J Immunol 1988, 140(9):3180-3185.
150. van der Poll T, Opal SM: Host-pathogen interactions in sepsis. Lancet Infect Dis 2008, 8(1):32-43.
151. Alexander C, Rietschel ET: Bacterial lipopolysaccharides and innate immunity. J Endotoxin Res
2001, 7(3):167-202.
152. Rietschel ET, Brade L, Brandenburg K, Flad HD, de Jong-Leuveninck J, Kawahara K, Lindner B,
Loppnow H, Luderitz T, Schade U et al: Chemical structure and biologic activity of bacterial and
synthetic lipid A. Rev Infect Dis 1987, 9 Suppl 5:S527-536.
153. Rietschel ET, Kirikae T, Schade FU, Mamat U, Schmidt G, Loppnow H, Ulmer AJ, Zahringer U,
Seydel U, Di Padova F et al: Bacterial endotoxin: molecular relationships of structure to activity and
function. FASEB J 1994, 8(2):217-225.
154. Kirikae T, Schade FU, Zahringer U, Kirikae F, Brade H, Kusumoto S, Kusama T, Rietschel ET: The
significance of the hydrophilic backbone and the hydrophobic fatty acid regions of lipid A for
macrophage binding and cytokine induction. FEMS Immunol Med Microbiol 1994, 8(1):13-26.
155. Suffredini AF, Fromm RE, Parker MM, Brenner M, Kovacs JA, Wesley RA, Parrillo JE: The
cardiovascular response of normal humans to the administration of endotoxin. N Engl J Med 1989,
321(5):280-287.
156. Mileski WJ, Winn RK, Harlan JM, Rice CL: Sensitivity to endotoxin in rabbits is increased after
hemorrhagic shock. J Appl Physiol 1992, 73(3):1146-1149.
157. Noshima S, Noda H, Herndon DN, Traber LD, Traber DL: Left ventricular performance during
continuous endotoxin-induced hyperdynamic endotoxemia in sheep. J Appl Physiol 1993,
74(4):1528-1533.
158. Bahrami S, Redl H, Leichtfried G, Yu Y, Schlag G: Similar cytokine but different coagulation
responses to lipopolysaccharide injection in D-galactosamine-sensitized versus nonsensitized rats.
Infect Immun 1994, 62(1):99-105.
Chapter 8: References
-181-
159. Holzer K, Thiel M, Moritz S, Kreimeier U, Messmer K: Expression of adhesion molecules on
circulating PMN during hyperdynamic endotoxemia. J Appl Physiol 1996, 81(1):341-348.
160. Levi M, ten Cate H, Bauer KA, van der Poll T, Edgington TS, Buller HR, van Deventer SJ, Hack
CE, ten Cate JW, Rosenberg RD: Inhibition of endotoxin-induced activation of coagulation and
fibrinolysis by pentoxifylline or by a monoclonal anti-tissue factor antibody in chimpanzees. J Clin
Invest 1994, 93(1):114-120.
161. Kneidinger R, Bahrami S, Redl H, Schlag G. Comparison of endothelial acitavion during endotoxic
and post-traumatic conditions by serum analysis of soluble E-selectin in non-human primates. J Lab
Clin Med 1996; 128:515-519.
162. Redl H, Schlag G, Bahrami S. Endotoxemia in primate models. In Endotoxin in Health and Disease,
Brade H, Opal S, Vogel S, Morrison D (editors), Marcel Dekker Limited. 1999; 795– 808.
163. van der Poll T, Levi M, van Deventer SJ, ten Cate H, Haagmans BL, Biemond BJ, Buller HR, Hack
CE, ten Cate JW: Differential effects of anti-tumor necrosis factor monoclonal antibodies on
systemic inflammatory responses in experimental endotoxemia in chimpanzees. Blood 1994,
83(2):446-451.
164. Takahashi I, Kotani S, Takada H, Shiba T, Kusumoto S: Structural requirements of endotoxic
lipopolysaccharides and bacterial cell walls in induction of interleukin-1. Blood Purif 1988,
6(3):188-206.
165. Erridge C, Bennett-Guerrero E, Poxton IR: Structure and function of lipopolysaccharides. Microbes
Infect 2002, 4(8):837-851.
166. Caroff M, Karibian D, Cavaillon JM, Haeffner-Cavaillon N: Structural and functional analyses of
bacterial lipopolysaccharides. Microbes Infect 2002, 4(9):915-926.
167. Mueller M, Lindner B, Dedrick R, Schromm AB, Seydel U: Endotoxin: physical requirements for
cell activation. J Endotoxin Res 2005, 11(5):299-303.
168. Shnyra A, Hultenby K, Lindberg AA: Role of the physical state of Salmonella lipopolysaccharide in
expression of biological and endotoxic properties. Infect Immun 1993, 61(12):5351-5360.
169. Kato N: Crystallization and electron microscopy of bacterial lipopolysaccharide. Micron 1993;
24:91-114.
170. Komuro T, Murai T, Kawasaki H: Effect of sonication on the dispersion state of lipopolysaccharide
and its pyrogenicity in rabbits. Chem Pharm Bull (Tokyo) 1987, 35(12):4946-4952.
171. Munford RS, Varley AW: Shield as signal: lipopolysaccharides and the evolution of immunity to
gram-negative bacteria. PLoS Pathog 2006, 2(6):e67.
172. Munford RS. Sensing Gram-Negative Bacterial Lipopolysaccharides: a Human Disease
Determinant? Infect. Immun. 2008 76 (2) 454-465
173. Dziarski R: Deadly plague versus mild-mannered TLR4. Nat Immunol 2006, 7(10):1017-1019.
174. Miller SI, Ernst RK, Bader MW: LPS, TLR4 and infectious disease diversity. Nat Rev Microbiol
2005, 3(1):36-46.
175. Park BS, Song DH, Kim HM, Choi BS, Lee H, Lee JO: The structural basis of lipopolysaccharide
recognition by the TLR4-MD-2 complex. Nature 2009, 458(7242):1191-1195.
176. Ulmer AJ, Feist W, Heine H, Kirikae T, Kirikae F, Kusumoto S, Kusama T, Brade H, Schade U,
Rietschel ET et al: Modulation of endotoxin-induced monokine release in human monocytes by lipid
A partial structures that inhibit binding of 125I-lipopolysaccharide. Infect Immun 1992, 60(12):51455152.
177. Van Amersfoort ES, Van Berkel TJ, Kuiper J: Receptors, mediators, and mechanisms involved in
bacterial sepsis and septic shock. Clin Microbiol Rev 2003, 16(3):379-414.
178. Beutler B, Rietschel ET: Innate immune sensing and its roots: the story of endotoxin. Nat Rev
Immunol 2003, 3(2):169-176.
179. Bryant CE, Spring DR, Gangloff M, Gay NJ: The molecular basis of the host response to
lipopolysaccharide. Nat Rev Microbiol 2010, 8(1):8-14.
-182-
Chapter 8: References
180. Cinel I, Opal SM: Molecular biology of inflammation and sepsis: a primer. Crit Care Med 2009,
37(1):291-304.
181. Staros EB: Innate immunity: New approaches to understanding its clinical significance. Am J Clin
Pathol 2005, 123(2):305-312.
182. Opal S M, Palardy J E, Marra M N, McKelligon B M, Scott R W, Fisher C J. (1994). Relative
concentrations of endotoxin-binding proteins in body fluids during infection. The Lancet, 344: 429431.
183. Hauser, A R., et al. Clinical significance of microbial infection and adaptation in cystic fibrosis.
Clinical microbiology reviews 24.1 (2011): 29-70.
184. West MA, Heagy W: Endotoxin tolerance: A review. Crit Care Med 2002, 30(1 Supp):S64-S73.
185. Greisman SE, Hornick RB: Mechanisms of endotoxin tolerance with special reference to man. J
Infect Dis 1973, 128:Suppl:265-276.
186. Cross AS: Endotoxin tolerance-current concepts in historical perspective. J Endotoxin Res 2002,
8(2):83-98.
187. Greisman SE, Carozza FA, Jr., Hills JD: Mechanisms of endotoxin tolerance. I. Relationship
between tolerance and reticuloendothelial system phagocytic activity in the rabbit. J Exp Med 1963,
117:663-674.
188. Greisman SE, DuBuy B: Mechanisms of endotoxin tolerance. IX. Effect of exchange transfusion.
Proc Soc Exp Biol Med 1975, 148(3):675-678.
189. Greisman SE, Hornick RB, Carozza FA, Jr., Woodward TE: Effect of endotoxin tolerance on human
typhoid fever. Trans Assoc Am Physicians 1962, 75:170-178.
190. Greisman SE, Hornick RB, Carozza FA, Jr., Woodward TE: The role of endotoxin during typhoid
fever and tularemia in man. I. Acquisition of tolerance to endotoxin. J Clin Invest 1963, 42:10641075.
191. Greisman SE, Hornick RB, Wagner HN, Jr., Woodward TE: The integrity of the endotoxin tolerance
mechanisms during typhoid fever in man. Trans Assoc Am Physicians 1967, 80:250-258.
192. Greisman SE, Hornick RB, Wagner HN, Jr., Woodward WE, Woodward TE: The role of endotoxin
during typhoid fever and tularemia in man. IV. The integrity of the endotoxin tolerance mechanisms
during infection. J Clin Invest 1969, 48(4):613-629.
193. Greisman SE, Wagner HN, Iio M, Hornick RB: Mechanisms of Endotoxin Tolerance. II.
Relationship between Endotoxin Tolerance and Reticuloendothelial System Phagocytic Activity in
Man. J Exp Med 1964, 119:241-264.
194. Greisman SE, Woodward CL: Mechanisms of endotoxin tolerance. VII. The role of the liver. J
Immunol 1970, 105(6):1468-1476.
195. Greisman SE, Woodward WE: Mechanisms of Endotoxin Tolerance. III. The Refractory State
during Continuous Intravenous Infusions of Endotoxin. J Exp Med 1965, 121:911-933.
196. Greisman SE, Young EJ, Carozza FA, Jr.: Mechanisms of endotoxin tolerance. V. Specificity of the
early and late phases of pyrogenic tolerance. J Immunol 1969, 103(6):1223-1236.
197. Greisman SE, Young EJ, DuBuy B: Mechanisms of endotoxin tolerance. VIII. Specificity of serum
transfer. J Immunol 1973, 111(5):1349-1360.
198. Greisman SE, Young EJ, Woodward WE: Mechanisms of endotoxin tolerance. IV. Specificity of the
pyrogenic refractory state during continuous intravenous infusions of endotoxin. J Exp Med 1966,
124(5):983-1000.
199. Biswas SK, Lopez-Collazo E: Endotoxin tolerance: new mechanisms, molecules and clinical
significance. Trends Immunol 2009, 30(10):475-487.
200. McCabe WR: Endotoxin Tolerance. II. Its Occurrence in Patients with Pyelonephritis. J Clin Invest
1963, 42(5):618-625.
201. Abernathy RS, Spink WW: Studies with Brucella endotoxin in humans: the significance of
susceptibility to endotoxin in the pathogenesis of brucellosis. J Clin Invest 1958, 37(2):219-231.
Chapter 8: References
-183-
202. Wiersinga WJ, van't Veer C, van den Pangaart PS, Dondorp AM, Day NP, Peacock SJ, van der Poll
T: Immunosuppression associated with interleukin-1R-associated-kinase-M upregulation predicts
mortality in Gram-negative sepsis (melioidosis). Crit Care Med 2009, 37(2):569-576.
203. Lehman JS, Jr., Bassily S: Endotoxin tolerance in patients with chronic bacteremia and bacteriuria
due to Salmonella. J Infect Dis 1971, 124(3):318-321.
204. Morgan HR: Resistance to the action of the endotoxins of enteric bacilli in man. J Clin Invest 1948,
27(6):706-709.
205. Adib-Conquy M, Adrie C, Moine P, Asehnoune K, Fitting C, Pinsky MR, Dhainaut JF, Cavaillon
JM: NF-kappaB expression in mononuclear cells of patients with sepsis resembles that observed in
lipopolysaccharide tolerance. Am J Respir Crit Care Med 2000, 162(5):1877-1883.
206. Buttenschoen K, Kornmann M, Berger D, Leder G, Beger HG, Vasilescu C: Endotoxemia and
endotoxin tolerance in patients with ARDS. Langenbecks Arch Surg 2008, 393(4):473-478.
207. Kox M, de Kleijn S, Pompe JC, Ramakers BP, Netea MG, van der Hoeven JG, Hoedemaekers CW,
Pickkers P: Differential ex vivo and in vivo endotoxin tolerance kinetics following human
endotoxemia. Crit Care Med, 39(8):1866-1870.
208. Hornick RB, Greisman SE, Woodward TE, DuPont HL, Dawkins AT, Snyder MJ: Typhoid fever:
pathogenesis and immunologic control. 2. N Engl J Med 1970, 283(14):739-746.
209. Hornick RB, Greisman SE, Woodward TE, DuPont HL, Dawkins AT, Snyder MJ: Typhoid fever:
pathogenesis and immunologic control. N Engl J Med 1970, 283(13):686-691.
210. Levin J, Bang FB: Clottable protein in Limulus; its localization and kinetics of its coagulation by
endotoxin. Thromb Diath Haemorrh 1968, 19(1):186-197.
211. Levin J, Bang FB: The Role of Endotoxin in the Extracellular Coagulation of Limulus Blood. Bull
Johns Hopkins Hosp 1964, 115:265-274.
212. Levin J, Tomasulo PA, Oser RS: Detection of endotoxin in human blood and demonstration of an
inhibitor. J Lab Clin Med 1970, 75(6):903-911.
213. Jorgensen JH, Smith RF: Rapid detection of contaminated intravenous fluids using the Limulus in
vitro endotoxin assay. Appl Microbiol 1973, 26(4):521-524.
214. Romaschin AD, Harris DM, Ribeiro MB, Paice J, Foster DM, Walker PM, Marshall JC: A rapid
assay of endotoxin in whole blood using autologous neutrophil dependent chemiluminescence. J
Immunol Methods 1998, 212(2):169-185.
215. Romaschin, A., Klein, D., & Marshall, J. (2012). Bench-to-bedside review: Clinical experience with
the endotoxin activity assay. Critical Care, 16(6), 248.
216. Rylatt D, Wilson K, Kemp BE, Elms MJ, Manickavasagam B, Shi W, Cox A, McArthur MJ, O'Hara
J, Corbett ME et al: A rapid test for endotoxin in whole blood. Prog Clin Biol Res 1995, 392:273284.
217. Rylatt DB, Kemp BE, Bundesen PG, John MA, O'Reilly EJ, Cottis LE, Miles SJ, Khan JM, Dinh DP,
Stapleton D et al: A rapid whole-blood immunoassay system. Med J Aust 1990, 152(2):75-77.
218. Ding, J L, Bow Ho. Endotoxin Detection–from Limulus Amebocyte Lysate to Recombinant Factor
C. Endotoxins: Structure, Function and Recognition (2010): 187-208.
219. Ding, J L., Bow Ho. A new era in pyrogen testing. TRENDS in Biotechnology 19.8 (2001): 277-281.
220. Roslansky PF, Dawson ME, Novitsky TJ: Plastics, endotoxins, and the Limulus amebocyte lysate
test. J Parenter Sci Technol 1991, 45(2):83-87.
221. Cooper JF, Levin J, Wagner HN, Jr.: Quantitative comparison of in vitro and in vivo methods for the
detection of endotoxin. J Lab Clin Med 1971, 78(1):138-148.
222. Takayama K, Qureshi N, Raetz CR, Ribi E, Peterson J, Cantrell JL, Pearson FC, Wiggins J, Johnson
AG: Influence of fine structure of lipid A on Limulus amebocyte lysate clotting and toxic activities.
Infect Immun 1984, 45(2):350-355.
-184-
Chapter 8: References
223. Hwang SH, Park DU, Joo SI, Park HH, Yoon CS: Comparison of endotoxin levels and gramnegative bacteria under different conditions in microbial laboratories and a biowaste site.
Chemosphere, 85(1):135-139.
224. Redl H, Bahrami S, Leichtfried G, Schlag G: Special collection and storage tubes for blood
endotoxin and cytokine measurements. Clin Chem 1992, 38(5):764-765.
225. Roslansky PF, Novitsky TJ: Sensitivity of Limulus amebocyte lysate (LAL) to LAL-reactive glucans.
J Clin Microbiol 1991, 29(11):2477-2483.
226. Albert O, Toubas D, Strady C, Cousson J, Delmas C, Vernet V, Villena I: Reactivity of (1-->3)-betad-glucan assay in bacterial bloodstream infections. Eur J Clin Microbiol Infect Dis, 30(11):14531460.
227. Ellis M, Al-Ramadi B, Finkelman M, Hedstrom U, Kristensen J, Ali-Zadeh H, Klingspor L:
Assessment of the clinical utility of serial beta-D-glucan concentrations in patients with persistent
neutropenic fever. J Med Microbiol 2008, 57(Pt 3):287-295.
228. Pickering JW, Sant HW, Bowles CA, Roberts WL, Woods GL: Evaluation of a (1->3)-beta-Dglucan assay for diagnosis of invasive fungal infections. J Clin Microbiol 2005, 43(12):5957-5962.
229. Mennink-Kersten MA, Ruegebrink D, Verweij PE: Pseudomonas aeruginosa as a cause of 1,3-betaD-glucan assay reactivity. Clin Infect Dis 2008, 46(12):1930-1931.
230. Al-Hasan MN, Eckel-Passow JE, Baddour LM: Bacteremia complicating gram-negative urinary tract
infections: a population-based study. J Infect, 60(4):278-285.
231. Elin RJ: Clinical Utility of the limulus test with blood, CSF and synovial fluid. Prog Clin Biol Res
1979, 29:279-292.
232. Elin RJ: Clinical Utility of the limulus ameocyte lysate (LAL) test. Prog Clin Biol Res 1985,
189:307-342
233. Berman NS, Siegel SE, Nachum R, Lipsey A, Leedom J: Cerebrospinal fluid endotoxin
concentrations in gram-negative bacterial meningitis. J Pediatr 1976, 88(4 Pt 1):553-556.
234. Dyson D, Cassady G: Use of Limulus lysate for detecting gram-negative neonatal meningitis.
Pediatrics 1976, 58(1):105-109.
235. Jorgensen JH, Smith RF: Rapid detection of contaminated intravenous fluids using the Limulus in
vitro endotoxin assay. Appl Microbiol 1973, 26(4):521-524.
236. Sarff LD, Platt LH, McCracken GH, Jr.: Cerebrospinal fluid evaluation in neonates: comparison of
high-risk infants with and without meningitis. J Pediatr 1976, 88(3):473-477.
237. Nachum R, Lipsey A, Siegel SE: Rapid detection of gram-negative bacterial meningitis by the
limulus lysate test. N Engl J Med 1973, 289(18):931-934.
238. Ross S, Rodriguez W, Controni G, Korengold G, Watson S, Khan W: Limulus lysate test for gramnegative bacterial meningitis. Bedside application. JAMA 1975, 233(13):1366-1369.
239. Tuazon CU, Perez AA, Elin RJ, Sheagren JN: Detection of endotoxin in cerebrospinal and joint
fluids by limulus assay. Arch Intern Med 1977, 137(1):55-56.
240. Clumeck N, Lauwers S, Butzler JP: Limulus test and meningitis. Br Med J 1977, 1(6063):777.
241. Jorgensen JH, Carvajal HF, Chipps BE, Smith RF: Rapid detection of gram-negative bacteriuria by
use of the Limulus endotoxin assay. Appl Microbiol 1973, 26(1):38-42.
242. Thomas LL, ten Cate H, Sturk A, Joop K, ten Cate JW: Quantitative chromogenic endotoxin
determination in cerebrospinal fluid. Clin Chim Acta 1983, 127(1):137-139.
243. Elin RJ, Knowles R, Barth WF, Wolff SM: Lack of specificity of the limulus lysate test in the
diagnosis of pyogenic arthritis. J Infect Dis 1978, 137(5):507-513.
244. Cesario T, Jason M, Andrews BS, Shanbron E, Juranas D, Katz L, Friou GJ: Limulus assay for
bacterial endotoxin in synovial fluid. Ann Rheum Dis 1983, 42(5):571-574.
245. Nachum R, Shanbrom E: Rapid detection of Gram-negative bacteriuria by Limulus amoebocyte
lysate assay. J Clin Microbiol 1981, 13(1):158-162.
Chapter 8: References
-185-
246. Jorgensen JH, Alexander GA: Rapid detection of significant bacteriuria by use of an automated
Limulus amoebocyte lysate assay. J Clin Microbiol 1982, 16(3):587-589.
247. DuBose DA, Lemaire M, Basamania K, Rowlands J: Comparison of plasma extraction techniques in
preparation of samples for endotoxin testing by the Limulus amoebocyte lysate test. J Clin Microbiol
1980, 11(1):68-72.
248. Hurley JC, Tosolini FA, Louis WJ: Quantitative Limulus lysate assay for endotoxin and the effect of
plasma. J Clin Pathol 1991, 44(10):849-854.
249. Novitsky TJ: Limitations of the Limulus amebocyte lysate test in demonstrating circulating
lipopolysaccharides. Ann N Y Acad Sci 1998, 851:416-421.
250. Warren HS, Novitsky TJ, Ketchum PA, Roslansky PF, Kania S, Siber GR: Neutralization of
bacterial lipopolysaccharides by human plasma. J Clin Microbiol 1985, 22(4):590-595.
251. Obayashi T, Tamura H, Tanaka S, Ohki M, Takahashi S, Arai M, Masuda M, Kawai T: A new
chromogenic endotoxin-specific assay using recombined limulus coagulation enzymes and its
clinical applications. Clin Chim Acta 1985, 149(1):55-65.
252. Lindsay GK, Roslansky PF, Novitsky TJ: Single-step, chromogenic Limulus amebocyte lysate assay
for endotoxin. J Clin Microbiol 1989, 27(5):947-951.
253. Riedemann NC, Guo RF, Ward PA: The enigma of sepsis. J Clin Invest 2003, 112(4):460-467.
254. Carlet J, Cohen J, Calandra T, Opal SM, Masur H: Sepsis: time to reconsider the concept. Crit Care
Med 2008, 36(3):964-966.
255. Vincent J-L. Dear SIRS, I'm sorry to say that I don't like you. Crit Care Med 25.2 (1997): 372-374.
256. Vincent JL, Martinez EO, Silva E: Evolving concepts in sepsis definitions. Crit Care Clin 2009,
25(4):665-675.
257. Thomas L. Germs. N Engl J Med 1972, 287:553-555.
258. Opal SM, Cohen J: Clinical gram-positive sepsis: does it fundamentally differ from gram-negative
bacterial sepsis? Crit Care Med 1999, 27(8):1608-1616.44.
259. Horn DL, Morrison DC, Opal SM, Silverstein R, Visvanathan K, Zabriskie JB: What are the
microbial components implicated in the pathogenesis of sepsis? Report on a symposium. Clin Infect
Dis 2000, 31(4):851-858
260. James SK, Armstrong P, Barnathan E, Califf R, Lindahl B, Siegbahn A, Simoons ML, Topol EJ,
Venge P, Wallentin L, GUSTO-IV-ACS Investigators. Troponin and C-reactive protein have
different relations to subsequent mortality and myocardial infarction after acute coronary syndrome:
GUSTO-IV substudy. J Am Coll Cardiol. 2003;41(6):916.
261. Mallett S, Halligan S, Thompson M, Collins GS, Altman DG: Interpreting diagnostic accuracy
studies for patient care. BMJ, 345:e3999.
262. Kramer AA, Zimmerman JE: Assessing the calibration of mortality benchmarks in critical care: The
Hosmer-Lemeshow test revisited. Crit Care Med 2007, 35(9):2052-2056.
263. Barriere SL, Lowry SF: An overview of mortality risk prediction in sepsis. Crit Care Med 1995,
23(2):376-393.
264. Cook NR: Statistical evaluation of prognostic versus diagnostic models: beyond the ROC curve. Clin
Chem 2008, 54(1):17-23.
265. Bone RC, Balk RA, Cerra FB, Dellinger RP, Fein AM, Knaus WA, Schein RM, Sibbald WJ:
Definitions for sepsis and organ failure and guidelines for the use of innovative therapies in sepsis.
The ACCP/SCCM Consensus Conference Committee. American College of Chest
Physicians/Society of Critical Care Medicine. Chest 1992, 101(6):1644-1655.
266. Bone RC: Sepsis, sepsis syndrome, and the systemic inflammatory response syndrome (SIRS).
Gulliver in Laputa. JAMA 1995, 273(2):155-156.
267. Bone RC, Fisher CJ, Jr., Clemmer TP, Slotman GJ, Metz CA, Balk RA: Sepsis syndrome: a valid
clinical entity. Methylprednisolone Severe Sepsis Study Group. Crit Care Med 1989, 17(5):389-393.
-186-
Chapter 8: References
268. Marshall JC, Cook DJ, Christou NV, Bernard GR, Sprung CL, Sibbald WJ: Multiple organ
dysfunction score: a reliable descriptor of a complex clinical outcome. Crit Care Med 1995,
23(10):1638-1652.
269. Beal AL, Cerra FB: Multiple organ failure syndrome in the 1990s. Systemic inflammatory response
and organ dysfunction. JAMA 1994, 271(3):226-233.
270. Annane D, Aegerter P, Jars-Guincestre MC, Guidet B: Current epidemiology of septic shock: the
CUB-Rea Network. Am J Respir Crit Care Med 2003, 168(2):165-172.
271. Bone RC: Toward an epidemiology and natural history of SIRS (systemic inflammatory response
syndrome). JAMA 1992, 268(24):3452-3455.
272. Brun-Buisson C, Doyon F, Carlet J, Dellamonica P, Gouin F, Lepoutre A, Mercier JC, Offenstadt G,
Regnier B: Incidence, risk factors, and outcome of severe sepsis and septic shock in adults. A
multicenter prospective study in intensive care units. French ICU Group for Severe Sepsis. JAMA
1995, 274(12):968-974.
273. Rangel-Frausto MS, Pittet D, Costigan M, Hwang T, Davis CS, Wenzel RP: The natural history of
the systemic inflammatory response syndrome (SIRS). A prospective study. JAMA 1995,
273(2):117-123.
274. Sands KE, Bates DW, Lanken PN, Graman PS, Hibberd PL, Kahn KL, Parsonnet J, Panzer R, Orav
EJ, Snydman DR et al: Epidemiology of sepsis syndrome in 8 academic medical centers. JAMA
1997, 278(3):234-240.
275. Bates DW, Lee TH: Projected impact of monoclonal anti-endotoxin antibody therapy. Arch Intern
Med 1994, 154(11):1241-1249.
276. Bates DW, Cook EF, Goldman L, Lee TH: Predicting bacteremia in hospitalized patients. A
prospectively validated model. Ann Intern Med 1990, 113(7):495-500.
277. Mozes B, Milatiner D, Block C, Blumstein Z, Halkin H: Inconsistency of a model aimed at
predicting bacteremia in hospitalized patients. J Clin Epidemiol 1993, 46(9):1035-1040.
278. Bates DW, Sands K, Miller E, Lanken PN, Hibberd PL, Graman PS, Schwartz JS, Kahn K,
Snydman DR, Parsonnet J et al: Predicting bacteremia in patients with sepsis syndrome. Academic
Medical Center Consortium Sepsis Project Working Group. J Infect Dis 1997, 176(6):1538-1551.
279. Peduzzi P, Shatney C, Sheagren J, Sprung C: Predictors of bacteremia and gram-negative
bacteremia in patients with sepsis. The Veterans Affairs Systemic Sepsis Cooperative Study Group.
Arch Intern Med 1992, 152(3):529-535.
280. Schechner V, Nobre V, Kaye KS, Leshno M, Giladi M, Rohner P, Harbarth S, Anderson DJ,
Karchmer AW, Schwaber MJ et al: Gram-negative bacteremia upon hospital admission: when
should Pseudomonas aeruginosa be suspected? Clin Infect Dis 2009, 48(5):580-586.
281. Sprung CL, Peduzzi PN, Shatney CH, Schein RM, Wilson MF, Sheagren JN, Hinshaw LB: Impact
of encephalopathy on mortality in the sepsis syndrome. The Veterans Administration Systemic
Sepsis Cooperative Study Group. Crit Care Med 1990, 18(8):801-806.
282. Knaus WA, Harrell FE, Jr., Lynn J, Goldman L, Phillips RS, Connors AF, Jr., Dawson NV,
Fulkerson WJ, Jr., Califf RM, Desbiens N et al: The SUPPORT prognostic model. Objective
estimates of survival for seriously ill hospitalized adults. Study to understand prognoses and
preferences for outcomes and risks of treatments. Ann Intern Med 1995, 122(3):191-203.
283. Bossink AW, Groeneveld J, Hack CE, Thijs LG: Prediction of mortality in febrile medical patients:
How useful are systemic inflammatory response syndrome and sepsis criteria? Chest 1998,
113(6):1533-1541.
284. Rhee JY, Kwon KT, Ki HK, Shin SY, Jung DS, Chung DR, Ha BC, Peck KR, Song JH: Scoring
systems for prediction of mortality in patients with intensive care unit-acquired sepsis: a comparison
of the Pitt bacteremia score and the Acute Physiology and Chronic Health Evaluation II scoring
systems. Shock 2009, 31(2):146-150.
285. Ghanem-Zoubi NO, Vardi M, Laor A, Weber G, Bitterman H: Assessment of disease-severity
scoring systems for patients with sepsis in general internal medicine departments. Crit Care,
15(2):R95.
Chapter 8: References
-187-
286. Vincent JL, Opal S, Torres A, Bonten M, Cohen J, Wunderink R: The PIRO concept: I is for
infection. Crit Care 2003, 7(3):252-255.
287. Cohen J, Cristofaro P, Carlet J, Opal S: New method of classifying infections in critically ill patients.
Crit Care Med 2004, 32(7):1510-1526.
288. Opal SM: Concept of PIRO as a new conceptual framework to understand sepsis. Pediatr Crit Care
Med 2005, 6(3 Suppl):S55-60.
289. Howell MD, Talmor D, Schuetz P, Hunziker S, Jones AE, Shapiro NI: Proof of principle: the
predisposition, infection, response, organ failure sepsis staging system. Crit Care Med, 39(2):322327.
290. Moreno RP, Metnitz B, Adler L, Hoechtl A, Bauer P, Metnitz PG: Sepsis mortality prediction based
on predisposition, infection and response. Intensive Care Med 2008, 34(3):496-504.
291. Rubulotta F, Marshall JC, Ramsay G, Nelson D, Levy M, Williams M: Predisposition,
insult/infection, response, and organ dysfunction: A new model for staging severe sepsis. Crit Care
Med 2009, 37(4):1329-1335.
292. Bates DW, Pruess KE, Lee TH: How bad are bacteremia and sepsis? Outcomes in a cohort with
suspected bacteremia. Arch Intern Med 1995, 155(6):593-598.
293. Abraham E, Laterre PF, Garg R, Levy H, Talwar D, Trzaskoma BL, Francois B, Guy JS, Bruckmann
M, Rea-Neto A et al: Drotrecogin alfa (activated) for adults with severe sepsis and a low risk of
death. N Engl J Med 2005, 353(13):1332-1341.
294. Kalil AC, LaRosa SP, Gogate J, Lynn M, Opal SM: Influence of severity of illness on the effects of
eritoran tetrasodium (E5564) and on other therapies for severe sepsis. Shock, 36(4):327-331.
295. Barochia AV, Cui X, Natanson C, Eichacker PQ: Risk of death and the efficacy of eritoran
tetrasodium (E5564): design considerations for clinical trials of anti-inflammatory agents in sepsis.
Crit Care Med, 38(1):306-308.
296. Kumar A, Safdar N, Kethireddy S, Chateau D: A survival benefit of combination antibiotic therapy
for serious infections associated with sepsis and septic shock is contingent only on the risk of death:
a meta-analytic/meta-regression study. Crit Care Med, 38(8):1651-1664.
297. Vincent JL, Opal SM, Marshall JC: Ten reasons why we should NOT use severity scores as entry
criteria for clinical trials or in our treatment decisions. Crit Care Med, 38(1):283-287.
298. Eichacker PQ, Parent C, Kalil A, Esposito C, Cui X, Banks SM, Gerstenberger EP, Fitz Y, Danner
RL, Natanson C: Risk and the efficacy of antiinflammatory agents: retrospective and confirmatory
studies of sepsis. Am J Respir Crit Care Med 2002, 166(9):1197-1205.
299. Macias WL, Nelson DR, Williams M, Garg R, Janes J, Sashegyi A: Lack of evidence for qualitative
treatment by disease severity interactions in clinical studies of severe sepsis. Crit Care 2005,
9(6):R607-622.
300. Ranieri VM, Thompson BT, Barie PS, Dhainaut JF, Douglas IS, Finfer S, Gardlund B, Marshall JC,
Rhodes A, Artigas A et al: Drotrecogin alfa (activated) in adults with septic shock. N Engl J Med,
366(22):2055-2064.
301. Kim PW, Perl TM, Keelaghan EF, Langenberg P, Perencevich EN, Harris AD, Song X, Roghmann
MC: Risk of mortality with blood stream infection is higher in the less severely ill at admission. Am
J Respir Crit Care Med 2005, 171:616-620.
302. Teres D, Lemeshow S: Why severity models should be used with caution. Crit Care Clin 1994,
10(1):93-110.
303. Pierrakos C, Vincent JL: Sepsis biomarkers: a review. Crit Care, 14(1):R15.
304. Kyzas PA, Denaxa-Kyza D, Ioannidis JP: Almost all articles on cancer prognostic markers report
statistically significant results. Eur J Cancer 2007, 43(17):2559-2579.
305. Pfeiffer CD, Fine JP, Safdar N: Diagnosis of invasive aspergillosis using a galactomannan assay: a
meta-analysis. Clin Infect Dis 2006, 42(10):1417-1427.
306. Tang BM, Eslick GD, Craig JC, McLean AS: Accuracy of procalcitonin for sepsis diagnosis in
critically ill patients: systematic review and meta-analysis. Lancet Infect Dis 2007, 7(3):210-217.
-188-
Chapter 8: References
307. Jones AE, Fiechtl JF, Brown MD, Ballew JJ, Kline JA: Procalcitonin test in the diagnosis of
bacteremia: a meta-analysis. Ann Emerg Med 2007, 50(1):34-41.
308. Karageorgopoulos DE, Vouloumanou EK, Ntziora F, Michalopoulos A, Rafailidis PI, Falagas ME:
beta-D-glucan assay for the diagnosis of invasive fungal infections: a meta-analysis. Clin Infect Dis,
52(6):750-770.
309. Avni T, Leibovici L, Paul M: PCR diagnosis of invasive candidiasis: systematic review and metaanalysis. J Clin Microbiol, 49(2):665-670.
310. Tacconelli E, Cataldo MA: Identifying risk factors for infections: the role of meta-analyses. Infect
Dis Clin North Am 2009, 23(2):211-224.
311. Pepe MS, Janes H, Longton G, Leisenring W, Newcomb P: Limitations of the odds ratio in gauging
the performance of a diagnostic, prognostic, or screening marker. Am J Epidemiol 2004, 159(9):882890.
312. Glas AS, Lijmer JG, Prins MH, Bonsel GJ, Bossuyt PM: The diagnostic odds ratio: a single indicator
of test performance. J Clin Epidemiol 2003, 56(11):1129-1135.
313. Davidoff F: Heterogeneity is not always noise: lessons from improvement. JAMA 2009,
302(23):2580-2586.
314. Fletcher J: What is heterogeneity and is it important? BMJ 2007, 334(7584):94-96.
315. Kibe S, Adams K, Barlow G: Diagnostic and prognostic biomarkers of sepsis in critical care. J
Antimicrob Chemother, 66 Suppl 2:ii33-40.
316. Bouadma, L., Luyt, C. E., Tubach, F., Cracco, C., Alvarez, A., Schwebel, C., ... & Wolff, M. (2010).
Use of procalcitonin to reduce patients' exposure to antibiotics in intensive care units (PRORATA
trial): a multicentre randomised controlled trial. The Lancet, 375(9713), 463-474.
317. Murray, P R., Masur H. Current approaches to the diagnosis of bacterial and fungal bloodstream
infections in the intensive care unit. Critical Care Medicine 40.12 (2012): 3277-3282.
318. Bauer, M., & Reinhart, K. (2010). Molecular diagnostics of sepsis—Where are we today?.
International Journal of Medical Microbiology, 300(6), 411-413.
319. Shah HN, Gharbia SE. (ed.). 2011. Mass spectrometry for microbial proteomics. Wiley, Chichester,
United Kingdom.
320. Wong HR: Clinical review: Sepsis and septic shock – the potential of gene arrays. Critical Care
2012, 16:204.
321. Reinhart K, Bauer M, Riedemann N C, & Hartog CS. New Approaches to Sepsis: Molecular
Diagnostics and Biomarkers. Clin Microbiol Rev, 2012, 25(4), 609-634.
322. Lvovschi V, Arnaud L, Parizot C, Freund Y, Juillien G, et al. (2011) Cytokine Profiles in Sepsis
Have Limited Relevance for Stratifying Patients in the Emergency Department: A Prospective
Observational Study. PLoS ONE 6(12): e28870. doi:10.1371/journal.pone.0028870.
323. Ioannidis, JPA. Expectations, validity, and reality in omics. J Clin Epidemiol 63(9) (2010): 945-949.
324. Dwivedi DJ, Toltl LJ, Swystun LL, Pogue J, Liaw KL, Weitz JI, Cook DJ, Fox-Robichaud AE, Liaw
PC, Group CC: Prognostic utility and characterization of cell-free DNA in patients with severe
sepsis. Crit Care, 16(4):R151.
325. Tang et al.: Genome-wide transcription profiling of human sepsis: a systematic review. Critical Care
2010 14:R237.
326. Afshari A, Schrenzel J, Ieven M, Harbarth S: Bench-to-bedside review: Rapid molecular diagnostics
for bloodstream infection - a new frontier? Crit Care, 16(3):222.
327. Coburn B, Morris AM, Tomlinson G, Detsky AS: Does this adult patient with suspected bacteremia
require blood cultures? JAMA, 308(5):502-511.
328. Brass JM: The cell envelope of gram-negative bacteria: new aspects of its function in transport and
chemotaxis. Curr Top Microbiol Immunol 1986, 129:1-92.
329. Lugtenberg B, Van Alphen L: Molecular architecture and functioning of the outer membrane of
Escherichia coli and other gram-negative bacteria. Biochim Biophys Acta 1983, 737(1):51-115.
Chapter 8: References
-189-
330. McCabe WR, Jackson GG (1962) Gram negative bacteremia. II Clinical laboratory and therapeutic
observations. Arch Intern Med 110:856-864.
331. Yagupsky P, Nolte FS: Quantitative aspects of septicemia. Clin Microbiol Rev 1990, 3(3):269-279.
332. Kellogg JA, Bankert DA, Manzella JP, Parsey KS, Scott SL, Cavanaugh SH: Occurrence and
documentation of low-level bacteremia in a community hospital's patient population. Am J Clin
Pathol 1995, 104(5):524-529.
333. Sleigh J, Cursons R, La Pine M: Detection of bacteraemia in critically ill patients using 16S rDNA
polymerase chain reaction and DNA sequencing. Intensive Care Med 2001, 27(8):1269-1273.
334. Ovstebo R, Brandtzaeg P, Brusletto B, Haug KB, Lande K, Hoiby EA, Kierulf P: Use of robotized
DNA isolation and real-time PCR to quantify and identify close correlation between levels of
Neisseria meningitidis DNA and lipopolysaccharides in plasma and cerebrospinal fluid from patients
with systemic meningococcal disease. J Clin Microbiol 2004, 42(7):2980-2987.
335. Corrigan JJ, Jr., Kiernat JF: Effect of heparin in experimental gram-negative septicemia. J Infect Dis
1975, 131(2):138-143.
336. Shenep JL, Barton RP, Mogan KA: Role of antibiotic class in the rate of liberation of endotoxin
during therapy for experimental gram-negative bacterial sepsis. J Infect Dis 1985, 151(6):1012-1018.
337. Quezado ZM, Natanson C, Alling DW, Banks SM, Koev CA, Elin RJ, Hosseini JM, Bacher JD,
Danner RL, Hoffman WD: A controlled trial of HA-1A in a canine model of gram-negative septic
shock. JAMA 1993, 269(17):2221-2227.
338. Myerowitz, R L., Medeiros AA, O'Brien TF. Recent experience with bacillemia due to gramnegative organisms. J Infect Dis 124.3 (1971): 239-246.
339. Setia U, Gross PA: Bacteremia in a community hospital: spectrum and mortality. Arch Intern Med
1977, 137(12):1698-1701.
340. Kreger BE, Craven DE, Carling PC, McCabe WR: Gram-negative bacteremia. III. Reassessment of
etiology, epidemiology and ecology in 612 patients. Am J Med 1980, 68(3):332-343.
341. Kreger BE, Craven DE, McCabe WR: Gram-negative bacteremia. IV. Re-evaluation of clinical
features and treatment in 612 patients. Am J Med 1980, 68(3):344-355.
342. Bryan CS, Reynolds KL, Brenner ER: Analysis of 1,186 episodes of gram-negative bacteremia in
non-university hospitals: the effects of antimicrobial therapy. Rev Infect Dis 1983, 5(4):629-638.
343. Roberts FJ, Geere IW, Coldman A: A three-year study of positive blood cultures, with emphasis on
prognosis. Rev Infect Dis 1991, 13(1):34-46.
344. Geerdes HF, Ziegler D, Lode H, Hund M, Loehr A, Fangmann W, Wagner J: Septicemia in 980
patients at a university hospital in Berlin: prospective studies during 4 selected years between 1979
and 1989. Clin Infect Dis 1992, 15(6):991-1002.
345. Weinstein MP, Towns ML, Quartey SM, Mirrett S, Reimer LG, Parmigiani G, Reller LB: The
clinical significance of positive blood cultures in the 1990s: a prospective comprehensive evaluation
of the microbiology, epidemiology, and outcome of bacteremia and fungemia in adults. Clin Infect
Dis 1997, 24(4):584-602.
346. Leibovici L, Shraga I, Drucker M, Konigsberger H, Samra Z, Pitlik SD: The benefit of appropriate
empirical antibiotic treatment in patients with bloodstream infection. J Intern Med 1998, 244(5):379386.
347. Garrouste-Orgeas M, Chevret S, Mainardi JL, Timsit JF, Misset B, Carlet J: A one-year prospective
study of nosocomial bacteraemia in ICU and non-ICU patients and its impact on patient outcome. J
Hosp Infect 2000, 44(3):206-213.
348. Retamar P, Portillo MM, Lopez-Prieto MD, Rodriguez-Lopez F, de Cueto M, Garcia MV, Gomez
MJ, Del Arco A, Munoz A, Sanchez-Porto A et al: Impact of inadequate empirical therapy on the
mortality of patients with bloodstream infections: a propensity score-based analysis. Antimicrob
Agents Chemother, 56(1):472-478.
-190-
Chapter 8: References
349. Ziegler EJ, McCutchan JA, Fierer J, Glauser MP, Sadoff JC, Douglas H, Braude AI: Treatment of
gram-negative bacteremia and shock with human antiserum to a mutant Escherichia coli. N Engl J
Med 1982, 307(20):1225-1230.
350. Ziegler EJ, Fisher CJ, Jr., Sprung CL, Straube RC, Sadoff JC, Foulke GE, Wortel CH, Fink MP,
Dellinger RP, Teng NN et al: Treatment of gram-negative bacteremia and septic shock with HA-1A
human monoclonal antibody against endotoxin. A randomized, double-blind, placebo-controlled trial.
The HA-1A Sepsis Study Group. N Engl J Med 1991, 324(7):429-436.
351. Greenman RL, Schein RM, Martin MA, Wenzel RP, MacIntyre NR, Emmanuel G, Chmel H, Kohler
RB, McCarthy M, Plouffe J et al: A controlled clinical trial of E5 murine monoclonal IgM antibody
to endotoxin in the treatment of gram-negative sepsis. The XOMA Sepsis Study Group. JAMA 1991,
266(8):1097-1102.
352. Singer C, Kaplan MH, Armstrong D: Bacteremia and fungemia complicating neoplastic disease. A
study of 364 cases. Am J Med 1977, 62(5):731-742.
353. Gonzalez-Barca E, Fernandez-Sevilla A, Carratala J, Granena A, Gudiol F. Prospective study of 288
episodes of bacteremia in neutropenic cancer patients in a single institution. Eur J Clin Microbiol
Infect Dis 1999; 15(4), 291-296.
354. Ehni WF, Reller LB, Ellison RT, 3rd: Bacteremia in granulocytopenic patients in a tertiary-care
general hospital. Rev Infect Dis 1991, 13(4):613-619.
355. Elting LS, Rubenstein EB, Rolston KV, Bodey GP: Outcomes of bacteremia in patients with cancer
and neutropenia: observations from two decades of epidemiological and clinical trials. Clin Infect
Dis 1997, 25(2):247-259.
356. Gonzalez-Barca E, Fernandez-Sevilla A, Carratala J, Salar A, Peris J, Granena A, Gudiol F:
Prognostic factors influencing mortality in cancer patients with neutropenia and bacteremia. Eur J
Clin Microbiol Infect Dis 1999, 18(8):539-544.
357. Klastersky J, Ameye L, Maertens J, Georgala A, Muanza F, Aoun M, Ferrant A, Rapoport B,
Rolston K, Paesmans M: Bacteraemia in febrile neutropenic cancer patients. Int J Antimicrob Agents
2007, 30 Suppl 1:S51-59.
358. Bryan CS, Reynolds KL: Bacteremic nosocomial pneumonia. Analysis of 172 episodes from a single
metropolitan area. Am Rev Respir Dis 1984, 129(5):668-671.
359. Rello J, Ricart M, Mirelis B, Quintana E, Gurgui M, Net A, Prats G: Nosocomial bacteremia in a
medical-surgical intensive care unit: epidemiologic characteristics and factors influencing mortality
in 111 episodes. Intensive Care Med 1994, 20(2):94-98.
360. Valles J, Leon C, Alvarez-Lerma F: Nosocomial bacteremia in critically ill patients: a multicenter
study evaluating epidemiology and prognosis. Spanish Collaborative Group for Infections in
Intensive Care Units of Sociedad Espanola de Medicina Intensiva y Unidades Coronarias (SEMIUC).
Clin Infect Dis 1997, 24(3):387-395.
361. Munford RS: Severe sepsis and septic shock: the role of gram-negative bacteremia. Annu Rev Pathol
2006, 1:467-496.
362. Joo EJ, Kang CI, Ha YE, Kim J, Kang SJ, Park SY, Lee NY, Wi YM, Chung DR, Peck KR et al:
Clinical predictors of Pseudomonas aeruginosa bacteremia among Gram-negative bacterial
infections in non-neutropenic patients with solid tumor. J Infect, 63(3):207-214.
363. Prowle JR, Echeverri JE, Ligabo EV, Sherry N, Taori GC, Crozier TM, Hart GK, Korman TM,
Mayall BC, Johnson PD et al: Acquired bloodstream infection in the intensive care unit: incidence
and attributable mortality. Crit Care, 15(2):R100.
364. Cosgrove SE, Carmeli Y: Studies of bloodstream infection outcomes: reading between the lines.
Infect Control Hosp Epidemiol 2003, 24(12):884-886.
365. Blot S, De Bacquer D, Hoste E, Depuydt P, Vandewoude K, De Waele J, Benoit D, De Schuijmer J,
Colardyn F, Vogelaers D: Influence of matching for exposure time on estimates of attributable
mortality caused by nosocomial bacteremia in critically ill patients. Infect Control Hosp Epidemiol
2005, 26(4):352-356.
Chapter 8: References
-191-
366. Vogelaers D, Lambert ML, Blot S: Impact estimates of nosocomial bloodstream infection: looking
from a different angle. Crit Care, 15(3):169.
367. Zahar JR, Timsit JF, Garrouste-Orgeas M, Francais A, Vesin A, Descorps-Declere A, Dubois Y,
Souweine B, Haouache H, Goldgran-Toledano D et al: Outcomes in severe sepsis and patients with
septic shock: pathogen species and infection sites are not associated with mortality. Crit Care Med
2011, 39(8):1886-1895.
368. Garrouste-Orgeas M, Timsit JF, Tafflet M, Misset B, Zahar JR, Soufir L, Lazard T, Jamali S,
Mourvillier B, Cohen Y et al: Excess risk of death from intensive care unit-acquired nosocomial
bloodstream infections: a reappraisal. Clin Infect Dis 2006, 42(8):1118-1126.
369. Januel JM, Harbarth S, Allard R, Voirin N, Lepape A, Allaouchiche B, Guerin C, Lehot JJ, Robert
MO, Fournier G et al: Estimating attributable mortality due to nosocomial infections acquired in
intensive care units. Infect Control Hosp Epidemiol, 31(4):388-394.
370. Sogaard M, Norgaard M, Pedersen L, Sorensen HT, Schonheyder HC: Blood culture status and
mortality among patients with suspected community-acquired bacteremia: a population-based cohort
study. BMC Infect Dis, 11:139.
371. Thompson DS: Estimates of the rate of acquisition of bacteraemia and associated excess mortality in
a general intensive care unit: a 10 year study. J Hosp Infect 2008, 69(1):56-61.
372. Hugonnet S, Harbarth S, Ferriere K, Ricou B, Suter P, Pittet D: Bacteremic sepsis in intensive care:
temporal trends in incidence, organ dysfunction, and prognosis. Crit Care Med 2003, 31(2):390-394.
373. van Deuren M, van der Ven-Jongekrijg J, Demacker PN, Bartelink AK, van Dalen R, Sauerwein RW,
Gallati H, Vannice JL, van der Meer JW: Differential expression of proinflammatory cytokines and
their inhibitors during the course of meningococcal infections. J Infect Dis 1994, 169(1):157-161.
374. Laupland KB, Gregson DB, Church DL, Ross T, Pitout JD: Incidence, risk factors and outcomes of
Escherichia coli bloodstream infections in a large Canadian region. Clin Microbiol Infect 2008,
14(11):1041-1047.
375. Lefort A, Panhard X, Clermont O, Woerther PL, Branger C, Mentre F, Fantin B, Wolff M, Denamur
E: Host factors and portal of entry outweigh bacterial determinants to predict the severity of
Escherichia coli bacteremia. J Clin Microbiol 2011, 49(3):777-783.
376. Johnson JR, Kuskowski MA, O'Bryan TT, Maslow JN: Epidemiological correlates of virulence
genotype and phylogenetic background among Escherichia coli blood isolates from adults with
diverse-source bacteremia. J Infect Dis 2002, 185(10):1439-1447.
377. Arditi, M., Ables, L., & Yogev, R. (1989). Cerebrospinal fluid endotoxin levels in children with H.
influenzae meningitis before and after administration of intravenous ceftriaxone. Journal of
Infectious Diseases, 160(6), 1005-1011.
378. Feldman, W. E. (1977). Relation of concentrations of bacteria and bacterial antigen in cerebrospinal
fluid to prognosis in patients with bacterial meningitis. New England Journal of Medicine, 296(8),
433-435.
379. Siempos, II, Vardakas KZ, Kyriakopoulos CE, Ntaidou TK, Falagas ME: Predictors of mortality in
adult patients with ventilator-associated pneumonia: a meta-analysis. Shock 2010, 33(6):590-601.
380. Kollef KE, Schramm GE, Wills AR, Reichley RM, Micek ST, Kollef MH: Predictors of 30-day
mortality and hospital costs in patients with ventilator-associated pneumonia attributed to potentially
antibiotic-resistant gram-negative bacteria. Chest2008, 134:281-287.
381. Agbaht K, Diaz E, Munoz E, Lisboa T, Gomez F, Depuydt PO, Blot SI, Rello J: Bacteremia in
patients with ventilator-associated pneumonia is associated with increased mortality: A study
comparing bacteremic vs. nonbacteremic ventilator-associated pneumonia. Crit Care Med 2007,
35(9):2064-2070.
382. Garnacho-Montero J, Ortiz-Leyba C, Fernández-Hinojosa E, Aldabó-Pallás T, Cayuela A, MarquezVácaro JA, Garcia-Curiel A, Jiménez-Jiménez FJ: Acinetobacter baumannii ventilator-associated
pneumonia: epidemiological and clinical findings. Intensive Care Med 31:649-655, 2005.
383. Iregui M, Ward S, Sherman G, Fraser VJ, Kollef MH: Clinical importance of delays in the initiation
of appropriate antibiotic treatment for ventilator-associated pneumonia. Chest 122:262-268, 2002.
-192-
Chapter 8: References
384. Crouch Brewer S, Wunderink RG, Jones CB, Leeper KV Jr: Ventilator-associated pneumonia due
to Pseudomonas aeruginosa. Chest 109:1019-1029, 1996.
385. Muscedere JG, McColl C, Shorr A, Jiang X, Marshall J, Heyland DK: Determinants of outcome in
patients with a clinical suspicion of ventilator-associated pneumonia. J Crit Care 2008, 23(1):41-49.
386. Luna CM, Videla A, Mattera J, Vay C, Famiglietti A, Vujacich P, Niederman MS: Blood cultures
have limited value in predicting severity of illness and as a diagnostic tool in ventilator-associated
pneumonia. Chest 1999, 116(4):1075-1084.
387. Fisher CJ, Jr., Dhainaut JF, Opal SM, Pribble JP, Balk RA, Slotman GJ, Iberti TJ, Rackow EC,
Shapiro MJ, Greenman RL et al: Recombinant human interleukin 1 receptor antagonist in the
treatment of patients with sepsis syndrome. Results from a randomized, double-blind, placebocontrolled trial. Phase III rhIL-1ra Sepsis Syndrome Study Group. JAMA 1994, 271(23):1836-1843.
388. From the Centers for Disease Control. Yersinia enterocolitica bacteremia and endotoxin shock
associated with red blood cell transfusions--United States, 1991. JAMA 1991, 265(17):2174-2175.
389. Schwalbe B, Spath-Schwalbe E: Endotoxin concentrations and cytokine responses in a patient with
fatal transfusion-associated sepsis. Transfusion 1998, 38(7):703-705.
390. Foreman NK, Wang WC, Cullen EJ, Jr., Stidham GL, Pearson TA, Shenep JL: Endotoxic shock
after transfusion of contaminated red blood cells in a child with sickle cell disease. Pediatr Infect Dis
J 1991, 10(8):624-626.
391. McAllister SK, Bland LA, Arduino MJ, Aguero SM, Wenger PN, Jarvis WR: Patient cytokine
response in transfusion-associated sepsis. Infect Immun 1994, 62(5):2126-2128.
392. Meers PD, Calder MW, Mazhar MM, Lawrie GM: Intravenous infusion of contaminated dextrose
solution: The Devonport incident. Lancet 1973, 2(7839):1189-1192.
393. Stevens AR, Jr., Legg JS, Henry BS, Dille JM, Kirby WM, Finch CA: Fatal transfusion reactions
from contamination of stored blood by cold growing bacteria. Ann Intern Med 1953, 39(6):12281239.
394. Maki DG, Rosenthal VD, Salomao R, Franzetti F, Rangel-Frausto MS: Impact of switching from
an open to a closed infusion system on rates of central line-associated bloodstream infection: a metaanalysis of time-sequence cohort studies in 4 countries. Infect Control Hosp Epidemiol, 32(1):50-58
395. Blot S, Vandewoude K, De Bacquer D, Colardyn F: Nosocomial bacteremia caused by antibioticresistant gram-negative bacteria in critically ill patients: clinical outcome and length of
hospitalization. Clin Infect Dis 2002, 34(12):1600-1606.
396. Blot S, Vandewoude K, Hoste E, Colardyn F: Reappraisal of attributable mortality in critically ill
patients with nosocomial bacteraemia involving Pseudomonas aeruginosa. J Hosp Infect 2003,
53(1):18-24.
397. Blot S, Vandewoude K, Hoste E, De Waele J, Kint K, Rosiers F, Vogelaers D, Colardyn F:
Absence of excess mortality in critically ill patients with nosocomial Escherichia coli bacteremia.
Infect Control Hosp Epidemiol 2003, 24(12):912-915.
398. Blot SI, Vandewoude KH, Colardyn FA: Clinical impact of nosocomial Klebsiella bacteremia in
critically ill patients. Eur J Clin Microbiol Infect Dis 2002, 21(6):471-473.
399. Blot SI, Vandewoude KH, Colardyn FA: Evaluation of outcome in critically ill patients with
nosocomial enterobacter bacteremia: results of a matched cohort study. Chest 2003, 123(4):12081213.
400. Blot S, Vandewoude K, Colardyn F: Nosocomial bacteremia involving Acinetobacter baumannii in
critically ill patients: a matched cohort study. Intensive Care Med 2003, 29(3):471-475.
401. Bryan CS: Clinical implications of positive blood cultures. Clin Microbiol Rev 1989, 2(4):329-353.
402. Peralta G, Roiz MP, Sanchez MB, Garrido JC, Ceballos B, Rodriguez-Lera MJ, Mateos F, De Benito
I: Time-to-positivity in patients with Escherichia coli bacteraemia. Clin Microbiol Infect 2007,
13(11):1077-1082.
403. Martinez JA, Soto S, Fabrega A, Almela M, Mensa J, Soriano A, Marco F, Jimenez de Anta MT,
Vila J: Relationship of phylogenetic background, biofilm production, and time to detection of
Chapter 8: References
-193-
growth in blood culture vials with clinical variables and prognosis associated with Escherichia coli
bacteremia. J Clin Microbiol 2006, 44(4):1468-1474.
404. Liao CH, Lai CC, Hsu MS, Huang YT, Chu FY, Hsu HS, Hsueh PR: Correlation between time to
positivity of blood cultures with clinical presentation and outcomes in patients with Klebsiella
pneumoniae bacteraemia: prospective cohort study. Clin Microbiol Infect 2009, 15(12):1119-1125.
405. Brun-Buisson C, Doyon F, Carlet J: Bacteremia and severe sepsis in adults: a multicenter
prospective survey in ICUs and wards of 24 hospitals. French Bacteremia-Sepsis Study Group. Am J
Respir Crit Care Med 1996, 154(3 Pt 1):617-624
406. Kang CI, Song JH, Chung DR, Peck KR, Ko KS, Yeom JS, Ki HK, Son JS, Lee SS, Kim YS et al:
Risk factors and pathogenic significance of severe sepsis and septic shock in 2286 patients with
gram-negative bacteremia. J Infect, 62(1):26-33.
407. Magret M, Lisboa T, Martin-Loeches I, Manez R, Nauwynck M, Wrigge H, Cardellino S, Diaz E,
Koulenti D, Rello J: Bacteremia is an independent risk factor for mortality in nosocomial
pneumonia: a prospective and observational multicenter study. Crit Care, 15(1):R62.
408. Wunsch H, Linde-Zwirble WT, Angus DC: Methods to adjust for bias and confounding in critical
care health services research involving observational data. J Crit Care 2006, 21(1):1-7.
409. Bekaert M, Timsit JF, Vansteelandt S, Depuydt P, Vesin A, Garrouste-Orgeas M, Decruyenaere J,
Clec'h C, Azoulay E, Benoit D: Attributable mortality of ventilator-associated pneumonia: a
reappraisal using causal analysis. Am J Respir Crit Care Med, 184(10):1133-1139.
410. Timsit JF, Fosse JP, Troche G, De Lassence A, Alberti C, Garrouste-Orgeas M, Bornstain C, Adrie
C, Cheval C, Chevret S: Calibration and discrimination by daily Logistic Organ Dysfunction scoring
comparatively with daily Sequential Organ Failure Assessment scoring for predicting hospital
mortality in critically ill patients. Crit Care Med 2002, 30(9):2003-2013.
411. MacArthur RD, Miller M, Albertson T, Panacek E, Johnson D, Teoh L, Barchuk W: Adequacy of
early empiric antibiotic treatment and survival in severe sepsis: experience from the MONARCS
trial. Clin Infect Dis 2004, 38(2):284-288.
412. McGregor JC, Rich SE, Harris AD, Perencevich EN, Osih R, Lodise TP, Jr., Miller RR, Furuno JP:
A systematic review of the methods used to assess the association between appropriate antibiotic
therapy and mortality in bacteremic patients. Clin Infect Dis 2007, 45(3):329-337.
413. Paul M, Shani V, Muchtar E, Kariv G, Robenshtok E, Leibovici L: Systematic review and metaanalysis of the efficacy of appropriate empiric antibiotic therapy for sepsis. Antimicrob Agents
Chemother, 54(11):4851-4863.
414. Safdar N, Handelsman J, Maki DG: Does combination antimicrobial therapy reduce mortality in
Gram-negative bacteraemia? A meta-analysis. Lancet Infect Dis 2004, 4(8):519-527.
415. Minneci PC, Deans KJ, Eichacker PQ, Natanson C: The effects of steroids during sepsis depend on
dose and severity of illness: an updated meta-analysis. Clin Microbiol Infect 2009, 15(4):308-318.
416. Rittirsch D, Hoesel LM, Ward PA: The disconnect between animal models of sepsis and human
sepsis. J Leukoc Biol 2007, 81(1):137-143.
417. Deitch EA: Animal models of sepsis and shock: a review and lessons learned. Shock 1998, 9(1):111.
418. Marshall JC: Sepsis: rethinking the approach to clinical research. J Leukoc Biol 2008, 83(3):471482.
419. Dyson A, Singer M: Animal models of sepsis: why does preclinical efficacy fail to translate to the
clinical setting? Crit Care Med 2009, 37(1 Suppl):S30-37.
420. Riedemann NC, Guo RF, Ward PA: Novel strategies for the treatment of sepsis. Nat Med 2003,
9(5):517-524.
421. Suffredini AF, Munford RS: Novel therapies for septic shock over the past 4 decades. JAMA,
306(2):194-199
422. Ziegler EJ: Protective antibody to endotoxin core: the emperor's new clothes? J Infect Dis 1988,
158(2):286-290.
-194-
Chapter 8: References
423. Warren HS, Danner RL, Munford RS: Anti-endotoxin monoclonal antibodies. N Engl J Med 1992,
326(17):1153-1157.
424. Cross AS, Opal SM, Bhattacharjee AK, Donta ST, Peduzzi PN, Furer E, Que JU, Cryz SJ:
Immunotherapy of sepsis: flawed concept or faulty implementation? Vaccine 1999, 17 Suppl 2:S1321.
425. Opal SM: The host response to endotoxin, antilipopolysaccharide strategies, and the management of
severe sepsis. Int J Med Microbiol 2007, 297(5):365-377.
426. Opal SM, Cross AS: Clinical trials for severe sepsis. Past failures, and future hopes. Infect Dis Clin
North Am 1999, 13(2):285-297, vii.
427. Greisman SE, Johnston CA. Review: Evidence against the hypothesis that antibodies to the inner
core of lipopolysaccharides in antisera raised by immunization with enterobacterial deep-rough
mutants confer broad-spectrum protection during Gram-negative bacterial sepsis. J Endotoxin Res
1997, 4(2) 123-153.
428. Chong KT, Huston M: Implications of endotoxin contamination in the evaluation of antibodies to
lipopolysaccharides in a murine model of gram-negative sepsis. J Infect Dis 1987, 156(5):713-719.
429. Wakelin SJ, Sabroe I, Gregory CD, Poxton IR, Forsythe JL, Garden OJ, Howie SE: "Dirty little
secrets"--endotoxin contamination of recombinant proteins. Immunol Lett 2006, 106(1):1-7.
430. McCabe WR, Kreger BE, Johns M: Type-specific and cross-reactive antibodies in gram-negative
bacteremia. N Engl J Med 1972, 287(6):261-267.
431. Zinner SH, McCabe WR: Effects of IgM and IgG antibody in patients with bacteremia due to gramnegative bacilli. J Infect Dis 1976, 133(1):37-45.
432. Pollack M, Huang AI, Prescott RK, Young LS, Hunter KW, Cruess DF, Tsai CM: Enhanced
survival in Pseudomonas aeruginosa septicemia associated with high levels of circulating antibody to
Escherichia coli endotoxin core. J Clin Invest 1983, 72(6):1874-1881.
433. Baumgartner JD, Glauser MP, McCutchan JA, Ziegler EJ, van Melle G, Klauber MR, Vogt M,
Muehlen E, Luethy R, Chiolero R et al: Prevention of gram-negative shock and death in surgical
patients by antibody to endotoxin core glycolipid. Lancet 1985, 2(8446):59-63.
434. McCutchan JA, Wolf JL, Ziegler EJ, Braude AI: Ineffectiveness of single-dose human antiserum to
core glycolipid (E. coli J5) for prophylaxis of bacteremic, gram-negative infections in patients with
prolonged neutropenia. Schweiz Med Wochenschr Suppl 1983, 14:40-45.
435. Prophylactic intravenous administration of standard immune globulin as compared with corelipopolysaccharide immune globulin in patients at high risk of postsurgical infection. The
Intravenous Immunoglobulin Collaborative Study Group. N Engl J Med 1992, 327(4):234-240.
436. Bennett-Guerrero E, Ayuso L, Hamilton-Davies C, White WD, Barclay GR, Smith PK, King SA,
Muhlbaier LH, Newman MF, Mythen MG: Relationship of preoperative antiendotoxin core
antibodies and adverse outcomes following cardiac surgery. JAMA 1997, 277(8):646-650.
437. J5 study Group. Treatment of severe infectious purpura in children with human plasma from donors
immunized with Escherichia coli J5: a prospective double-blind study. J Infect Dis 1992,
165(4):695-701.
438. Derkx B, Wittes J, McCloskey R: Randomized, placebo-controlled trial of HA-1A, a human
monoclonal antibody to endotoxin, in children with meningococcal septic shock. European Pediatric
Meningococcal Septic Shock Trial Study Group. Clin Infect Dis 1999, 28(4):770-777.
439. Levin M, Quint PA, Goldstein B, Barton P, Bradley JS, Shemie SD, Yeh T, Kim SS, Cafaro DP,
Scannon PJ et al: Recombinant bactericidal/permeability-increasing protein (rBPI21) as adjunctive
treatment for children with severe meningococcal sepsis: a randomised trial. rBPI21 Meningococcal
Sepsis Study Group. Lancet 2000, 356(9234):961-967.
440. Calandra T, Glauser MP, Schellekens J, Verhoef J: Treatment of gram-negative septic shock with
human IgG antibody to Escherichia coli J5: a prospective, double-blind, randomized trial. J Infect
Dis 1988, 158(2):312-319.
441. Grundmann R, Hornung M: Immunoglobulin therapy in patients with endotoxemia and
postoperative sepsis--a prospective randomized study. Prog Clin Biol Res 1988, 272:339-349.
Chapter 8: References
-195-
442. Schedel I, Dreikhausen U, Nentwig B, Hockenschnieder M, Rauthmann D, Balikcioglu S, Coldewey
R, Deicher H: Treatment of gram-negative septic shock with an immunoglobulin preparation: a
prospective, randomized clinical trial. Crit Care Med 1991, 19(9):1104-1113.
443. Behre G, Ostermann H, Schedel I: Endotoxin concentration and therapy with polyclonal
immunoglobulin M-enriched immunoglobulins in neutropenic patients with sepsis syndrome. Pilot
study and interim analysis of a randomized trial Anti-infect Drug Chemother 1995, 13:129-34.
444. Fisher CJ, Jr., Zimmerman J, Khazaeli MB, Albertson TE, Dellinger RP, Panacek EA, Foulke GE,
Dating C, Smith CR, LoBuglio AF: Initial evaluation of human monoclonal anti-lipid A antibody
(HA-1A) in patients with sepsis syndrome. Crit Care Med 1990, 18(12):1311-1315.
445. The French National Registry of HA-1A (Centoxin) in septic shock. A cohort study of 600 patients.
The National Committee for the Evaluation of Centoxin. Arch Intern Med 1994, 154(21):2484-2491.
446. McCloskey RV, Straube RC, Sanders C, Smith SM, Smith CR: Treatment of septic shock with
human monoclonal antibody HA-1A. A randomized, double-blind, placebo-controlled trial. CHESS
Trial Study Group. Ann Intern Med 1994, 121(1):1-5.
447. Angus DC, Birmingham MC, Balk RA, Scannon PJ, Collins D, Kruse JA, Graham DR, Dedhia HV,
Homann S, MacIntyre N: E5 murine monoclonal antiendotoxin antibody in gram-negative sepsis: a
randomized controlled trial. E5 Study Investigators. Jama 2000, 283(13):1723-1730.
448. Daifuku R, Haenftling K, Young J, Groves ES, Turrell C, Meyers FJ: Phase I study of
antilipopolysaccharide human monoclonal antibody MAB-T88. Antimicrob Agents Chemother 1992,
36(10):2349-2351.
449. Greenberg RN, Wilson KM, Kunz AY, Wedel NI, Gorelick KJ: Randomized, double-blind phase II
study of anti-endotoxin antibody (E5) as adjuvant therapy in humans with serious gram-negative
infections. Prog Clin Biol Res 1991, 367:179-186.
450. Greenberg RN, Wilson KM, Kunz AY, Wedel NI, Gorelick KJ: Observations using antiendotoxin
antibody (E5) as adjuvant therapy in humans with suspected, serious, gram-negative sepsis. Crit
Care Med 1992, 20(6):730-735.
451. Albertson TE, Panacek EA, MacArthur RD, Johnson SB, Benjamin E, Matuschak GM, Zaloga G,
Maki D, Silverstein J, Tobias JK et al: Multicenter evaluation of a human monoclonal antibody to
Enterobacteriaceae common antigen in patients with Gram-negative sepsis. Crit Care Med 2003,
31(2):419-427.
452. Willatts SM, Radford S, Leitermann M: Effect of the antiendotoxic agent, taurolidine, in the
treatment of sepsis syndrome: a placebo-controlled, double-blind trial. Crit Care Med 1995,
23(6):1033-1039.
453. Reinhart K, Meier-Hellmann A, Beale R, Forst H, Boehm D, Willatts S, Rothe KF, Adolph M,
Hoffmann JE, Boehme M et al: Open randomized phase II trial of an extracorporeal endotoxin
adsorber in suspected Gram-negative sepsis. Crit Care Med 2004, 32(8):1662-1668.
454. Bennett-Guerrero E, Grocott HP, Levy JH, Stierer KA, Hogue CW, Cheung AT, Newman MF,
Carter AA, Rossignol DP, Collard CD: A phase II, double-blind, placebo-controlled, ascending-dose
study of Eritoran (E5564), a lipid A antagonist, in patients undergoing cardiac surgery with
cardiopulmonary bypass. Anesth Analg 2007, 104(2):378-383.
455. Tidswell M, Tillis W, Larosa SP, Lynn M, Wittek AE, Kao R, Wheeler J, Gogate J, Opal SM: Phase
2 trial of eritoran tetrasodium (E5564), a toll-like receptor 4 antagonist, in patients with severe sepsis.
Crit Care Med, 38(1):72-83.
456. Mitka M: Drug for severe sepsis is withdrawn from market, fails to reduce mortality. JAMA,
306(22):2439-2440.
457. Dellinger RP, Tomayko JF, Angus DC, Opal S, Cupo MA, McDermott S, Ducher A, Calandra T,
Cohen J: Efficacy and safety of a phospholipid emulsion (GR270773) in Gram-negative severe
sepsis: results of a phase II multicenter, randomized, placebo-controlled, dose-finding clinical trial.
Crit Care Med 2009, 37(11):2929-2938.
458. Heemskerk S, Masereeuw R, Moesker O, Bouw MP, van der Hoeven JG, Peters WH, Russel FG,
Pickkers P: Alkaline phosphatase treatment improves renal function in severe sepsis or septic shock
patients. Crit Care Med 2009, 37(2):417-423
-196-
Chapter 8: References
459. Pickkers P, Heemskerk S, Schouten J, Laterre PF, Vincent JL, Beishuizen A, Jorens PG, Spapen H,
Bulitta M, Peters WH et al: Alkaline phosphatase for treatment of sepsis-induced acute kidney
injury: a prospective randomized double-blind placebo-controlled trial. Crit Care, 16(1):R14.
460. Warren HS, Amato SF, Fitting C, Black KM, Loiselle PM, Pasternack MS, Cavaillon JM:
Assessment of ability of murine and human anti-lipid A monoclonal antibodies to bind and
neutralize lipopolysaccharide. J Exp Med 1993, 177(1):89-97.
461. Opal, SM, Patrozou E. Translational research in the development of novel sepsis therapeutics:
Logical deductive reasoning or mission impossible? Crit Care Med 37.1 (2009): S10-S15.
462. Lamontagne F, Briel M, Duffett M, Fox-Robichaud A, Cook DJ, Guyatt G, Lesur O, & Meade MO.
(2010). Systematic review of reviews including animal studies addressing therapeutic interventions
for sepsis. Crit Care Med, 38(12), 2401.
463. LJ Berry, Introduction, in, Handbook of Endotoxin, Vol 3 Cellular biology of endotoxin. pages xviixxi, Editor LJ Berry, Elsevier 1985
464. Michie HR: The value of animal models in the development of new drugs for the treatment of the
sepsis syndrome. J Antimicrob Chemother 1998, 41 Suppl A:47-49.
465. Cross AS, Opal SM, Sadoff JC, Gemski P: Choice of bacteria in animal models of sepsis. Infect
Immun 1993, 61(7):2741-2747.
466. Parker SJ, Watkins PE: Experimental models of gram-negative sepsis. Br J Surg 2001, 88(1):22-30.
467. Danner RL, Eichacker PQ, Doerfler ME, Hoffman WD, Reilly JM, Wilson J, MacVittie TJ, Stuetz P,
Parrillo JE, Natanson C: Therapeutic trial of lipid X in a canine model of septic shock. J Infect Dis
1993, 167(2):378-384.
468. Eichacker PQ, Hoffman WD, Farese A, Danner RL, Suffredini AF, Waisman Y, Banks SM,
Mouginis T, Wilson L, Rothlein R et al: Leukocyte CD18 monoclonal antibody worsens
endotoxemia and cardiovascular injury in canines with septic shock. J Appl Physiol 1993,
74(4):1885-1892.
469. Hoffman WD, Danner RL, Quezado ZM, Banks SM, Elin RJ, Hosseini JM, Natanson C: Role of
endotoxemia in cardiovascular dysfunction and lethality: virulent and nonvirulent Escherichia coli
challenges in a canine model of septic shock. Infect Immun 1996, 64(2):406-412.
470. Natanson C, Danner RL, Elin RJ, Hosseini JM, Peart KW, Banks SM, MacVittie TJ, Walker RI,
Parrillo JE: Role of endotoxemia in cardiovascular dysfunction and mortality. Escherichia coli and
Staphylococcus aureus challenges in a canine model of human septic shock. J Clin Invest 1989,
83(1):243-251.
471. Natanson C, Fink MP, Ballantyne HK, MacVittie TJ, Conklin JJ, Parrillo JE: Gram-negative
bacteremia produces both severe systolic and diastolic cardiac dysfunction in a canine model that
simulates human septic shock. J Clin Invest 1986, 78(1):259-270.
472. Natanson C, Hoffman WD, Koev LA, Dolan DP, Banks SM, Bacher J, Danner RL, Klein HG,
Parrillo JE: Plasma exchange does not improve survival in a canine model of human septic shock.
Transfusion 1993, 33(3):243-248.
473. Quezado ZM, Natanson C, Banks SM, Alling DW, Koev CA, Danner RL, Elin RJ, Hosseini JM,
Parker TS, Levine DM et al: Therapeutic trial of reconstituted human high-density lipoprotein in a
canine model of gram-negative septic shock. J Pharmacol Exp Ther 1995, 272(2):604-611.
474. Quezado ZM, Hoffman WD, Winkelstein JA, Yatsiv I, Koev CA, Cork LC, Elin RJ, Eichacker PQ,
Natanson C: The third component of complement protects against Escherichia coli endotoxininduced shock and multiple organ failure. J Exp Med 1994, 179(2):569-578.
475. Freeman BD, Quezado Z, Zeni F, Natanson C, Danner RL, Banks S, Quezado M, Fitz Y, Bacher J,
Eichacker PQ: rG-CSF reduces endotoxemia and improves survival during E. coli pneumonia. J
Appl Physiol 1997, 83(5):1467-1475.
476. Solomon SB, Cui X, Gerstenberger E, Danner RL, Fitz Y, Banks SM, Natanson C, Eichacker PQ:
Effective dosing of lipid A analogue E5564 in rats depends on the timing of treatment and the route
of Escherichia coli infection. J Infect Dis 2006, 193(5):634-644.
Chapter 8: References
-197-
477. Rogy MA, Moldawer LL, Oldenburg HS, Thompson WA, Montegut WJ, Stackpole SA, Kumar A,
Palladino MA, Marra MN, Lowry SF: Anti-endotoxin therapy in primate bacteremia with HA-1A
and BPI. Ann Surg 1994, 220(1):77-85.
478. Barclay GR: Endotoxin-core antibodies: time for a reappraisal? Intensive Care Med 1999,
25(5):427-429.
479. Cometta A, Baumgartner JD, Glauser MP: Polyclonal intravenous immune globulin for prevention
and treatment of infections in critically ill patients. Clin Exp Immunol 1994, 97 Suppl 1:69-72.
480. Opal SM, Jhung JW, Keith JC, Jr., Goldman SJ, Palardy JE, Parejo NA: Additive effects of human
recombinant interleukin-11 and granulocyte colony-stimulating factor in experimental gram-negative
sepsis. Blood 1999, 93(10):3467-3472.
481. Turgeon AF, Hutton B, Fergusson DA, McIntyre L, Tinmouth AA, Cameron DW, Hebert PC: Metaanalysis: intravenous immunoglobulin in critically ill adult patients with sepsis. Ann Intern Med
2007, 146(3):193-203.
482. Werdan K, Pilz G, Bujdoso O, Fraunberger P, Neeser G, Schmieder RE, Viell B, Marget W,
Seewald M, Walger P et al: Score-based immunoglobulin G therapy of patients with sepsis: the
SBITS study. Crit Care Med 2007, 35(12):2693-2701.
483. Flynn PM, Shenep JL, Stokes DC, Fairclough D, Hildner WK: Polymyxin B moderates acidosis and
hypotension in established, experimental gram-negative septicemia. J Infect Dis 1987, 156(5):706712.
484. Cavaillon JM, Haeffner-Cavaillon N: Polymyxin-B inhibition of LPS-induced interleukin-1
secretion by human monocytes is dependent upon the LPS origin. Mol Immunol 1986, 23(9):965969.
485. Davies B, Cohen J: Endotoxin removal devices for the treatment of sepsis and septic shock. Lancet
Infect Dis, 11(1):65-71.
486. Opal SM: Hemofiltration-absorption systems for the treatment of experimental sepsis: is it possible
to remove the "evil humors" responsible for septic shock? Crit Care Med 2000, 28(5):1681-1682.
487. Cruz DN, Bellomo R, Ronco C: Clinical effects of polymyxin B-immobilized fiber column in septic
patients. Contrib Nephrol 2007, 156:444-451.
488. Cruz DN, Perazella MA, Bellomo R, de Cal M, Polanco N, Corradi V, Lentini P, Nalesso F, Ueno T,
Ranieri VM et al: Effectiveness of polymyxin B-immobilized fiber column in sepsis: a systematic
review. Crit Care 2007, 11(2):R47.
489. Aoki H, Kodama M, Tani T, Hanasawa K: Treatment of sepsis by extracorporeal elimination of
endotoxin using polymyxin B-immobilized fiber. Am J Surg 1994, 167(4):412-417.
490. Buttenschoen K, Radermacher P, Bracht H: Endotoxin elimination in sepsis: physiology and
therapeutic application. Langenbecks Arch Surg, 395(6):597-605.
491. Cruz DN, Antonelli M, Fumagalli R, Foltran F, Brienza N, Donati A, Malcangi V, Petrini F, Volta
G, Bobbio Pallavicini FM et al: Early use of polymyxin B hemoperfusion in abdominal septic shock:
the EUPHAS randomized controlled trial. JAMA 2009, 301(23):2445-2452.
492. Novelli G, Ferretti G, Poli L, Pretagostini R, Ruberto F, Perrella SM, Levi Sandri GB, Morabito V,
Berloco PB: Clinical results of treatment of postsurgical endotoxin-mediated sepsis with polymyxinB direct hemoperfusion. Transplant Proc, 42(4):1021-1024.
493. Novelli G, Ferretti G, Ruberto F, Morabito V, Pugliese F: Early management of endotoxemia using
the endotoxin activity assay and polymyxin B-based hemoperfusion. Contrib Nephrol, 167:91-101.
494. Vincent JL, Laterre PF, Cohen J, Burchardi H, Bruining H, Lerma FA, Wittebole X, De Backer D,
Brett S, Marzo D et al: A pilot-controlled study of a polymyxin B-immobilized hemoperfusion
cartridge in patients with severe sepsis secondary to intra-abdominal infection. Shock 2005,
23(5):400-405.
495. Cavaillon JM: Polymyxin B for endotoxin removal in sepsis. Lancet Infect Dis, 11(6):426-427.
-198-
Chapter 8: References
496. Ullrich H, Jakob W, Frohlich D, Rothe G, Prasser C, Drobnik W, Taeger K, Meier-Hellmann A,
Reinhart K, Zimmermann M et al: A new endotoxin adsorber: first clinical application. Ther Apher
2001, 5(5):326-334.
497. Hurley JC, Guidet B, Offenstadt G, Maury E: Endotoxemia and mortality prediction in ICU and
other settings: underlying risk and co-detection of gram negative bacteremia are confounders. Crit
Care, 16(4):R148.
Surgery
498. Baigrie RJ, Lamont PM, Whiting S, Morris PJ: Portal endotoxin and cytokine responses during
abdominal aortic surgery. Am J Surg 1993, 166(3):248-251.
499. Berger D, Bolke E, Seidelmann M, Beger HG: Time-scale of interleukin-6, myeloid related proteins
(MRP), C reactive protein (CRP), and endotoxin plasma levels during the postoperative acute phase
reaction. Shock 1997, 7(6):422-426.
500. Berger D, Ott S, Schmidt UM, Bolke E, Seidelmann M, Beger HG: Determination of endotoxinneutralizing capacity of plasma in postsurgical patients. Eur Surg Res 1996, 28(2):130-139.
501. Berger D, Schmidt UM, Ott S, Seidelmann M, Martin R, Beger HG: Incidence and
pathophysiological relevance of postoperative endotoxemia. FEMS Immunol Med Microbiol 1995,
11(4):285-290.
502. Boelke E, Storck M, Buttenschoen K, Berger D, Hannekum A: Endotoxemia and mediator release
during cardiac surgery. Angiology 2000, 51(9):743-749.
503. Bolke E, Jehle PM, Storck M, Nothnagel B, Stanescu A, Orth K: Endotoxin release and endotoxin
neutralizing capacity during colonoscopy. Clin Chim Acta 2001, 303(1-2):49-53.
504. Brinkmann A, Wolf CF, Berger D, Kneitinger E, Neumeister B, Buchler M, Radermacher P, Seeling
W, Georgieff M: Perioperative endotoxemia and bacterial translocation during major abdominal
surgery: evidence for the protective effect of endogenous prostacyclin? Crit Care Med 1996,
24(8):1293-1301.
505. Buttenschoen K, Buttenschoen DC, Berger D, Vasilescu C, Schafheutle S, Goeltenboth B,
Seidelmann M, Beger HG: Endotoxemia and acute-phase proteins in major abdominal surgery. Am J
Surg 2001, 181(1):36-43.
506. Jansen NJ, van Oeveren W, Gu YJ, van Vliet MH, Eijsman L, Wildevuur CR: Endotoxin release and
tumor necrosis factor formation during cardiopulmonary bypass. Ann Thorac Surg 1992, 54(4):744747; discussion 747-748.
507. Kato M, Kaise M, Obata T, Yonezawa J, Toyoizumi H, Yoshimura N, Yoshida Y, Kawamura M,
Tajiri H: Bacteremia and endotoxemia after endoscopic submucosal dissection for gastric neoplasia:
pilot study. Gastric Cancer, 15(1):15-20.
508. Kats S, Schonberger JP, Brands R, Seinen W, van Oeveren W: Endotoxin release in cardiac surgery
with cardiopulmonary bypass: pathophysiology and possible therapeutic strategies. An update. Eur J
Cardiothorac Surg, 39(4):451-458.
509. Kokuba Y, Nakao A, Takagi H: Changes in the blood endotoxin concentration after digestive
surgery. Surg Today 1993, 23(2):100-104.
510. Konickova Z, Konigova R, Likovsky Z: Endotoxemia and endotoxemia diagnostics. Scand J Plast
Reconstr Surg 1979, 13(1):79-80.
511. Kraft AR, Jones RB, Homer LD, Merecki BE, Herman CM: Detection of detoxified endotoxin by
the Limulus lysate assay. Surg Forum 1973, 24:68-70.
512. Lau JY, Lo SY, Ng EK, Lee DW, Lam YH, Chung SC: A randomized comparison of acute phase
response and endotoxemia in patients with perforated peptic ulcers receiving laparoscopic or open
patch repair. Am J Surg 1998, 175(4):325-327.
513. Lau LL, Halliday MI, Lee B, Hannon RJ, Gardiner KR, Soong CV: Intestinal manipulation during
elective aortic aneurysm surgery leads to portal endotoxaemia and mucosal barrier dysfunction. Eur
J Vasc Endovasc Surg 2000, 19(6):619-624.
Chapter 8: References
-199-
514. Lequier LL, Nikaidoh H, Leonard SR, Bokovoy JL, White ML, Scannon PJ, Giroir BP: Preoperative
and postoperative endotoxemia in children with congenital heart disease. Chest 2000, 117(6):17061712.
515. Palmer KR, Duerden BI, Holdsworth CD: Bacteriological and endotoxin studies in cases of
ulcerative colitis submitted to surgery. Gut 1980, 21(10):851-854.
516. Pathan N, Burmester M, Adamovic T, Berk M, Ng KW, Betts H, Macrae D, Waddell S, Paul-Clark
M, Nuamah R et al: Intestinal injury and endotoxemia in children undergoing surgery for congenital
heart disease. Am J Respir Crit Care Med, 184(11):1261-1269.
517. Plachouras D, Stamatakos M, Baziaka F, Giamarellos-Bourboulis E, Tsaganos T, Giamarellou H,
Safioleas M: Portal and systemic endotoxemia in abdominal operations: the significance of acute
abdomen. J Surg Res 2006, 134(1):133-137.
518. Riddington DW, Venkatesh B, Boivin CM, Bonser RS, Elliott TS, Marshall T, Mountford PJ, Bion
JF: Intestinal permeability, gastric intramucosal pH, and systemic endotoxemia in patients
undergoing cardiopulmonary bypass. JAMA 1996, 275(13):1007-1012.
519. Rocke DA, Gaffin SL, Wells MT, Koen Y, Brock-Utine JG: Endotoxemia associated with
cardiopulmonary bypass. J Thorac Cardiovasc Surg 1987, 93(6):832-837.
520. Roscher R, Oettinger W, Beger HG: Bacterial microflora, endogenous endotoxin, and prostaglandins
in small bowel obstruction. Am J Surg 1988, 155(2):348-355.
521. Roumen RM, Frieling JT, van Tits HW, van der Vliet JA, Goris RJ: Endotoxemia after major
vascular operations. J Vasc Surg 1993, 18(5):853-857.
522. Rowe MI, Buckner DM, Newmark S: The early diagnosis of gram negative septicemia in the
pediatric surgical patient. Ann Surg 1975, 182(3):280-286.
523. Schietroma M, Cappelli S, Carlei F, Di Giuro G, Agnifili A, Recchia CL, Antonellis M, Amicucci
G: [Intestinal and systemic endotoxaemia after laparotomic or laparoscopic cholecystectomy]. Chir
Ital 2006, 58(2):171-177.
524. Schietroma M, Carlei F, Cappelli S, Amicucci G: Intestinal permeability and systemic endotoxemia
after laparotomic or laparoscopic cholecystectomy. Ann Surg 2006, 243(3):359-363.
525. Schoeffel U, Pelz K, Haring RU, Amberg R, Schandl R, Urbaschek R, von Specht BU, Farthmann
EH: Inflammatory consequences of the translocation of bacteria and endotoxin to mesenteric lymph
nodes. Am J Surg 2000, 180(1):65-72.
526. Shimizu T, Endo Y, Tabata T, Mori T, Hanasawa K, Tsuchiya M, Tani T: Diagnostic and predictive
value of the silkworm larvae plasma test for postoperative infection following gastrointestinal
surgery. Crit Care Med 2005, 33(6):1288-1295.
527. Soong CV, Blair PH, Halliday MI, McCaigue MD, Campbell GR, Hood JM, Rowlands BJ, Barros
D'Sa AA: Endotoxaemia, the generation of the cytokines and their relationship to intramucosal
acidosis of the sigmoid colon in elective abdominal aortic aneurysm repair. Eur J Vasc Surg 1993,
7(5):534-539.
528. Wardle EN: Endotoxinemia in renal transplant recipients. Clin Nephrol 1975, 3(1):31-33.
529. Welch M, Douglas JT, Smyth JV, Walker MG: Systemic endotoxaemia and fibrinolysis during
aortic surgery. Eur J Vasc Endovasc Surg 1995, 9(2):228-232.
530. Zuhlke HV, Lorenz EP, Harnoss BM, Haring R, Rodloff A: [Endotoxinemia and bacteremia in
manual oral decompression of ileus]. Chirurg 1988, 59(5):349-356.
Pancreatitits
531. Ammori BJ, Barclay GR, Larvin M, McMahon MJ: Hypocalcemia in patients with acute
pancreatitis: a putative role for systemic endotoxin exposure. Pancreas 2003, 26(3):213-217.
532. Ammori BJ, Fitzgerald P, Hawkey P, McMahon MJ: The early increase in intestinal permeability
and systemic endotoxin exposure in patients with severe acute pancreatitis is not associated with
systemic bacterial translocation: molecular investigation of microbial DNA in the blood. Pancreas
2003, 26(1):18-22.
-200-
Chapter 8: References
533. Ammori BJ, Leeder PC, King RF, Barclay GR, Martin IG, Larvin M, McMahon MJ: Early increase
in intestinal permeability in patients with severe acute pancreatitis: correlation with endotoxemia,
organ failure, and mortality. J Gastrointest Surg 1999, 3(3):252-262.
534. Buttenschoen K, Berger D, Hiki N, Buttenschoen DC, Vasilescu C, Chikh-Torab F, Seidelmann M,
Beger HG: Endotoxin and antiendotoxin antibodies in patients with acute pancreatitis. Eur J Surg
2000, 166(6):459-466.
535. Giamarellos-Bourboulis EJ, Nikou GC, Matsaggoura M, Toumpanakis C, Grecka P, Giannikopoulos
G, Katsilambros N: Alterations of systemic endotoxemia over the course of acute edematous
pancreatitis. correlation to the advent of an infection? Pancreatology 2003, 3(4):323-328.
536. Rahman SH, Ammori BJ, Larvin M, McMahon MJ: Increased nitric oxide excretion in patients with
severe acute pancreatitis: evidence of an endotoxin mediated inflammatory response? Gut 2003,
52(2):270-274.
537. Windsor JA, Fearon KC, Ross JA, Barclay GR, Smyth E, Poxton I, Garden OJ, Carter DC: Role of
serum endotoxin and antiendotoxin core antibody levels in predicting the development of multiple
organ failure in acute pancreatitis. Br J Surg 1993, 80(8):1042-1046.
Cirrhosis, obstructive jaundice & liver disease
538. Bauer TM, Schwacha H, Steinbruckner B, Brinkmann FE, Ditzen AK, Aponte JJ, Pelz K, Berger D,
Kist M, Blum HE: Small intestinal bacterial overgrowth in human cirrhosis is associated with
systemic endotoxemia. Am J Gastroenterol 2002, 97(9):2364-2370.
539. Bercoff E, Le Bihan G, Birembaut JC, Bourreille J, Caline M, Lemeland JF: [Endotoxins and liver
cirrhosis: results of 118 studies using the limulus test]. Rev Med Interne 1982, 3(2):137-141.
540. Bode C, Kugler V, Bode JC: Endotoxemia in patients with alcoholic and non-alcoholic cirrhosis and
in subjects with no evidence of chronic liver disease following acute alcohol excess. J Hepatol 1987,
4(1):8-14.
541. Cahill CJ, Pain JA, Bailey ME: Bile salts, endotoxin and renal function in obstructive jaundice. Surg
Gynecol Obstet 1987, 165(6):519-522.
542. Chan CC, Hwang SJ, Lee FY, Wang SS, Chang FY, Li CP, Chu CJ, Lu RH, Lee SD: Prognostic
value of plasma endotoxin levels in patients with cirrhosis. Scand J Gastroenterol 1997, 32(9):942946.
543. Fukui H, Brauner B, Bode JC, Bode C: Plasma endotoxin concentrations in patients with alcoholic
and non-alcoholic liver disease: reevaluation with an improved chromogenic assay. J Hepatol 1991,
12(2):162-169.
544. Fukui H, Matsumoto M, Bode C, Bode JC, Tsujita S, Tsujii T: Endotoxaemia in patients with liver
cirrhosis and upper gastrointestinal bleeding: detection by the chromogenic assay with plasma
Tween 80 pretreatment. J Gastroenterol Hepatol 1993, 8(6):577-581.
545. Gaeta GB, Perna P, Adinolfi LE, Utili R, Ruggiero G: Endotoxemia in a series of 104 patients with
chronic liver diseases: prevalence and significance. Digestion 1982, 23(4):239-244.
546. Guarner C, Soriano G, Tomas A, Bulbena O, Novella MT, Balanzo J, Vilardell F, Mourelle M,
Moncada S: Increased serum nitrite and nitrate levels in patients with cirrhosis: relationship to
endotoxemia. Hepatology 1993, 18(5):1139-1143.
547. Kim BI, Kim HJ, Park JH, Park DI, Cho YK, Sohn CI, Jeon WK, Kim HS, Kim DJ: Increased
intestinal permeability as a predictor of bacterial infections in patients with decompensated liver
cirrhosis and hemorrhage. J Gastroenterol Hepatol, 26(3):550-557.
548. Kimmings AN, van Deventer SJ, Obertop H, Rauws EA, Huibregtse K, Gouma DJ: Endotoxin,
cytokines, and endotoxin binding proteins in obstructive jaundice and after preoperative biliary
drainage. Gut 2000, 46(5):725-731.
549. Kimmings AN, van Deventer SJ, Rauws EAJ, Huibregtse K, Gouma DJ: Systemic inflammatory
response in acute cholangitis and after subsequent treatment. Eur J Surg 2000, 166(9):700-705.
Chapter 8: References
-201-
550. Lumsden AB, Henderson JM, Kutner MH: Endotoxin levels measured by a chromogenic assay in
portal, hepatic and peripheral venous blood in patients with cirrhosis. Hepatology 1988, 8(2):232236.
551. Nolan JP: The role of intestinal endotoxin in liver injury: a long and evolving history. Hepatology,
52(5):1829-1835.
552. Ohshio G, Manabe T, Tobe T, Yoshioka H, Hamashima Y: Circulating immune complex, endotoxin,
and biliary infection in patients with biliary obstruction. Am J Surg 1988, 155(2):343-347.
553. Rao RK, Seth A, Sheth P: Recent Advances in Alcoholic Liver Disease I. Role of intestinal
permeability and endotoxemia in alcoholic liver disease. Am J Physiol Gastrointest Liver Physiol
2004, 286(6):G881-884.
554. Roughneen PT, Kumar SC, Pellis NR, Rowlands BJ: Endotoxemia and cholestasis. Surg Gynecol
Obstet 1988, 167(3):205-210.
555. Schafer C, Parlesak A, Schutt C, Bode JC, Bode C: Concentrations of lipopolysaccharide-binding
protein, bactericidal/permeability-increasing protein, soluble CD14 and plasma lipids in relation to
endotoxaemia in patients with alcoholic liver disease. Alcohol Alcohol 2002, 37(1):81-86.
556. Tarao K, Moroi T, Hirabayashi Y, Ikeuchi T, Endo O, Takamura Y: Effect of paromomycin sulfate
on endotoxemia in patients with cirrhosis. J Clin Gastroenterol 1982, 4(3):263-267.
557. Tarao K, Moroi T, Nagakura Y, Ikeuchi T, Suyama T, Endo O, Fukushima K: Relationship between
endotoxaemia and protein concentration of ascites in cirrhotic patients. Gut 1979, 20(3):205-210.
558. Tarao K, So K, Moroi T, Ikeuchi T, Endo O: [Studies of endotoxin in liver cirrhosis. 2. Clinical
significance of the appearance of endotoxin during the cirrhotic change of the liver]. Nihon
Shokakibyo Gakkai Zasshi 1977, 74(6):748-754.
559. Thompson JN, Blumgart LH: Endotoxaemia in obstructive jaundice. Br J Surg 1984, 71(3):247-248.
560. Thompson JN, Cohen J, Moore RH, Blenkharn JI, McConnell JS, Matkin J, Blumgart LH:
Endotoxemia in obstructive jaundice. Observations on cause and clinical significance. Am J Surg
1988, 155(2):314-321.
561. Triger DR, Boyer TD, Levin J: Portal and systemic bacteraemia and endotoxaemia in liver disease.
Gut 1978, 19(10):935-939.
562. Violi F, Ferro D, Basili S, Saliola M, Quintarelli C, Alessandri C, Cordova C: Association between
low-grade disseminated intravascular coagulation and endotoxemia in patients with liver cirrhosis.
Gastroenterology 1995, 109(2):531-539.
563. Wang SS, Lee FY, Chan CC, Lu RH, Chao Y, Lin HC, Wu SL, Tsai YT, Lee SD: Sequential
changes in plasma cytokine and endotoxin levels in cirrhotic patients with bacterial infection. Clin
Sci (Lond) 2000, 98(4):419-425.
Liver transplantation; donors and recipients
564. Fugger R, Hamilton G, Steininger R, Mirza D, Schulz F, Muhlbacher F: Intraoperative estimation of
endotoxin, TNF alpha, and IL-6 in orthotopic liver transplantation and their relation to rejection and
postoperative infection. Transplantation 1991, 52(2):302-306.
565. Hamilton G, Prettenhofer M, Zommer A, Hofbauer S, Gotzinger P, Gnant FX, Fugger R:
Intraoperative course and prognostic significance of endotoxin, tumor necrosis factor-alpha and
interleukin-6 in liver transplant recipients. Immunobiology 1991, 182(5):425-439.
566. McNicol L, Liu G, Hurley J, Hardy KJ, Jones RM: Endotoxin and tumor necrosis factor levels and
hemodynamics during human liver transplantation. Transplant Proc 1993, 25(2):1828-1829.
567. Miyata T, Yokoyama I, Todo S, Tzakis A, Selby R, Starzl TE: Endotoxaemia, pulmonary
complications, and thrombocytopenia in liver transplantation. Lancet 1989, 2(8656):189-191.
568. Pillay SP, Moore TD, Lynch SV, Whiting RF, Balderson GA, Strong RW: Endotoxin levels in adult
liver donors. Aust N Z J Surg 1994, 64(9):615-617.
569. Ronholm E, Tomasdottir H, Runeborg J, Mattsby-Baltzer I, Olausson M, Aneman A, Bengtsson A:
Gastro-intestinal complement activation during human liver transplantation: impact on postoperative
liver function. Acta Anaesthesiol Scand 2000, 44(7):850-857.
-202-
Chapter 8: References
570. Winchurch RA, Thupari JN, Munster AM: Endotoxemia in burn patients: levels of circulating
endotoxins are related to burn size. Surgery 1987, 102(5):808-812.
571. Yokoyama I, Todo S, Miyata T, Selby R, Tzakis AG, Starzl TE: Endotoxemia and human liver
transplantation. Transplant Proc 1989, 21(5):3833-3841.
572. Zipfel A, Schenk M, You MS, Lauchart W, Bode C, Viebahn R: Endotoxemia in organ donors: graft
function following liver transplantation. Transpl Int 2000, 13 Suppl 1:S286-287.
Urological
573. Bag S, Kumar S, Taneja N, Sharma V, Mandal AK, Singh SK: One week of nitrofurantoin before
percutaneous nephrolithotomy significantly reduces upper tract infection and urosepsis: a
prospective controlled study. Urology, 77(1):45-49.
574. Garibaldi RA, Allman GW, Larsen DH, Smith CB, Burke JP: Detection of endotoxemia by the
limulus test in patients with indwelling urinary catheters. J Infect Dis 1973, 128(4):551-554.
575. Kumar S, Bag S, Ganesamoni R, Mandal AK, Taneja N, Singh SK: Risk factors for urosepsis
following percutaneous nephrolithotomy: role of 1 week of nitrofurantoin in reducing the risk of
urosepsis. Urol Res, 40(1):79-86.
576. Li L, Shen Z, Wang H, Fu S, Cheng G: Investigation of infection risk and the value of urine
endotoxin during extracorporeal shock wave lithotripsy. Chin Med J (Engl) 2001, 114(5):510-513.
577. Luchi M, Morrison DC, Opal S, Yoneda K, Slotman G, Chambers H, Wiesenfeld H, Lemke J, Ryan
JL, Horn D: A comparative trial of imipenem versus ceftazidime in the release of endotoxin and
cytokine generation in patients with gram-negative urosepsis. Urosepsis Study Group. J Endotoxin
Res 2000, 6(1):25-31.
578. Tanaka M, Matsumoto T, Kitada S, Kumazawa J, Hara S, Yamaguchi A: Endotoxemia in patients
who underwent ultrasonic lithotripsy and extracorporeal shock wave lithotripsy. Eur Urol 1988,
14(3):173-177.
579. Rao PN, Dube DA, Weightman NC, Oppenheim BA, Morris J: Prediction of septicemia following
endourological manipulation for stones in the upper urinary tract. J Urol 1991, 146(4):955-960.
580. Thompson PM, Pryor JP, Williams JP, Eyers DE, Dulake C, Scully MF, Kakkar VV: The problem
of infection after prostatic biopsy: the case for the transperineal approach. Br J Urol 1982,
54(6):736-740.
Trauma
581. Buttenschoen K, Berger D, Hiki N, Strecker W, Seidelmann M, Beger HG: Plasma concentrations of
endotoxin and antiendotoxin antibodies in patients with multiple injuries: a prospective clinical study.
Eur J Surg 1996, 162(11):853-860.
582. Buttenschoen K, Berger D, Strecker W, Buttenschoen DC, Stenzel K, Pieper T, Beger HG:
Association of endotoxemia and production of antibodies against endotoxins after multiple injuries. J
Trauma 2000, 48(5):918-923.
583. Dobke MK, Simoni J, Ninnemann JL, Garrett J, Harnar TJ: Endotoxemia after burn injury: effect of
early excision on circulating endotoxin levels. J Burn Care Rehabil 1989, 10(2):107-111.
584. Endo S, Inada K, Kikuchi M, Yamada Y, Kasoi T, Kondo M, Hoshi S, Yamashita H, Suzuki M,
Yoshida M: Are plasma endotoxin levels related to burn size and prognosis? Burns 1992, 18(6):486489.
585. Endo S, Inada K, Yamada Y, Takakuwa T, Kasai T, Nakae H, Yoshida M, Ceska M: Plasma
endotoxin and cytokine concentrations in patients with hemorrhagic shock. Crit Care Med 1994,
22(6):949-955.
586. Heggers JP, Goodheart R, Carino E, McCoy L, Ramirez RJ, Maness C: Is the limulus amebocyte
lysate the sole predictor of septic episodes in major thermal injuries? J Burn Care Rehabil 1998,
19(6):512-515.
Chapter 8: References
-203-
587. Hiki N, Berger D, Buttenschoen K, Boelke E, Seidelmann M, Strecker W, Kinzl L, Beger HG:
Endotoxemia and specific antibody behavior against different endotoxins following multiple injuries.
J Trauma 1995, 38(5):794-801.
588. Jiang J, Tian K, Chen H, Zhu P, Wang Z: Kinetics of plasma cytokines and its clinical significance
in patients with severe trauma. Chin Med J (Engl) 1997, 110(12):923-926.
589. Konigova R, Konickova Z, Bouska I: Clinical forms of endotoxemia in burns. Scand J Plast
Reconstr Surg 1979, 13(1):61-62.
590. McManus AT, Lindberg RB, Pruitt BA, Jr., Mason AD, Jr.: Association of endotoxemia and
bacteremia in burn patients: a prospective study. Prog Clin Biol Res 1979, 29:275-278.
591. Moore FA, Moore EE, Poggetti R, McAnena OJ, Peterson VM, Abernathy CM, Parsons PE: Gut
bacterial translocation via the portal vein: a clinical perspective with major torso trauma. J Trauma
1991, 31(5):629-636; discussion 636-628.
592. Roumen RM, Hendriks T, Wevers RA, Goris JA: Intestinal permeability after severe trauma and
hemorrhagic shock is increased without relation to septic complications. Arch Surg 1993,
128(4):453-457.
593. Rush BF, Jr., Sori AJ, Murphy TF, Smith S, Flanagan JJ, Jr., Machiedo GW: Endotoxemia and
bacteremia during hemorrhagic shock. The link between trauma and sepsis? Ann Surg 1988,
207(5):549-554.
Critical illness including ARDS
594. Maskin B, Fontan PA, Spinedi EG, Gammella D, Badolati A: Evaluation of endotoxin release and
cytokine production induced by antibiotics in patients with Gram-negative nosocomial pneumonia.
Crit Care Med 2002, 30(2):349-354.
595. McCartney AC, Piotrowicz BI, Edlin SE, Ledingham IM: Evaluation of the chromogenic Limulus
lysate assay in septic shock. Prog Clin Biol Res 1987, 231:459-474.
596. Parsons PE, Moore FA, Moore EE, Ikle DN, Henson PM, Worthen GS: Studies on the role of tumor
necrosis factor in adult respiratory distress syndrome. Am Rev Respir Dis 1992, 146(3):694-700.
597. Parsons PE, Worthen GS, Moore EE, Tate RM, Henson PM: The association of circulating
endotoxin with the development of the adult respiratory distress syndrome. Am Rev Respir Dis 1989,
140(2):294-301.
598. Pugin J, Auckenthaler R, Delaspre O, van Gessel E, Suter PM: Rapid diagnosis of gram negative
pneumonia by assay of endotoxin in bronchoalveolar lavage fluid. Thorax 1992, 47(7):547-549.
599. Raij L, Shapiro FL, Michael AF: Endotoxemia in febrile reactions during hemodialysis. Kidney Int
1973, 4(1):57-60.
600. Taniguchi T, Katsushima S, Lee K, Hidaka A, Konishi J, Ideguchi H, Kawaguchi Y: Endotoxemia in
patients on hemodialysis. Nephron 1990, 56(1):44-49.
601. Wardle EN: Endotoxinaemia and the pathogenesis of acute renal failure. Q J Med 1975,
44(174):389-398.
602. Wildfeuer A, Heymer B, Schleifer KH, Seidel HP, Haferkamp O: [Contribution to the diagnosis of
shock: demonstration of endotoxin and mucopeptide by the Limulus polyphemus-lysate test (author's
transl)]. Klin Wochenschr 1974, 52(4):175-178.
Pediatric and neonate
603. Goldberg PK, Kozinn PJ, Kodsi B, Lewitis R, Ackerman B, Aschenbrand L, Feldman F:
Endotoxemia and hyperbilirubinemia in the neonate. Am J Dis Child 1982, 136(9):845-848.
604. Goldstein JA, Reller LB, Wang WL: Limulus amebocyte lysate test in neonates. Am J Clin Pathol
1976, 66(6):1012-1015.
605. Hirthler M, Simoni J, Dickson M: Elevated levels of endotoxin, oxygen-derived free radicals, and
cytokines during extracorporeal membrane oxygenation. J Pediatr Surg 1992, 27(9):1199-1202.
606. Imura K, Fukui Y, Yagi M, Nakai S, Hasegawa T, Kawahara H, Kamata S, Okada A. Perioperative
change of plasma endotoxin levels in early infants. J Pediatr Surg. 1989 Dec;24(12) 1232-5.
-204-
Chapter 8: References
607. Magliulo E, Scevola D, Fumarola D: Limulus test in Escherichia coli enteritis. Lancet 1975,
1(7903):407.
608. Sauer M, Tiede K, Fuchs D, Gruhn B, Berger D, Zintl F: Procalcitonin, C-reactive protein, and
endotoxin after bone marrow transplantation: identification of children at high risk of morbidity and
mortality from sepsis. Bone Marrow Transplant 2003, 31(12):1137-1142.
609. Scheifele DW, Fussell S, Olsen E: Bacterial endotoxins in umbilical cord blood of neonates. Biol
Neonate 1984, 45(3):119-124.
610. Scheifele DW, Melton PW, Ebelt V: Endotoxinaemia per se as cause of neonatal morbidity. Lancet
1981, 1(8215):337.
611. Scheifele DW, Olsen E, Fussell S, Pendray M: Spontaneous endotoxinemia in premature infants:
correlations with oral feeding and bowel dysfunction. J Pediatr Gastroenterol Nutr 1985, 4(1):67-74.
612. Duffy LC, Zielezny MA, Carrion V, Griffiths E, Dryja D, Hilty M, Rook C, Morin F, 3rd:
Concordance of bacterial cultures with endotoxin and interleukin-6 in necrotizing enterocolitis. Dig
Dis Sci 1997, 42(2):359-365.
613. Taylor G, Stern C, Silverman M, Mearns M: Endotoxaemia in cystic fibrosis: response to antibiotics.
Arch Dis Child 1983, 58(10):826-828.
614. Willetts IE, Kite P, Barclay GR, Banks RE, Rumley A, Allgar V, Stringer MD: Endotoxin, cytokines
and lipid peroxides in children with intussusception. Br J Surg 2001, 88(6):878-883.
Sepsis and specific infections
615. Caridis DT, Reinhold RB, Woodruff PW, Fine J: Endotoxaemia in man. Lancet 1972, 1(7765):13811385.
616. Catchpole BN: Letter: Limulus assay for gram-negative endotoxin. Lancet 1974, 2(7874):225.
617. Dofferhoff AS, Bom VJ, van Ingen J, de Vries-Hospers HG, Hazenberg BP, van der Meer J, Mulder
PO, Weits J: Patterns of cytokines, plasma endotoxin, and acute phase proteins during the treatment
of severe sepsis in humans. Prog Clin Biol Res 1991, 367:43-54.
618. Fumarola D: Letter: The Limulus test and gram-negative bacillary sepsis. Am J Dis Child 1975,
129(5):644.
619. Endo S, Inada K, Inoue Y, Kuwata Y, Suzuki M, Yamashita H, Hoshi S, Yoshida M: Two types of
septic shock classified by the plasma levels of cytokines and endotoxin. Circ Shock 1992, 38(4):264274.
620. Fink PC, Lehr L, Urbaschek RM, Kozak J: Limulus amebocyte lysate test for endotoxemia:
investigations with a femtogram sensitive spectrophotometric assay. Klin Wochenschr 1981,
59(5):213-218.
621. Fossard DP, Kakkar VV: The Limulus test in experimental and clinical endotoxaemia. Br J Surg
1974, 61(10):798-804.
622. Galloway RE, Levin J, Butler T, Naff GB, Goldsmith GH, Saito H, Awoke S, Wallace CK:
Activation of protein mediators of inflammation and evidence for endotoxemia in Borrelia
recurrentis infection. Am J Med 1977, 63(6):933-938.
623. Gelfand JA, Elin RJ, Berry FW, Jr., Frank MM: Endotoxemia associated with the JarischHerxheimer reaction. N Engl J Med 1976, 295(4):211-213.
624. Glew RH, Levin J: Failure to demonstrate circulating endotoxin in malaria (38572). Proc Soc Exp
Biol Med 1975, 148(2):508-510.
625. Hamilton G, Hofbauer S, Hamilton B: Endotoxin, TNF-alpha, interleukin-6 and parameters of the
cellular immune system in patients with intraabdominal sepsis. Scand J Infect Dis 1992, 24(3):361368.
626. Kinsey SE, Machin SJ: Endotoxaemia in neutropenic patients. Lancet 1988, 2(8606):345.
Chapter 8: References
-205-
627. Koster F, Levin J, Walker L, Tung KS, Gilman RH, Rahaman MM, Majid MA, Islam S, Williams
RC, Jr.: Hemolytic-uremic syndrome after shigellosis. Relation to endotoxemia and circulating
immune complexes. N Engl J Med 1978, 298(17):927-933.
628. Rintala E, Pulkki K, Mertsola J, Nevalainen T, Nikoskelainen J: Endotoxin, interleukin-6 and
phospholipase-A2 as markers of sepsis in patients with hematological malignancies. Scand J Infect
Dis 1995, 27(1):39-43.
629. Roman-Marchant O, Orellana-Jimenez CE, De Backer D, Melot C, Vincent JL: Septic shock of early
or late onset: does it matter? Chest 2004, 126(1):173-178.
630. Shands JW, Jr., McKimmey C: Plasma endotoxin: increased levels in neutropenic patients do not
correlate with fever. J Infect Dis 1989, 159(4):777-780.
631. Tani T, Hanasawa K, Kodama M, Imaizumi H, Yonekawa M, Saito M, Ikeda T, Yagi Y, Takayama
K, Amano I et al: Correlation between plasma endotoxin, plasma cytokines, and plasminogen
activator inhibitor-1 activities in septic patients. World J Surg 2001, 25(5):660-668.
632. Tubbs H: Endotoxin in human and murine malaria. Trans R Soc Trop Med Hyg 1980, 74(1):121-123.
633. Tubbs HR: Endotoxin in meningococcal infections. Arch Dis Child 1980, 55(10):808-810.
634. van Deuren M, Santman FW, van Dalen R, Sauerwein RW, Span LF, van der Meer JW: Plasma and
whole blood exchange in meningococcal sepsis. Clin Infect Dis 1992, 15(3):424-430.
635. van Dissel JT, van Langevelde P, Westendorp RG, Kwappenberg K, Frolich M: Anti-inflammatory
cytokine profile and mortality in febrile patients. Lancet 1998, 351(9107):950-953.
636. Were T, Davenport GC, Hittner JB, Ouma C, Vulule JM, Ong'echa JM, Perkins DJ: Bacteremia in
Kenyan children presenting with malaria. J Clin Microbiol, 49(2):671-676.
Treatment studies
637. Bode C, Schafer C, Fukui H, Bode JC: Effect of treatment with paromomycin on endotoxemia in
patients with alcoholic liver disease--a double-blind, placebo-controlled trial. Alcohol Clin Exp Res
1997, 21(8):1367-1373.
638. Jackson SK, Parton J, Barnes RA, Poynton CH, Fegan C: Effect of IgM-enriched intravenous
immunoglobulin (Pentaglobin) on endotoxaemia and anti-endotoxin antibodies in bone marrow
transplantation. Eur J Clin Invest 1993, 23(9):540-545.
639. Munster AM, Moran KT, Thupari J, Allo M, Winchurch RA: Prophylactic intravenous
immunoglobulin replacement in high-risk burn patients. J Burn Care Rehabil 1987, 8(5):376-380.
640. Poynton CH, Jackson S, Fegan C, Barnes RA, Whittaker JA: Use of IgM enriched intravenous
immunoglobulin (Pentaglobin) in bone marrow transplantation. Bone Marrow Transplant 1992,
9(6):451-457.
641. Taggart DP, Sundaram S, McCartney C, Bowman A, McIntyre H, Courtney JM, Wheatley DJ:
Endotoxemia, complement, and white blood cell activation in cardiac surgery: a randomized trial of
laxatives and pulsatile perfusion. Ann Thorac Surg 1994, 57(2):376-382.
642. Wellmann W, Fink PC, Schmidt FW: Whole-gut irrigation as antiendotoxinaemic therapy in
inflammatory bowel disease. Hepatogastroenterology 1984, 31(2):91-93.
Other
643. Das J, Folkman J: Letter: Limulus test. N Engl J Med 1975, 293(21):1102-1103.
644. Fumarola D: [Preliminary observations on the Limulus amebocyte lysate test and relative
considerations]. G Batteriol Virol Immunol Microbiol 1973, 66(7-12):318-325.
645. Geerts SO, Nys M, De MP, Charpentier J, Albert A, Legrand V, Rompen EH. Systemic release of
endotoxins induced by gentle mastication association with periodontitis severity. J Periodontol. 2002
Jan;73(1) 73-8.
646. Goto T, Eden S, Nordenstam G, Sundh V, Svanborg-Eden C, Mattsby-Baltzer I: Endotoxin levels in
sera of elderly individuals. Clin Diagn Lab Immunol 1994, 1(6):684-688.
-206-
Chapter 8: References
647. Levin J: Endotoxin and endotoxemia. N Engl J Med 1973, 288(24):1297-1298.
648. Nadhazi Z, Takats A, Offenmuller K, Bertok L: Plasma endotoxin level of healthy donors. Acta
Microbiol Immunol Hung 2002, 49(1):151-157.
649. Nakao A, Yasui M, Kawagoe T, Tamura H, Tanaka S, Takagi H. False-positive endotoxemia
derives from gauze glucan after hepatectomy for hepatocellular carcinoma with cirrhosis.
Hepatogastroenterology. 1997 Sep-Oct;44(17) 1413-8.
650. Robinson JA, Klondnycky ML, Loeb HS, Racic MR, Gunnar RM: Endotoxin, prekallikrein,
complement and systemic vascular resistance. Sequential measurements in man. Am J Med 1975,
59(1):61-67.
651. Roth RI, Levin FC, Levin J: Optimization of detection of bacterial endotoxin in plasma with the
Limulus test. J Lab Clin Med 1990, 116(2):153-161.
652. Roth RI, Levin J, Behr S: A modified Limulus amebocyte lysate test with increased sensitivity for
detection of bacterial endotoxin. J Lab Clin Med 1989, 114(3):306-311.
653. Sturk A, Janssen ME, Muylaert FR, Joop K, Thomas LL, ten Cate JW: Endotoxin testing in blood.
Prog Clin Biol Res 1987, 231:371-385.
654. Sturk A, Joop K, ten Cate JW, Thomas LL: Optimalization of a chromogenic assay for endotoxin in
blood. Prog Clin Biol Res 1985, 189:117-137.
655. Thomas LL, Henny CP, Buller HR, Sturk A, Joop K, ten Cate JW: Chromogenic endotoxin
determination in blood--clinical relevance. Klin Wochenschr 1982, 60(14):759-760.
656. Thomas LL, Sturk A, Kahle LH, ten Cate JW: Quantitative endotoxin determination in blood with a
chromogenic substrate. Clin Chim Acta 1981, 116(1):63-68.
657. Valtonen M, Lommi J, Pohjola-Sintonen S, Ljungberg P, Hynninen M, Vaara M: Low levels of
cytokines and endotoxin in a fatal case of myocardial depression and septic shock due to Yersinia
pseudotuberculosis. Scand J Infect Dis 1995, 27(5):533-535.
658. Videm V, Svennevig JL, Fosse E, Mohr B, Aasen AO. Plasma endotoxin concentrations during
cardiac surgery may be related to atherosclerosis. Perfusion. 2000 Sep;15(5) 421-6.
659. Wildfeuer A, Heymer B, Schleifer KH, Haferkamp O: Investigations on the specificity of the
Limulus test for the detection of endotoxin. Appl Microbiol 1974, 28(5):867-871.
660. Young LS: Opsonizing antibodies, host factors, and the limulus assay for endotoxin. Infect Immun
1975, 12(1):88-92.
Metabolic and heart failure
661. Brock-Utne JG, Gaffin SL, Wells MT, Gathiram P, Sohar E, James MF, Morrell DF, Norman RJ:
Endotoxaemia in exhausted runners after a long-distance race. S Afr Med J 1988, 73(9):533-536.
662. Camus G, Poortmans J, Nys M, Deby-Dupont G, Duchateau J, Deby C, Lamy M: Mild
endotoxaemia and the inflammatory response induced by a marathon race. Clin Sci (Lond) 1997,
92(4):415-422.
663. Harte AL, Varma MC, Tripathi G, McGee KC, Al-Daghri NM, Al-Attas OS, Sabico S, O'Hare JP,
Ceriello A, Saravanan P et al: High fat intake leads to acute postprandial exposure to circulating
endotoxin in type 2 diabetic subjects. Diabetes Care, 35(2):375-382.
664. Lassenius MI, Pietilainen KH, Kaartinen K, Pussinen PJ, Syrjanen J, Forsblom C, Porsti I, Rissanen
A, Kaprio J, Mustonen J et al: Bacterial endotoxin activity in human serum is associated with
dyslipidemia, insulin resistance, obesity, and chronic inflammation. Diabetes Care, 34(8):1809-1815.
665. Niebauer J, Volk HD, Kemp M, Dominguez M, Schumann RR, Rauchhaus M, Poole-Wilson PA,
Coats AJ, Anker SD: Endotoxin and immune activation in chronic heart failure: a prospective cohort
study. Lancet 1999, 353(9167):1838-1842.
666. Pussinen PJ, Havulinna AS, Lehto M, Sundvall J, Salomaa V: Endotoxemia is associated with an
increased risk of incident diabetes. Diabetes Care, 34(2):392-397.
667. Rauchhaus M, Coats AJ, Anker SD: The endotoxin-lipoprotein hypothesis. Lancet 2000,
356(9233):930-933.
Chapter 8: References
-207-
668. Rauchhaus M, Doehner W, Francis DP, Davos C, Kemp M, Liebenthal C, Niebauer J, Hooper J,
Volk HD, Coats AJ et al: Plasma cytokine parameters and mortality in patients with chronic heart
failure. Circulation 2000, 102(25):3060-3067.
669. Zhu BL, Ishikawa T, Michiue T, Quan L, Maeda H: Postmortem serum endotoxin level in relation to
the causes of death. Leg Med (Tokyo) 2005, 7(2):103-109.
Veterinary and animals models
670. Breuhaus BA, DeGraves FJ: Plasma endotoxin concentrations in clinically normal and potentially
septic equine neonates. J Vet Intern Med 1993, 7(5):296-302.
671. Fessler JF, Bottoms GD, Coppoc GL, Gimarc S, Latshaw HS, Noble JK: Plasma endotoxin
concentrations in experimental and clinical equine subjects. Equine Vet J Suppl 1989(7):24-28.
672. Peek SF, Borah S, Semrad S, McGuirk S, Slack JA, Patton E, Coombs D, Lien L, Darien BJ: Plasma
endotoxin concentration in horses: a methods study. Vet Clin Pathol 2004, 33(1):29-31.
673. Takahashi T, Fujii H, Shimizu H, Omori E, Uehara K, Takeuchi M, Matsumi M, Yokoyama M,
Akagi R, Morita K: Site- and cell-type- specific induction of intestinal inducible nitric oxide
synthase in a rat model of endotoxemia. Med Chem 2005, 1(6):643-647.
674. Editorial. Endotoxaemia or endotoxinemia? Lancet 1992, 340:1323.
675. Falkow, S. Molecular Koch's postulates applied to microbial pathogenicity. Reviews of Infectious
Diseases (1988): 274-276.
676. Falkow, S. Molecular Koch's postulates applied to bacterial pathogenicity—a personal recollection
15 years later. Nature Reviews Microbiology 2.1 (2004): 67-72.
Statistical methods
677. Sweeney DA, Danner RL, Eichacker PQ, Natanson C: Once is not enough: clinical trials in sepsis.
Intensive Care Med 2008, 34(11):1955-1960.
678. Polderman KH, Girbes AR: Drug intervention trials in sepsis: divergent results. Lancet 2004,
363(9422):1721-1723.
679. Bellomo R, Warrillow SJ, Reade MC: Why we should be wary of single-center trials. Crit Care Med
2009, 37(12):3114-3119.
680. Borenstien M, Hedges LV, Higgins JPT, Rotherstien HR. Introduction to meta-analysis. 2009. Wiley
Interscience.
681. L'Abbe KA, Detsky AS, O'Rourke K: Meta-analysis in clinical research. Ann Intern Med 1987,
107:224-233.
682. Higgins JP, Thompson SG, Spiegelhalter DJ: A re-evaluation of random-effects meta-analysis. J R
Stat Soc Ser A Stat Soc 2009, 172(1):137-159.
683. Harris RJ, Bradburn MJ, Deeks JJ, Harbord RM, Altmamn DG, Sterne JAC (2008) Metan: fixed and
random effects-meta-analysis. Stata J 8:3-28
684. Moher D, Liberati A, Tetzlaff J, Altman DG, The PRISMA Group (2009) Preferred Reporting Items
for Systematic Reviews and Meta-Analyses: The PRISMA Statement. PLoS Med 6(7): e1000097.
doi:10.1371/journal.pmed.1000097.
685. Pearson K. Report on certain enteric fever inoculation statistics. BMJ 1904;3:1243-1246.
686. Higgins JP, Thompson SG, Deeks JJ, Altman DG: Measuring inconsistency in meta-analyses. BMJ
2003, 327(7414):557-560.
687. Song F: Exploring Heterogeneity in Meta-Analysis: Is the L'Abbé Plot Useful? J Clin Epidemiol
1999, 58:717-723
688. Ioannidis, J. (2010). Meta‐research: The art of getting it wrong. Research Synthesis Methods, 1(3
‐4), 169-184.
689. Wu, Y., Wang, F., Fan, X., Bao, R., Bo, L., Li, J., & Deng, X. (2012). Accuracy of plasma sTREM-1
for sepsis diagnosis in systemic inflammatory patients: a systematic review and meta-analysis.
Critical Care, 16(6), R229
-208-
Chapter 8: References
690. Whiting P, Harbord R, Kleijnen J: No role for quality scores in systematic reviews of diagnostic
accuracy studies. BMC Med Res Methodol 2005, 5:19.
691. Herbison P, Hay-Smith J, Gillespie WJ: Adjustment of meta-analyses on the basis of quality scores
should be abandoned. J Clin Epidemiol 2006, 59(12):1249-1256.
692. Collett D: Modelling Binary data. 2nd edition 2003. Chapman and Hall/CRC
693. Fleiss JL, Levin B, Paik MC: Statistical methods for rates and proportions. 3rd edition 2003. Wiley
Interscience.
694. Agresti A: Categorical data analysis. 2nd edition 2002 Wiley Interscience.
695. Cummings P: The relative merits of risk ratios and odds ratios. Arch Pediatr Adolesc Med 2009,
163(5):438-445.
696. Greenland S: Interpretation and choice of effect measures in epidemiologic analyses. Am J
Epidemiol 1987, 125(5):761-768.
697. Sharp SJ, Thompson SG: Analysing the relationship between treatment effect and underlying risk in
meta-analysis: comparison and development of approaches. Stat Med 2000, 19(23):3251-3274.
698. Sharp SJ, Thompson SG, Altman DG: The relation between treatment benefit and underlying risk in
meta-analysis. BMJ 1996, 313(7059):735-738.
699. Thompson SG, Smith TC, Sharp SJ: Investigating underlying risk as a source of heterogeneity in
meta-analysis. Stat Med 1997, 16(23):2741-2758.
700. Mutapi F, Roddam A (2002) p values for pathogens: statistical inference for infectious-disease data.
Lancet Infect Dis 2:219-230.
701. Kuss O, Gromann C: An exact test for meta-analysis with binary endpoints. Methods Inf Med 2007,
46(6):662-668.
702. Venkataraman G, Ananthanarayanan V: Demystifying "exact" logistic regression for pathologists. J
Clin Pathol 2008, 61(2):237-238.
703. Bradburn MJ, Deeks JJ, Berlin JA, Russell Localio A: Much ado about nothing: a comparison of the
performance of meta-analytical methods with rare events. Stat Med 2007, 26(1):53-77.
704. Sweeting MJ, Sutton AJ, Lambert PC: What to add to nothing? Use and avoidance of continuity
corrections in meta-analysis of sparse data. Stat Med 2004, 23(9):1351-1375.
705. Vandermeer B, Bialy L, Hooton N, Hartling L, Klassen TP, Johnston BC, Wiebe N: Meta-analyses
of safety data: a comparison of exact versus asymptotic methods. Stat Methods Med Res 2009,
18(4):421-432.
706. Clarke P: When can group level clustering be ignored? Multilevel models versus single-level models
with sparse data. J Epidemiol Community Health 2008, 62(8):752-758.
707. Hanley JA, Negassa A, Edwardes MD, Forrester JE: Statistical analysis of correlated data using
generalized estimating equations: an orientation. Am J Epidemiol 2003, 157(4):364-375.
Discussion
708. Woodruff PW, O'Carroll DI, Koizumi S, Fine J: Role of the intestinal flora in major trauma. J Infect
Dis 1973, 128:Suppl:290-294.
709. Klein DJ, Derzko A, Foster D, Seely AJ, Brunet F, Romaschin AD, Marshall JC: Daily variation in
endotoxin levels is associated with increased organ failure in critically ill patients. Shock 2007,
28(5):524-529.
710. Maury E, Blanchard HS, Chauvin P, Guglielminotti J, Alzieu M, Guidet B, Offenstadt G:
Circulating endotoxin and antiendotoxin antibodies during severe sepsis and septic shock. J Crit
Care 2003, 18:115-120
711. Koch T, Duncker HP, Axt R, Schiefer GH, Van Ackern K, Neuhof H (1993) Alterations of bacterial
clearance induced by endotoxin and tumor necrosis factor. Infect Immun 61:3143-3148
712. Young LS: Opsonizing antibodies, host factors, and the limulus assay for endotoxin. Infect Immun
1975, 12(1):88-92.
Chapter 8: References
-209-
713. Yahav D, Paul M, Fraser A, Sarid N, Leibovici L: Efficacy and safety of cefepime: a systematic
review and meta-analysis. Lancet Infect Dis 2007, 7(5):338-348.
714. Kim PW, Wu YT, Cooper C, Rochester G, Valappil T, Wang Y, Kornegay C, Nambiar S: Metaanalysis of a possible signal of increased mortality associated with cefepime use. Clin Infect Dis,
51(4):381-389.
715. Yahav D, Paul M, Fraser A, Sarid N, Leibovici L: Efficacy and safety of cefepime: a systematic
review and meta-analysis. Lancet Infect Dis 2007, 7(5):338-348.
716. Paul M, Yahav D, Fraser A, Leibovici L: Cefepime and all-cause mortality. Clin Infect Dis 2009,
49(4):640-641; author reply 641-642.
717. Leibovici L, Yahav D, Paul M: Meta-analysis of a possible signal of increased mortality associated
with cefepime use. Clin Infect Dis, 51(11):1350-1351; author reply 1351-1352.
718. Leibovici L, Yahav D, Paul M: Excess mortality related to cefepime. Lancet Infect Dis, 10(5):293294.
719. Seng, P., M. Drancourt, F. Gouriet, S. B. La, P. E. Fournier, J. M. Rolain, and D. Raoult. 2009.
Ongoing revolution in bacteriology: routine identification of bacteria by matrix-assisted laser
desorption ionization time-of-flight mass spectrometry. Clin. Infect. Dis. 49:543-551.
720. Lehmann, L E., et al. Cost and mortality prediction using polymerase chain reaction pathogen
detection in sepsis: evidence from three observational trials. Crit Care 14 (2010): R186.
721. Amaral, AC. Polymyxin B hemoperfusion and mortality in abdominal septic shock. JAMA,
2009:302(18), 1968-1969.
722. Vincent J-L. Polymyxin B hemoperfusion and mortality in abdominal septic shock. JAMA,
2009:302(18), 1968.
723. Gutsmann T, Razquin-Olazarán I, Kowalski I, Kaconis Y, Howe J, Bartels R. Brandenburg K.
(2010). New antiseptic peptides to protect against endotoxin-mediated shock. Antimicrobial agents
and chemotherapy, 54(9), 3817-3824.
724. Schuerholz T, Brandenburg K, Marx G. (2012). Antimicrobial peptides and their potential
application in inflammation and sepsis. Critical care (London, England), 16(2), 207-207.
725. Brandenburg K, Andrä J, Garidel P, Gutsmann T. (2011). Peptide-based treatment of sepsis. Applied
microbiology and biotechnology, 90(3), 799-808
726. Altemeier W A, Todd J C, Inge W W (1967). Gram-negative septicemia: a growing threat. Annals of
Surgery, 166(4), 530.
727. DuPont H L, Spink W W (1969). Infections due to gram-negative organisms: an analysis of 860
patients with bacteremia at the University of Minnesota Medical Center, 1958-1966. Medicine, 48(4),
307-332.
728. Bryant R E, Hood A F, Hood C E, Koenig M G. (1971). Factors affecting mortality of gram-negative
rod bacteremia. Archives of Internal Medicine, 127(1), 120.
729. Hodgin U G, Sanford J P. (1965). Gram-negative rod bacteremia: an analysis of 100 patients. The
American journal of medicine, 39(6), 952-960.
730. Watt P J, Okubadejo O A. (1967). Changes in incidence and aetiology of bacteraemia arising in
hospital practice. BMJ, 1(5534), 210-211.
-210-
7.3.
Acknowledgement of correspondence with authors
1. Dr D Berger, Klinikum der Universitat Ulm, Germany (1 page)
2. Dr P. Brandtzaeg, Ulleval University Hospital, Oslo, Norway (5 pages)
3. Dr B Byl, Erasme University Hospital, Belgium;
4. Dr R Danner, NIH, Bethesda, MD, USA (2 pages)
5. Dr P Engervall, Karolinska Hospital, Stockholm, Sweden (3 pages)
6. Dr KCH Fearon, Royal Infirmary, Edinburgh, UK (3 pages)
7. Dr EJ Giamarellos-Bourboulis, Laiko General Hospital, Athens (3 pages)
8. Dr B. Guidet, Hopital Saint-Antoine, Paris, France (2 pages)
9. Dr P Ketchum (Associates of Cape Cod) and Dr D Bates, Brigham and Women‘s
Hospital, Boston, MA, USA. (4 pages)
10. Dr J. Levin, VA Medical Center, San Francisco, USA (1 page)
11. Dr D. Massignon, Center Hospitalier Lyon Sud, France (3 pages)
12. Dr E Maury, Hopital Saint Antoine, Paris, France. (1 page)
13. Dr MJ McMahon, Leeds Teaching Hospital, Leeds, UK (1 page).
14. Pr G. Offenstadt, Hopital Saint Antoine, Paris, France. (3 pages)
15. Prof Steven M Opal, MD, Infectious Disease Division, The Alpert Medical
School of Brown University, Providence, Rhode Island;
16. Dr. Jan M. Prins, Academic Medical Center, Amsterdam (2 pages)
17. Dr JT van Dissel, Leiden University Medical Center, The Netherlands (6 pages)
18. Dr SM Willatts, Bristol royal Infirmary, Bristol, UK (1 page)
19. Dr M Yoshida, Jichi Medical School, Japan, (3 pages via Dr J Levin)
-211-
-212-
9. Appendix: the publications
Original research publications
(Contribution of Hurley JC to each publication is 100% unless stated otherwise)
Code Citation
JCM ‘94 Hurley JC. 1994. Endotoxemia: concordance with Gram negative
bacteremia in patients with gram negative sepsis: a reappraisal
with meta-analysis. J. Clin. Microbiol. 32:2120-2127.
JCM ‘95 Hurley JC. 1995. Bacteremia, endotoxemia and mortality in Gram
negative sepsis: a reappraisal with meta-analysis. J. Clin.
Microbiol. 33: 1278-82.
EIH&D ‘99 Hurley JC, & Levin J* (1999). Relevance of endotoxin detection in
sepsis. In Endotoxin in Health and Disease., Helmut Brade,
Stephen Opal Stefanie Vogel, David Morrison (editors), Marcel
Dekker Limited. 841 – 854. * Contribution of Hurley JC to
publication is 90%.
AP&LM ‘00 Hurley JC. (2000). Endotoxemia: concordance with gram negative
bacteremia: a meta – analysis using ROC curves. Arch. Pathol
Lab Med. 124: 1157 – 64.
JER ‘03 Hurley JC. (2003). Endotoxemia and Gram negative bacteremia as
predictors of outcome in sepsis: a meta-analysis using ROC
curves. J Endotoxin Res. 9; 271-279.
JCM ‘09 Hurley JC. (2009). The detection of endotoxemia with Gram negative
bacteremia is bacterial species dependent: A meta-analysis of
clinical studies. J. Clin. Microbiol. 47: 3826-3831
EJCM&ID Hurley JC. (2009). Does gram negative bacteremia occur without
endotoxemia? A meta-analysis using hierarchical summary ROC
‘10
curves. Eur. J. Clin. Microbiol. Infect Dis. 29: 207-215.
CC ‘12 Hurley JC, Guidet B, Offenstadt G, Maury E. (2012). Co-dependence
of endotoxemia, gram negative bacteremia and underlying risk as
predictors of mortality: a meta-analysis using L‘Abbé plots. Crit
Care 16: R148 * Contribution of Hurley JC to publication is
90%.
-213-
-214-
Background publications
Code Citation
Lancet ‘93 Hurley JC. (1993) Reappraisal of the role of endotoxin in the sepsis
syndrome (Medical hypothesis). Lancet 341: 1133-5.
CMR ‘95 Hurley JC. (1995) Endotoxin: methods of detection and clinical
correlates. Clin. Microbiol. Rev. 8:268-292.
Drugs ‘94 Hurley JC. (1994) Sepsis management and anti-endotoxin therapy
after nebacumab; A reappraisal. (leading article). Drugs 47:855861.
Drug Safety ‘95 Hurley JC. (1995) Antibiotic induced release of endotoxin: a
therapeutic paradox. Drug Safety. 12:183-195
EOID ‘95 Hurley JC. (1995) Endotoxaemia and novel therapies for the
treatment of sepsis. Expert Opinion Invest. Drugs. 4(3):163-174
EOID ‘96 Hurley JC. (1996). Meta analysis and investigation of anti –
infective therapies. Expert Opinion Invest. Drugs 6: 159 – 167.
JER ‘01 Hurley JC. (2003). Endotoxemia and Gram negative bacteremia as
predictors of outcome in sepsis: a meta-analysis using ROC
curves: call for data (letter). J Endotoxin Res. 7(6) 467.
AP&LM ‘11 Hurley JC. (2011). Meta-analysis of Clinical Studies of
Diagnostic Tests: Developments in How the Receiver
Operating Characteristic ―Works‖. Arch. Pathol Lab Med. 135:
1585-1590