St ckl et al.: Desirable routine analytical goals
157
Eur J Clin Chem Clin Biochem
1995; 33:157-169
© 1995 Walter de Gruyter & Co.
Berlin · New York
Desirable Routine Analytical Goals for Quantities Assayed in Serum
Discussion paper from the members of the External Quality Assessment (EQA)
Working Group A1) on analytical goals in laboratory medicine
By Dietmar St ckl1, Henk Baadenhuijsen2, Callum G. Fr ser*, Jean-Claude LibeetA, Per Hyloft Petersen5 and
Carmen Ricos6
1
2
3
4
5
6
Institut f r Standardisierung und Dokumentation im medizinischen Laboratorium e. V. (INSTAND e. V.),
D sseldorf, Germany
Academisch Ziekenhuis Sint Radboud, Nijmegen, The Netherlands
Department ofBiomedical Medicine, Ninewells Hospital and Medical School, Dundee, Scotland, U. K.
Instituut voor Hygiene en Epidemiologie, Klinische Biologie, Brussels, Belgium
Klinisk Kemisk Afdeling (Centrallaboratoriet), Odense Universitetshospital, Odense, Denmark
Servicio de Bioquimica, Hospital General Universitario Vall d'Hebron, Po Vall d1Hebron, Barcelona, Spain
(Received May 9/December 12, 1994)
Summary: The aim of the Working Group was to describe guidelines for deriving desirable analytical goals in
laboratory medicine. First, a literature review is given of the different approaches used until now, and some of the
most important studies are presented in detail. These approaches are then discussed critically, and the analytical
goals proposed by the group are outlined with respect to monitoring and diagnostic testing. The group recommends
that, most realistically, analytical quality specifications be biologically based. For diagnostic testing, the aim is
achievement of accuracy, allowing the use of common reference intervals when populations are homogeneous for a
given quantity. For monitoring (within an individual laboratory and performed with the same instrument), analytical
performance should aim at stable operation and low imprecision compared with the within-subject biological variation. Method accuracy is also very important for the comparability of results from different laboratories or instruments.
Foreword
patient treatment strategies. In view of the complex and
_.
.
- . „, , . ,
.,
.,
partly controversial nature of this topic, the literature
The aim of the Working Group was to describe guide- Λ
., .
, ,™
,
,
.
..
. _ . . , . f,
f · ,
, · ι ι
describing the different approaches used so rfar is rehnes for deriving desirable analytical
goals in laboratory
.
,
.
-,
.
^.^^j·,·
J
,. .
_ .~,
.,
..
viewed, and some of the most important studies are dist
Λ ·
medicme. Desirable means that analyses of the quality
^ · j * ·ι rr
*
·
ι
rr / -r r>
" * ,. ,
, ,
,. .,,
l - i cussed m detail [for recent reviews see also: Hyltoft Petdemanded by these goals have a negligible analytical
o r j j / i \ AT?
ο ττ /, /· η
/ο\ι
^
, „ ,.-i .
.
ersen & H0rder (1) and Fr ser & Hyltoft Petersen (2)1.
uncertaintyJ (due to method bias, unspecificity, or im- τ . , ,, ,
. , ,u . . ...
.
.
. '
.
^
.
'
It should be noted that in this review part, we cite the
precision)J for a Mquantity
,
,, u
.
.
,., .
J present in a specimen. The at^.
*
*
nomenclature used by the
respective authors which, m
tainment of such goals would increase measurement
^ ^ inconsistendes in wordi ^ terminol.
comp^rabuity between Afferent methods and laboratodevelopment of the Working Group's conD
nes, allowing& the use of common reference intervals and cept,. we soon realized
,. , that
., . no universally
.
,. U1 apn
applicable
proach could be established. Every application requires
!
) The Working Group is one of four ad hoc groups meeting under
the auspices of the European external quality assessment (EQA)
orgamzers group. The Working Groups were initiated by A. Uldall
following a meeting of EQA organizers and interested individuals
in Cracow, Poland, in 1991 at Eurolab '91.
Eur J Clin Chem Clin Biochem 1995; 33 (No 3)
special consideration. Analytical quality specifications
for
monitoring may be different from those for diagnos.
. _ - . . .
, .
,.0
tlc
testing. The situation in emergency analysis may differ from that in routine analysis (other techniques might
158
be preferable in the former dae to the necessity for a
short response time). Special population screening programmes may require detailed studies on the relation of
cost to analytical performance. Specific clinical strategies require discussions with the clinicians involved.
Therefore, the Working Group restricted itself to outlining general analytical goals for monitoring and diagnostic testing. These recommendations are presented in
detail here. The setting of analytical goals for clinical
strategies is, however, not discussed [for more information see I.e. (1)].
Stock! et al.: Desirable routine analytical goals
contributions to goal setting for quantitative measurements in laboratory medicine reflect both analytical and
medical points of view. Strategies for defining maximum analytical bias and imprecision include "state of
the art" of performance (7), opinions of experts (8, 9),
national recommendations (10, -Irl), needs for quality
control of analytical processes (12-16), proposals derived from the viewpoint of clinicians (17-22), analysis
of the effect of performance on interpretation in specific
clinical situations (1, 23-26), and derivation of criteria
based on intra- or inter-individual biological variation
(27-37).
Introduction
Analyses in laboratory medicine are only useful if measurements of quantities allow the evaluation of a person's state of health. This means that diagnosis or exclusion of a disease, detection of a person's predisposition
to a disease, or monitoring during medical treatment is
based on comparison of a value with a set of appropriate
population based reference intervals or with values previously observed from the same individual. Two general
aspects have to be considered when any kind of measurement is performed: a) the "medical requirements"
(clinical usefulness), i. e. what quality is needed for the
measurement from the medical point of view, and b) the
"analytical performance", i. e. how this desirable analytical quality can be specified, achieved, and controlled.
Factors connected with medical requirements are: within- and between-subject biological variations of a quantity, distribution of analytical results between sick and
healthy subpopulations, prevalence of disease, use of an
analysis (e.g. single point testing, patient follow up,
population screening), and time of analysis. The distribution of analytical results between sick and healthy
subpopulations and the prevalence of a disease determine the "a priori" relevance of an analytical test used
to decide whether a particular disease is present or not.
A widely used approach to assess the relevance of a
clinical test is the predictive value model which defines
the diagnostic performance of a laboratory test by its
diagnostic sensitivity, diagnostic specificity, predictive
value, and efficiency. This model and other approaches
are discussed in detail in recent articles by Linnet (3),
Büttner (4), Christensen (5), and Henderson (6). Since
they are not directly related to analytical goal setting,
they will not be outlined here. Analytical performance
is associated with: bias and imprecision of an analytical
procedure, analytical specificity and detection limit,
analysis costs, internal and external control, etc. Analytical performance and medical requirements both influence each other. Medical knowledge can guide analytical
strategies, whereas analytical progress can promote better understanding of biological processes. Therefore,
Description of the Different Strategies
Analytical performance and
"the state of the art"
According to some authors, the current state of the art
achieved in laboratory medicine is already sufficient (7).
Although this argument may be correct in many situations, it is untrue in many other situations, e. g. in certain
fields of immunochemical determinations, in the casojof
recently developed analytical techniques without thorough validation, or in cases where inherently superior
methods seem to disappear from the market because
faster and cheaper assays become available. This means
that the state of the art is not stable: it can change for
the better but it can also deteriorate. Therefore, clear
criteria should be defined concerning the analytical
quality that is desired in laboratory medicine. Where this
is impossible, at least criteria should be elaborated
which can aid in the distinction between good and poor
methods.
Analytical goals advocated by experts or
expert groups
Analytical goals advocated by experts (8, 9) are generally empirical in nature. They mostly reflect the experience gained in the special field in which the experts
are involved. They often include compromises taking
the state of the art very much into account and are not
real goals for desirable analytical quality. The latter is
especially true for the goals used by organizations responsible for external quality assessment or proficiency
testing (10, 11,38).
Analytical goals and the needs of
quality control
Discussion of analytical goals has become more and
more influenced by the need for the quality control of
analytical processes. Some general comments on quality
Eur J Clin Chem Clin Biochem 1995; 33 (No 3)
159
Stöckl et al.: Desirable routine analytical goals
control in laboratory medicine can be found in a series
of recommendations of the International Federation for
Clinical Chemistry (IFCC) (39, 40), and in an article by
de Verdier et al. (41). The general statement from the
viewpoint of quality control is that the better methods
are, the less effort and money has to be invested in their
control. On this basis, analytical imprecision expressed
as standard deviation (sa, see Annex for a list of abbreviations) of 1/4 of the total allowable error is recommended for internal quality control (14), which is half
that originally proposed by the same group (42). Results
exceeding the total allowable error indicate that the analytical method is out of control. Usually no estimates for
the total allowable error (TAE) are derived in such
works but the currently used strategies are assessed as
to whether they fulfil the needs for internal quality control (12, 13). In addition, no theoretical goals are derived
for method imprecision, but the state of the art is often
used as a basis for discussion, in order to estimate the
sa/TAE relation necessary for appropriate internal quality control. The benefit of these studies is that they provide a variety of statistical methods for evaluating
method performance in terms of bias and imprecision
(42-^44), they discuss the quality control procedures
necessary for specific analytical performance (15, 16),
and they compare these strategies with those used in
other scientific fields, e.g. industrial process management (14).
Analytical goals derived from the viewpoint
of clinicians
As discussed in the introduction, different types of medical decisions may be based on the results of analyses
done in laboratory medicine. Therefore, the necessary
quality of an analytical result depends both on the actual
clinical situation and on the physician who requested the
analysis. In some studies, attempts have therefore been
made to derive analytical goals from the viewpoint of
clinicians by sending questionnaires to experienced
physicians in many different fields of laboratory medicine (17—22). They were asked for decision levels for
common quantities in particular clinical situations or for
changes which would cause their medical action (medical significant differences = Amed). In the work of
Skendzel et al. (20), medically useful analytical coefficients of variation (CVa's) were calculated in the
following way for monitoring: CVa = Amed/[l.65x^2].
When a patient's value was compared with a reference
value, the calculation was as follows: CVa = Ame(/l-65.
Compared with other proposals, these coefficients of
variation were often quite large and were the subject of
severe criticism (45). On the other hand, it was stated
(13) that the medically significant differences reported
Eur J Clin Chem Clin Biochem 1995; 33 (No 3)
in the work of Skendzel et al. (20) were very useful,
provided that these values were interpreted so that they
reflect the sum of preanalytical (sp), within-subject biological (Sj), and analytical standard deviation (sa): st =
[Sp + sf 4- sa]1/2. If values for maximum analytical coefficients of variation were then calculated from the medically significant differences, they would be much lower
than those proposed by Skendzel et al. (20). On the other
hand, it is possible that the medical decision limits themselves are based on the experience of the physicians with
the performance of the analytical tests, and so more or
less reflect the analytical state of the art. In other words,
it may not be possible to derive analytical goals from
medical decision limits because they themselves may be
strongly influenced by analytical performance.
A n a l y t i c a l goals derived from specific
clinical situations
Every clinical situation requires a specific analytical approach. Therefore, it is very difficult to establish a uniform concept for the estimation of allowable imprecision
and bias. Examples of specific situations are: determination of haemoglobin AI C for long term diabetic control
(25), determination of theophylline in serum for therapeutic drug monitoring (26), determination of creatine
kinase isoenzyme in the diagnosis of acute myocardial
infarction, determination of blood thyroid-stimulating
hormone in screening for congenital hypothyroidism,
and monitoring cholesterol as a measure for the risk of
coronary heart disease (23). Further applications can be
found in a recent review by Hyltoft Petersen & Border
(1). In a recent article, Schectman & Sasse (46) investigated the influence of analytical imprecision on the
medical usefulness of lipid measurements. The interpretation of the analytical performance was based on the
use of clinical cut-off points or comparison with desired
concentrations. All of these approaches have in common
that they evaluate the clinical consequences caused by
the analytical imprecision and bias either together or independently. Many different statistical approaches are
used to calculate the effects at different levels of complexity: e. g. sum of misclassifications, the number of
false negative results, or economical optimization. These
approaches usually do not generate general figures for
analytical bias and imprecision, but they present models
for optimizing the analytical performance for the specific clinical situations. Therefore, they are more or less
restricted to specialists in the field. Nevertheless, the statistical background outlined in these studies can also be
used for the optimization of analytical methods in other
applications.
160
A similar approach, which mainly discusses overlap of
healthy and sick subpopulations and prevalence of diseases in the context of analytical needs is shown by Lott
et al. (47). They discuss the effect of analytical imprecision on the percentage of correct diagnoses at different levels of prevalence and population overlap in
three different applications, namely, population screening, long term, and short term trend detection.
A general theory for the allowable analytical coefficient
of variation in situations that specifically involve the
monitoring of individuals was proposed recently (24).
The following formula was derived: CVa < [(Δ2/2Ζ2)
- CV?]1/2, where Ac is the percentage of change with
clinical significance in monitoring serial results from an
individual, Ζ is a statistical value for the probability of
significance, and CVj is the within-subject biological coefficient of variation. A critical point of this approach is
the estimation of the Ac value. In the work mentioned
above, Ac was often derived from the viewpoint of clinicians (questionnaires), which perhaps does not reflect
the desirable analytical quality but might be much influenced by the analytical state of the art as mentioned
above.
St ckl et al.: Desirable routine analytical goals
Following this hypothesis, data on biological variation
were collected and goals for a variety of quantities were
agreed on at two international conferences (30, 31). This
data base was expanded and collected in a recent review
(49), although data on individual within-subject biological variations have only be describpd for a few quantities
(33, 50). It was further recommended (31) that the composite biological variation be used for group screening,
and the within-subject biological variation for single individual testing.
The approach of relating analytical performance to the
biological variation was very much inspired by the ideas
of Tonks (27). He empirically stated that the total allowable error (TAE) should not exceed one quarter of the
reference interval (R), using the following formula: TAE
= ± l/4[of the reference interval]/[mean of the reference
interval] X 100%.
The concepts outlined above mainly deal with method
imprecision, but it has become increasingly apparent
that method bias (expressed as the deviation from the
"true value") should be included when analytical goals
in laboratory medicine are derived. Therfore, Harris (51)
expanded his original work (32) so that bias (B) and
imprecision were both taken into account. The original
equation: sa ^ 0.5 Si (sc) was thus changed into
[s^ + B2]172 < 0.5 Si (sc). In addition, the author focused
in this work on diagnostic testing, especially when comparing patients' data with a reference interval of the
healthy subpopulation, an idea originally proposed by
Tonks (27) and later by Click (29) [a series of reeommendations on the theory of reference values was recently published by the IFCC, see I. c. (52) and articles
cited therein]. As stated before, the reference interval
reflects the within- and between-subject biological variation and, if the 95% interval is chosen, then R = 4 sc.
The formula for the allowable standard deviation (taking
into account method bias) of an analytical method for
single diagnostic testing becomes: sa ^ R[l/80 — 4/5
X B2/R2]172, or B < 0.125 R, if sa is zero. As a rule of
thumb, he still proposed sa ^ 0.5 Sj for single serial testing, and sa < 0.1 R for single diagnostic testing in the
absence of method bias.
Cotlove et al. (28) proposed that analytical goals be
based not on the reference interval, because this reflects
both the analytical and the biological variation (48), but
on the composite biological standard deviation (sc),
which includes the within- (s,·) and the between-subject
or group standard deviation (sg). They stated that the
analytical standard deviation should not exceed 0.5
times the composite biological standard deviation:
sa ^ 0.5 sc, with sc = [s? + s2]172, because then the analytical imprecision adds only 12% to the total test result
variability: st = [(0.5 sc)2 + s2]172 = 1.12 sc. In fact, this
formulation is similar to the one originally proposed by
Tonks (27) because the total allowable error defined by
him was meant as 2 sa. In addition, 95% of the reference
interval represents the mean ± 2 sc = 4 sc, if corrected
for the analytical imprecision. Tonks' formula can then
be written as: 2 sa < 0.25 X 4 X sc, which is the same
as the formula derived by Cotlove et al.
Ross (53) uses the preanalytical test diagnostic efficiency as a starting point. He considers in particular the
case in which the best performance is needed. This is
the case when the means of the healthy and sick subpopulations are only 1.88 SD's apart. The preanalytical
diagnostic efficiency is then 82.5%, which means that
82.5% true diagnostic statements can be made in the
ideal case. In that situation, he examines the loss of
diagnostic efficiency depending on the bias or the imprecision of the analytical method used, the test population variability and the probability for false rejection.
The output of the model is the limit for allowable analytical error of a single result for a quantity in the clinical
population under study. The advantage of this model is
that it not only relates method bias and imprecision to
the biological variance, but it additionally shows how
both variables affect the confidence level of medical decisions. The disadvantage is its limitation to the situation
Analytical goals based on within- and
between-subject biological variation
Eur J din Chem Clin Biochem 1995; 33 (No 3)
161
Stöckl et al.: Desirable routine analytical goals
of a bimodal distribution with a 50% prevalence where
the means are 1.88 SD's apart from each other. For individual single testing he proposes sa ^ 0.64 Sj. In a recent
article (54), this concept was expanded and sa < 0.5 sj
and B < 0.25 to 0.33 Si were advocated. Further, total
allowable analytical error should not exceed 1.25 to
1.40 times Sj.
Gowans et al. (35) based goals on the recommendations
of the IFCC regarding reference intervals (52). They derived functions which allow the investigation of the influence of analytical bias and analytical imprecision
(separately and in combination) on the percentage of the
population outside reference limits. For a population
sample size of 120, the IFCC stated that the maximum
acceptable percentage outside each reference limit
should be 4.6%. From this, a maximum analytical standard deviation (with no bias) of sa ^ 0.58 sc, and a maximum bias (with no imprecision) of B ^ 0.25 sc is calculated.
A similar approach was proposed by Stamm (55). For
diagnostic testing he proposed a maximum method bias
of 0.33 sg together with a maximum standard deviation
of 0.33 sg, in order to keep the sum of false classifications below 5% on each side of the reference interval.
For monitoring, on the other hand, only limits for
method imprecision were proposed. Further, these limits
were not directly related to the biological variation, but
were expressed as 1/4 of a predefined medical decision
limit.
Klee (56) investigated the influence of imprecision and
bias of a test on its clinical specificity. He defined an
analytical "error budget" as the squared sums of the imprecision and bias errors. The maximum limit for this
error budget was set at a value corresponding to a 50%
increase in the false-positive rate for classifying healthy
subjects. For Gaussian distributions with ± 2 s used as
decision limits, this budget equates to 0.45 sc. He recommended that this budget be allocated as 0.36 sc for bias,
and 0.18 se for the analytical standard deviation (these
fractions do not add up to 0.45 because bias and imprecision are not additive; the bias has a greater influence on false classification than imprecision).
A striking feature is the fact that all of the individual
approaches described above recommend numbers for
analytical standard deviation near or equal to 0.5 times
the biological standard deviation. In addition, the later
studies split this number into portions for imprecision
and bias, if both are present (35, 51, 55, 56); only
Ross & Fräser (54) use the sum of the numbers
derived for maximum bias and imprecision, if both
are present.
Eur J Clin Chem Clin Biochem 1995; 33 (No 3)
Modern Concepts for Assessing Method Bias
As shown above, only the newer concepts relating analytical performance to medical needs discuss the influence of bias on the medical usefulness of the analytical
result. The reason for the late introduction of the bias
component in the discussions on desirable performance
standards in laboratory medicine is that for a long time
only a limited accuracy base existed. This changed,
when so called "Definitive Methods" were developed,
which are the "best approximation to the true value"
of a quality. Based on this approach, a comprehensive
measurement system in laboratory medicine was proposed by Tietz (57). An important role in this concept is
played by the "Reference Methods", which are intended
to transfer the accuracy base of the Definitive Methods
to the routine methods (58—60). The bias of methods is
of particular importance when intermethod or interlaboratory comparability is required. Both features are primarily assessed in external quality assessment (EQA)
surveys and therefore the approach of Tietz was recommended early in this field (9, 61—63). Comparison of
routine methods in laboratory medicine with accuracy
based methods have been published (64—66) and meanwhile some countries use an accuracy base for performance evaluation for a variety of quantities in their internal and external quality assessment programmes (10,
67, 68).
Comparison of the Different Approaches
In table 1, some examples of the maximum medically
useful analytical coefficients of variation proposed in the
literature for single individual testing are compared with
Tab. 1 Maximum medically useful analytical coefficients of variation proposed in the literature for single individual testing and
analytical "state of the art".
Quantity
S-Sodium
S-Chloride
S-Calcium
S-Protein
S-Glucose
S-Creatinine
S-Potassium
S-ChoIesterol
S-Phosphate
S-Uric acid
S-Urea
Proposed analytical coefficient of variation
from references*
(69)
(30)
(10)
0.9
1.2
1.5
1.5
2.1
2.0
1.4
2.6
2.2
2.2
2.5
0.3
0.7
0.9
1.4
2.2
2.2
2.4
2.7
4.0
4.2
6.3
2.0
2.0
3.3
3.0
5.0
6.0
2.7
6.0
5.0
6.0
8.0
(9)
1.1
1.5
1.5
3.3
3.0
6.7
5.0
6.0
5.0
6.3
10.0
(20)a
(27)
1.7
—
4.8
8.3
0.9
1.0
—
3.5
5.0
—
—
5.0
5.0
—
5.0
11.2
10.1
4.8
12.3
14.3
—
12.2
* Median CVa obtained from the Netherlands external/internal
quality assessment scheme (69); Aspen conference (30); German
Guidelines (10); Gilbert (9); Skendzel et al. (20), "whenever two
values are given, the lower is listed; Tonks (27).
162
the state of the art as derived-from external quality assessment data (69). The concepts related to these examples were discussed earlier in the review part. In all
cases the between day analytical coefficient of variation
is considered. The data in columns 4 + 5 refer to medical decision limits as specified in the respective publications, while the data in columns 1 -3 were given by the
respective authors without restriction to specific concentrations. In this way, the decision levels are included in
the latter, which allows the comparison of both types of
data. The data of Tonks (27) in column 6 are shown only
for historical reasons. With the exception of S-urea, the
goals derived from the within-subject biological variation are the most stringent (column 2). The most liberal
goals are those obtained from physicians' interviews
(column 5). Between them lie the goals advocated by
experts and national recommendations (columns 3 + 4).
Interestingly, most of these follow in general the demands given by the biological variation of the quantities,
e. g. for S-sodium the proposed CV are always the lowest ones. For S-uric acid or S-urea usually quite high CV
are tolerated. But there are examples where quantities
with comparable biological variations show pronounced
differences, e. g. S-creatinine and S-potassium. The values for S-potassium are considerably lower (see columns
3—5). This could be due to the fact that many methods
for S-creatinine today still have accuracy and precision
problems, while S-potassium normally is measured with
methods of higher precision. It seems that the goals for
S-creatinine and S-potassium derived by experts, external quality assessment organizers or questionnaires of
clinicians reflect more or less the analytical state of the
art for these two analytes. The comparison of the state
of the art CV (69) (column 1) with the most stringent
goals proposed at the Aspen Conference (30) (column 2)
shows that the analytical precision is now satisfactory
for most of the presented quantities, provided the methods have no bias. Exceptions are analytes such as Ssodium, S-chloride and S-calcium, for which the biological variation is very small. An interesting feature in the
numbers advocated by Tonks (27) (column 6) is the tendency to set an upper imprecision limit, even if biology
would allow higher values. This might reflect the fact
that different quantities are often measured with similar
methods and equipment, which results in similar imprecision.
Comments of the Working Group on the Approaches
in the Literature for Deriving Analytical Quality
Specification from:
(i) The state of the art
Advantages of this concept are that analytical quality
specifications can easily be established and most
Stöckl et al.: Desirable routine analytical goals
laboratories will accept them. Disadvantages are that
they might not be related to optimal patient care and that
these analytical quality specifications change with time.
The state of the art can improve, but it can also deteriorate.
(ii) The views of experts and expert groups
Advantages are that they can be easily established on a
local basis and will gain a high acceptance in laboratories. Disadvantages are that these analytical quality
specifications often cannot be transferred to a broader
community and that they often are influenced by the
analytical state of the art.
(iii) Opinions of clinicians
The advantages of this concept are that it may be related
to the clinicians' perception when to react. In addition,
the analytical quality specifications derived represent
more or less the situation used in practice. Disadvantages are that often different opinions among clinicians
exist and that it is often impossible to fix the probability
level for significance of results. In addition, the analytical quality specifications derived might be very much
influenced by the analytical state of the art. The latter
will be discussed in more detail below. Some studies
made an attempt to derive analytical quality specifications from the viewpoint of clinicians by sending questionnaires to experienced physicians in many different
fields of laboratory medicine (17—22). They were asked
for decision levels for common quantities in particular
clinical situations or for changes in concentrations
which would cause their medical action. It was shown
that the decision limits given in the work of Skendzel et
al. (20) will cause < 5% false positive alarms for most
of the common quantities, when the observed analytical
results exceed these action limits, provided the results
were obtained with state of the art analytical performance (13). This supports the conclusion that these action limits might be influenced to a great extent by the
analytical state of the art. But is was also stated that the
medical significant differences may be too narrow for
some quantities, because the clinicians seem to underestimate the biological variation of those quantities (13).
In addition, it can be questioned, whether others are not
unnecessarily high. To investigate this problem in more
detail, the within-subject biological variation (CVi), the
analytical variation (CVa), and the medical decision values must be viewed in relation to each other. Table 2
compares the quotients of CVa and CVi5 Amed and CVj
and CVj for some of the common quantities [they are
arranged by descending CVa/CVi values, and the data
of Linnet (13) are used]. From;these calculations it is
Eur J Clin Chem Clin Biochem 1995; 33 (No 3)
163
Stöckl et al.: Desirable routine analytical goals
Tab. 2 Comparison of CVa/CVi$ Amed/CVj, and CV} of 17 common quantities arranged by descending CVa/CVj values [the data
of 1. c. (13) are used for the calculation].
Quantity
CV0/
CVj
S-Sodium
S-Creatinine
S-Calcium
S-Thyroxine
Prothrombin time
S-Glucose
S-Cholesterol
S-Total protein
S-Phosphate
S-Potassium
S-Bilirubin
S-Aspartate aminotransferase
S-Urea
Haemoglobin
S-Triacylglycerols
Leukocyte count
S-Iron
1.49
1.39
1.33
1.18
0.85
0.79
0.79
0.74
0.60
0.48
0.48
0.43
0.37
0.24
0.21
0.17
0.12
&
4.5
5.89
5.39
5.0
5.58
5.09
4.23
4.57
4.43
2.70
1.87
2.39
2.48
1.69
1.02
1.71
1.09
CVj
0.8
4.4
1.8
7.6
4.5
4.4
4.8
3.0
5.8
4.4
23
14
12.4
4.5
26
Advantages of this approach are that generally analytical
imprecision and bias are evaluated. In addition, the goals
derived are directly related to the clinical use. They are
most useful when one particular test is used for one specific clinical situation. Disadvantages are that they are
complicated, often require much effort for their production, mostly can be used only on a local basis, and
that the clinical strategy must be clearly defined. In addition, this approach lacks practicability for laboratory
tests that are used for many different clinical situations.
Nevertheless, the group advocates this approach wherever it can be applied. A detailed description of its application in various clinical strategies can be found in Hyltoft Petersen & H0rder (1).
14.7
26
surprising that the quantities with a low (< 0.5) CVa/
CVj value (which indicates sufficient analytical
precision) also have a low Amed/CVj value (the "a priori"
least medical significant difference which analytically
can be detected for two consecutive measurements
(P = 0.05) is: Amcd = 2.77 X CVi? which would result
in an ideal Amed/CVi value of 2.77). It seems that the
medical decision limits have been chosen to be quite
narrow despite the fact that these quantities mostly have
a great biological variation. In other words, this varia-^
tion seems to be underestimated by clinicians with the
consequence that the number of false positive results is
high (13). On the other hand, many quantities with a
high (> 0.5) CVa/CVi value (which indicates insufficient analytical precision) have high Amed/CVi values.
In these cases, it can be questioned whether the medical
decision limits should not be lowered. Maybe the poor
analytical imprecision has been overestimated here by
the clinicians. A result of this is a relatively high
number of false negative results. Therefore, analytical
goal setting on the basis of the viewpoint of clinicians
seems to be more influenced by "the state of the art"
than by what is desirable. In addition, it may lead to
incongruent decision limits when different quantities
are compared.
(iv) Analytical goals and the needs of quality
control
As stated above, this approach usually uses only currently defined goals and investigates whether they are in
accordance with the needs for quality control.
Eur J Clin Chem Clin Biochem 1995; 33 (No 3)
(v) Analytical goals derived from specific
clinical situations
(vi) Analytical goals based on within- and
between-subject biological variation
The basis of this concept is that the mean within-subject
biological variation is nearly constant in all investigations (49) including the elderly (70) and groups of patients in certain stable states (49). This allows the
transfer of data over time and geography. Further, where
the population is homogeneous for a quantity, common
reference intervals can be used within large regions or
within ethnic groups. This can save considerable resources in laboratories which otherwise have to establish
their own reference intervals. In addition, the concept of
common "reference changes" (33) could be applied in
the whole of Europe (and even more widely). For a more
recent article on the concept of reference changes see
Queralto et al. (50).
The Concept of the Working Group
Clarification of the terms "goal" and
"desirable analytical performance standard"
Sometimes confusion arises over the use of the word
"goal" for both medical and analytical aims; in fact they
should be clearly separated. A medical goal is, e. g. the
use of common reference intervals. For this medical
goal, a "desirable analytical performance standard"
(= "analytical goal") must be derived. This desirable analytical performance standard must then be specified in
terms of maximum method bias and imprecision. When
achieved, these performance standards should guarantee
optimal patient care. To illustrate this way of thinking,
Scheme 1 presents some analytical goals for different
application levels (we use the term "analytical goal" in
the following always in the sense of "desirable performance standards", as described above). Two main levels
Stockt et al.: Desirable routine analytical goals
164
Scheme 1 Goals and related desirable analytical performance standards at different levels
Level
Goal
Achieving
medical utility
1. Make use of:
a> Common reference changes
b. Common reference intervals
Accreditation
Desirable performance standards
In the absence of systematic changes: sa ^ 0.5 S{
(see Annex for detailed discussion)
When imprecision is negligible: B ^ 0.25 sj
(see Annex for detailed discussion)
2. Analytical performance should not
influence the outcome of clinical
strategies
The demands implied for desirable performance should be evaluated
for each clinical strategy according to accepted models for the
specific purpose (diagnosis, screening, monitoring, etc.)
Guard against unprofessional
performance
The specifications must consider the analytical state of the art, but
should gradually be converted in accordance with the above
recommendations.
are shown in Scheme 1, which indicate differences in
approach. Achieving medical utility is always related to
biology whereas accreditation systems are mostly related to the actual analytical performance attained in
laboratories. The latter uses goals from biological variation only if convenient.This is practical, since it reflects
what is possible with the methods and equipment in use,
but it may not reflect the medical utility. It is obvious
that also the EQA of accreditation/licensing testing has
to consider the state of the art performance, but a gradual conversion of systems should be undertaken in accordance with medical utility concepts whenever possible.
Recommendations of the Working Group
From the above discussions, the Working Group recommends that analytical quality specifications be based on
biology as a realistic basis, and that the approaches (i)
to (iii) are reasonable only for providing provisional solutions: the first in assessing surveys and the other two
as interesting approaches in individual laboratories. The
starting point for our work was, therefore, the concept
of Harris (32) and Cotlove et al. (28) for patient monitoring: sa < 0.5 Si, and the approach of Gowans et al.
(35) for diagnostic testing: B < 0.25 sc. We expanded
the concept of Cotlove et al. (28) because, during monitoring, unidirectional systematic changes (Δ8Ε) (=
drifts) can occur, and we accepted the concept of Gowans et al. (35), who based their goals on the recommendations of the IFCC for the establishment of reference
intervals (52). This concept is especially useful for diagnostic testing. A detailed discussion of the derivation
of analytical quality specifications for monitoring and
diagnostic testing can be found in the Annex. In particular, the combined effect of bias and imprecision is presented in a graphical form. In short, we recommend that
desirable performance standards should be calculated
for monitoring as:
sa < 0.5 Si
(in the absence of unidirectional
changes), or
systematic
SE ^ 0.33 Si (when imprecision is negligible);
see also Annex;
for diagnostic testing as:
B ^ 0.25 sc (when the imprecision is negligible),
or
sa ^ 0.58 sc (when bias is negligible);
see also Annex.
Data on biological variation for most of the common
serum quantities are available (49, 71-73) and the desired performance standards can be calculated as shown
in table 3 (74). If data for sg are not available, an approximation for bias can be 1/16 of the range of the reference
Tab. 3 Specifications for desirable analytical performance (from
I.e. (74))
Quantity
Maximum CVa in
the absence of
syst. changes
(using CVa
^ 0.5 CVj)
S-Sodium
S-Calcium
S-Creatinine
S-IgM
S-Potassium
S-Transferrin
S-Cholesterol
S-Lactate
S-Phosphate
S-Bilirubin
S-Triacylglycerols
S-Alanine
aminotransferase
0.3
0.9
2.2
2.3
2.4
2.4
2.7
3.9
4.0
Maximum bias
(%)ifthe
imprecision
is negligible
(using B
<0.25CV C )
0.2
0.7
2.8
12.7
11.3
1.6
2.3
4.1
4.1
3.1
9.8
13.6
15,6
13.6
Π.6
-
Eur J Clin Chem Clin Biochern 1995; 33 (No 3)
Stöckl et al.: Desirable routine analytical goals
165
interval (this assumes that sa is small in relation to Sj
and sg). We recommend that these desirable performance
standards should also be introduced in external quality
assessment as the basis for calculation of acceptance
limits. When more demanding specifications based on
documented medical needs are stated, these may be substituted for those based on biological variation. The
goals outlined above may be viewed as aims for the
future if they cannot be reached with current analytical
equipment and methodology. Last, but not least, the
group recommends that limits for interference and unspecificity should be related to the biological variation
of the respective quantity (75). This is more relevant
than the fixed limits used previously [e.g. 5% for
electrolytes and substrates and 10% for enzymes (76);
or 3% (77), and 10% (78) for all quantities]. Further, the
group recommends the use of goals derived from specific clinical situations (1, 79) whenever they are more
appropriate.
monitored, because densities of solutions, and therefore their concentrations, change with temperature.
This has an effect of about 0.12% within a temperature
change of 5 degrees. In addition buoyancy corrections
should be made for weighings, and so on (80). Another consequence would be the necessity to keep
preanalytical errors below 0.5%, otherwise the final
result would be dominated by the preanalytical errors.
Taking all these sources of error together, routine
method inaccuracies below 1% are extremely difficult
to achieve, it may even not be possible to assess them
and they might be outweighed by the preanalytical
errors. Therefore, accuracy seems to be a major challenge for future developments.
Comparison of the Recommendations
with the Actual Situation
a) the model is simple to understand and apply,
It should be noted that, for a number of quantities, the
"state of the art" analytical precision is better than the
goals. Consider for example S-triacylglycerols. The proposed maximum coefficient of variation was 11.6% (see
tab. 3). If unidirectional systematic changes are also
taken into account, it must be reduced to maybe 6 or
8%. Much lower values are attainable today. Therefore,
in the opinion of the Working Group, it would neither
be wise nor reasonable if a new triacylglycerol method
with a CV in the range of 6% was introduced on the
market. First, it must not be forgotten that an analytical
standard deviation of 0.5 Sj, still adds 12% variation to
the true test variation. In addition, it generally can be
stated that the better a method performs, the less time
and money has to be invested in its control (14). Therefore, in the above cited example, more stringent criteria
than those derived from the biological variation are advisable. On the other hand, there are many quantities
whose analytical precision is insufficient according to
the above cited concept, especially when method bias is
taken into account. Insufficient analytical precision is
normally associated with quantities which have narrow
biological variation like S-sodium, S-chloride, Scalcium or S-protein, despite the attainment of quite low
CV values for these quantities. Therefore, in these
cases, performing more than one measurement might
be more effective than aiming for a CV below 0.5%.
Then less stringent criteria than those derived above
could be accepted. Even more difficult to achieve is a
method bias below 0.5%, because then the temperature
during analysis must be kept constant or be constantly
Eur J Clin Chem Clin Biochem 1995; 33 (No 3)
Conclusion
The proposed "desirable performance standards" are
based on within- and between-subject biological variation for several reasons, namely
b) there are many data on biological variation,
c) the within- and between-subject biological variations
are nearly constant over geography, and
d) when these criteria are fulfilled, the analytical quality
will satisfy most clinical needs.
Furthermore, it is often difficult to evaluate clinical
needs in complicated clinical situations in which other
considerations may be more relevant in the process of
decision-making. However, when more demanding clinically derived specifications are stated, these might be
substituted for those based on biological variation. This
may be the case for S-cholesterol (tab. 3) for which the
National Cholesterol Education Program (11) recommended a bias of less than 3%. Further, we recommend
that where the state of the art is much better than the
desirable performance standards derived from the biological variation, this should be welcomed because it
gives the user additional benefits, especially in the internal quality control of such methods.
ANNEX
Detailed Concept of the Working Group for Deriving
Analytical Quality Specifications from Biology
Abbreviations used:
Within-subject (individual) biological standard
deviation:
Si
St ckl et al.: Desirable routine analytical goals
166
Between-subject (group) biological standard deviation:
s
g
Composite biological standard deviation:
Sc = (Sf + S|)"2
Analytical imprecision expressed as standard deviation:
sa
Coefficient of variation:
CV (the same subscripts as above are used, e. g. CVi)
Bias: B
Drift (unidirectional changes in bias):
ASE
Shift (difference in bias when two methods are used for
monitoring the same patient, see I.e. (81)):
Δ8Ε
Reference interval (for normal distributed quantity):
Κ ( = μ ± 1.96s)
Medically significant difference (reference change):
General assumptions underlying the concepts
We wanted to develop a concept to be generally applicable in the routine laboratory. Consider, for example, the
case of a potassium test to be used in a general medical
laboratory. The test should cover potassium determination in plasma in emergency situations, such as digitalis
poisoning, heart arrhythmias, acute renal failure, or diarrhoea; in therapy decisions, such as treatment with angiotensin-converting enzyme blocker, digitalis, or diuretics; or in monitoring illness, such as diabetes mellitus, chronic renal failure, or gastrointestinal diseases.
These are situations, which today, can normally all be
covered by one and the same test. Developing general
goals for such a test means that the proposed goals
should be independent of the concentration of potassium present in a specimen. It also follows that we
always considered a Gaussian distribution and 5% testlevel. Therefore, these goals are not ideal because they
are not the best choice for each particular application.
But, on the other hand, it is our conviction that they
are especially useful for the daily routine operation of
a typical medical laboratory. We advocate to use %
bias and CV to express the maximum analytical bias
and imprecision derived by the concept below, because
laboratories are more familiar with this way of presentation than with absolute numbers of bias or standard deviations (for examples see tab. 3).
change") concept (33). This concept considers the case
of two consecutive measurements (xi and x2) performed
under identical analytical conditions, with the assumption of normal distributed χ values. In addition both
measurements are performed on the same underlying
population (μι = μ2). Then, the smallest medically significant difference (Amed) which analytically can be detected for two consecutive measurements (P = 0.05)
(sa = 0) is:
Amed = 1.96 χ >/2 * Si = 2.77 χ s{ (i).
Two important cases have to be considered for monitoring: a) Δ SE > 0; b) sa > 0. Therefore, equation (i) has
to be rewritten as follows (25):
Amed = 2.77 χ (si + s?)I/2 + Δ SE (ii).
If Δ SE = 0, Amed is only dependent on sa and s{, and
we can use the concept of Harris sa < 0.5 s{ (32). From
this follows:
Amed * 2.77 (0.25 s? + s?)1'2
Amed ^ 3.10 si (iii).
Now the case sa = 0 shall be considered. We start again
with equation (ii); with sa = 0, and Amed < 3.10 s~as
calculated in (iii):
2.77 χ Si + Δ SE < 3.10 si5 giving
Δ SE < 0.33 ^ (iv).
At this stage, we have derived maximum values for sa
(with Δ SE = 0) and Δ SE (with sa = 0). Now the function describing the combined effect of both is presented.
The combined effects of drift and analytical imprecision,
related to biology, can be represented by figure 1 (25).
It covers the case for which, due to analytical uncertainty, the ideal medically significant difference of
2.77 Si has to be widened to 3.10 Sj. It should be noted
0.1
Analytical quality specifications for
monitoring
The analytical specifications shall be derived from the
least "medically significant difference" ("reference
Sa/Sj
Fig. 1 * Reference change (Amcfi) concept: relationship between
maximum allowable drift and analytical standard deviation in fraction of Si, (Δ SE/SI) and (sa/Si), respectively. The figure represents
the case Amed = 3.10 Sj [from 1. c. (25), with permission].
Eur J Clin Chem Clin Biochem 1995; 33 (No 3)
167
Stock! et al.: Desirable routine analytical goals
that any combination of bias and standard deviation below this curve would allow reduction of Ameci to values
between 2.77 and 3.10 Sj. Further, the values of the function for SE = 0 and sa = 0 are the same as calculated above.
Analytical quality specifications for
diagnostic testing
The analytical specifications are outlined for sharing
common reference intervals, according to the concept of
Gowans et al. (35). This concept investigates the influence of method bias and imprecision on the percentage
of individuals outside each reference limit for a normal
distributed reference population. In figure 2 (35), the
combined effects of bias and imprecision place 4.6% of
a population outside of each reference limit (with zero
imprecision and bias, this is 2.5%). Again, the two extreme cases, bias = 0 or sa = 0 can be read from figure 2:
sa < 0.58 sc (v), for bias = 0
B < 0.25 sc (vi), for imprecision = 0.
0.4
0.3
5 0.2
0.1
0.1
0.2
0.3
sa/sc
0.4
0.5
0.6
Fig. 2 Common reference interval concept: relationship between
maximum allowable bias and analytical standard deviation in fraction of sc, (|B|/sc) and (sa/sc), respectively. The figure represents
the case that 4.6% of the individuals are outside each reference
limit due to analytical uncertainty [from 1. c. (35), with permission].
Acknowledgement
We are indebted to A. Uldall for continuous help for meeting facilities and stimulation of the project. We thank all members of the
Working Groups for fruitful discussions during our common meetings and for exchange of ideas by numerous letters.
References
1. Hyltoft Petersen P, H0rder M. Influence of analytical quality
on test results. Scand J Clin Lab Invest 1992; 52 Suppl
208:65-87.
2. Fräser CG, Hyltoft Petersen P. Desirable standards for laboratory tests if they are to fulfill medical needs. Clin Chem 1993;
39:1447-55.
3. Linnet K. A review on the methodology for assessing diagnostic tests. Clin Chem 1988; 34:1379-86.
4. Büttner J. Laboratory findings: structure, validity and significance for medical cognitive processes. Eur J Clin Chem Clin
Biochem 1991; 29:507-19.
5. Christensen M. Clinical test evaluation. Basic concepts. Scand
J Clin Lab Invest 1992; 52 Suppl 208:13-29.
6. Henderson AR. Assessing test accuracy and its clinical consequences: a primer for receiver operating characteristic curve
analysis. Ann Clin Biochem 1993; 30:521-39.
7. Barnett RN. Error limits and quality control [editorial]. Arch
Pathol Lab Med 1989; 113:829-30.
8. Barnett RN. Medical significance of laboratory results. Am J
Clin Pathol 1968; 50:671-6.
9. Gilbert RK. Progress and analytical goals in clinical chemistry.
Am J Clin Pathol 1975; 63:960-73.
10. Richtlinien der Bundesärztekammer zur Qualitätssicherung in
medizinischen Laboratorien. Dt Ärztebl 1988; 85:A699A712.
11. Naito HK. Current status of blood cholesterol measurement
in clinical laboratories of the United States: a report from the
Laboratory Standardization Panel of the National Cholesterol
Education Program. Clin Chem 1988; 34:193-201.
12. Campbell BG. Evaluation of two types of "medically significant error limits" and two quality control procedures on a
multichannel analyzer. Arch Pathol Lab Med 1989;
113:834-7.
13. Linnet K. Choosing quality control systems to detect maximum clinically allowable analytical errors. Ciin Chem 1989;
35:284-8.
Eur J Clin Chem Clin Biocbem 1995; 33 (No 3)
14. Westgard JO, Burnett RW. Precision requirements for cost-effective operation of analytical processes. Clin Chem 1990;
36:1629-32.
15. Westgard JO, Petersen PH, Wiebe DA. Laboratory process
specifications for assuring quality in the U. S. National Cholesterol Education Program. Clin Chem 1991; 37:656-61.
16. Bennett ST, Connelly DP, Eckfeldt JH. Assessment of splitsample proficiency testing for cholesterol by use of a computer
simulation model. Clin Chem 1991; 37:497-503.
17. Campbell DG, Owen JA. The physician's view of laboratory
performance. Aust Ann Med 1969; 18:4-6.
18. Barrett AE, Cameron SJ, Fräser CG, Penberthy LA, Shand KL.
A clinical view of analytical goals in clinical biochemistry. J
Clin Pathol 1979; 32:893-6.
19. Elion Gerritzen WE. Analytical precision in clinical chemistry
and medical decisions. Am J Clin Pathol 1980; 73:183-95.
20. Skendzel LP, Barnett RN, Platt R. Medical useful criteria for
analyte performance of laboratory tests. Am J Clin Pathol
1985; 83:200-5.
21. Lott JA, Surufka N, Massion CG. Proficiency testing of serum
enzymes based on medical needs criteria. Arch Pathol Lab Med
1991; 115:11-4.
22. Thue G, Sandberg S, Fugelli P. Clinical assessment of haemoglobin values by general practitioners related to analytical and
biological variation. Scand J Clin Lab Invest 1991; 51:453-9.
23. Petersen PH, Horder M. Ways of assessing quality goals for
diagnostic tests in clinical situations. Arch Pathol Lab Med
1988; 112:435-43.
24. Fräser CG, Petersen PH, Larsen ML. Setting analytical goals
for random analytical error in specific clinical monitoring situations. Clin Chem 1990; 36:1625-8.
25. Larsen ML, Fräser CG, Petersen PH. A comparison of analytical goals for haemoglobin A lc assays derived using different
strategies. Ann Clin Biochem 1991; 28:272-8.
168
26. Jenny RW. Analytical goals for determination of theophylline
concentration in serum. Clin Chem 1991; 37:154-8.
27. Tonks DB. A study of accuracy and precision of clinical chemistry determinations in 170 Canadian laboratories. Clin Chem
1963; 9:217-33.
28. Cotlove E, Harris EK, Williams GZ. Components of variation
in long term studies of serum constituents in normal subjects.
111. Physiological and medical implications. Clin Chem 1970;
16:1028-32.
29. Click JH Jr. Expression of random analytical error as a percentage of the range of clinical interest. Clin Chem 1976;
22:475-83.
30. Elevitch FR, editor: Analytical goals in clinical chemistry. Proceedings of the CAP Aspen Conference Aug 25-27 1976,
Skokie (IL). Skokie (IL): College of American Pathologists,
1977.
31. Analytical goals in clinical chemistry: Their relationship to
medical care. Proceedings of the Subcommittee on Analytical
Goals in Clinical Chemistry, World Association of Societies of
Pathology, CIBA Foundation; 1978 Apr 25-28; London. Am
J Clin Pathol 1979; 71:624-30.
32. Harris EK. Statistical principles underlying analytical goal-setting in clinical chemistry. Am J Clin Pathol 1979; 72:374-82.
33. Harris EK, Yasaka T. On the calculation of a "reference
change" for comparing two consecutive measurements. Clin
Chem 1983; 29:25-30.
34. Tietz NW. Textbook of clinical chemistry. Philadelphia: Saunders, 1986.
35. Gowans EMS, Hyltoft Petersen P, Blaabjerg 0, H0rder M. Analytical goals for the acceptance of common reference intervals
for laboratories throughout a geographical area. Scand J Clin
Lab Invest 1988; 48:757-64.
36. Hyltoft Petersen P, Gowans EMS, Blaabjerg 0, H0rder M. Analytical goals for the estimation of non-Gaussian reference intervals. Scand J Clin Lab Invest 1989; 49:727-37.
37. Fräser CG, Harris EK. Generation and application of data on
biological variation in clinical chemistry. Crit Rev Clin Lab
Sei 1989; 27:409-37.
38. Department of Health and Human Services Health Care Financing Administration. Medicare, Medicaid and CLIA programs; regulations implementing the Clinical Laboratory Improvement Amendments of 1988 (CLIA). Fed Regist 1992;
57:7002-186.
39. Büttner J, Borth R, Boutwell JH, Broughton PMG, Bowyer
RC. Approved recommendations on quality control in clinical
chemistry. Part 6. Quality requirements from the point of
health care. J Clin Chem Clin Biochem 1980; 18:861-6.
40. Büttner J, Borth R, Boutwell JH, Broughton PMG, Bowyer
RC. Approved recommendations on quality control in clinical
chemistry. Part 5. External quality control. J Clin Chem Clin
Biochem 1983; 21:885-92.
41. de Verdier CH, Aronsson T, Nyberg A. Quality control in clinical chemistry. Efforts to find an efficient strategy. Scand J
Clin Lab Invest 1984; 44 Suppl 72:9-241.
42. Westgard JO, Carey RN, Wold S. Criteria for judging precision
and accuracy in method development and evaluation. Clin
Chem 1974; 20:825-33.
43. Westgard JO, Hunt MR. Use and interpretation of common
statistical tests in method comparison studies. Clin Chem
1973; 19:49-57.
44. Westgard JO, Groth T. Power functions for statistical control
rules. Clin Chem 1979; 25:863-9.
45. Fräser CG. Better criteria for desirable laboratory performance
exist [letter]. Am J Clin Pathol 1986; 85:251.
46. Schectman G, Sasse E. Variability of lipid measurements: relevance for the clinician. Clin Chem 1993; 39:1495-503.
47. Lott JA, Manning NR, Kyler MK. Proficiency testing in a
medical needs context. Clin Chem 1989; 35:347-54.
48. Gowenlock AH, Broughton PMG. The influence of accuracy
and precision on the normal range and on acceptable limits for
an analytical result. Z Anal Chem 1968; 243:774-80.
StöckI et al: Desirable routine analytical goals
49. Fräser CG. The application of theoretical goals based on biological variation data in proficiency testing. Arch Pathol Lab
Med 1988; 112:404-15.
50. Queralto JM, Boyd JC, Harris EK. On the calculation of reference change values, with examples from a long-term study.
Clin Chem 1993; 39:1398-1403.
51. Harris EK. Proposed goals for analytical precision and accuracy in single point testing. Archf Pathol Lab Med 1988;
112:416-20.
52. Solberg HE, Petit Clerc C. Approved recommendation on the
theory of reference values. Part 3. Preparation of individuals
and collection of specimens for the production of reference
values. J Clin Chem Clin Biochem 1988; 26:593-8.
53. Ross JW. A theoretical basis for clinically relevant proficiency
testing evaluation limits: sensitivity analysis of the effect of
inherent test variability on acceptable method error. Arch Pathol Lab Med 1988; 112:421-34.
54. Ross JW, Fräser MD. Analytical goals developed from inherent
error of medical tests. Clin Chem 1993; 39:1481-94.
55. Stamm D. Die neuen Richtlinien der Bundesärztekammer zur
Qualitätssicherung in medizinischen Laboratorien. DG Klin
Chem Mitteil 1990; 21:69-77.
56. Klee GG. Tolerance limits for short-term analytical bias and
analytical imprecision derived from clinical assay specificity.
Clin Chem 1993; 39:1514-8.
57. Tietz NW. A model for a comprehensive measurement system
in clinical chemistry. Clin Chem 1979; 25:833-9.
58. Boutwell JH, editor: A national understanding for the development of reference materials and methods for clinical chemistry.
Proceedings of a conference. Washington: American Association for Clinical Chemistry, 1978.
.^
59. Büttner J. Reference methods as a basis for accurate measuring
systems. Eur J Clin Chem Clin Biochem 1991; 29:223-35.
60. Stöckl D, Reinauer H. Development of criteria for the evaluation of reference method values. Scand J Clin Lab Invest 1993;
53 Suppl 212:16-8.
61. Gilbert RK. CAP interlaboratory survey data and analytical
goals. Elevitch FR, editor: Analytical goals in clinical chemistry. Proceedings of the CAP Aspen Conference Aug 25-27
1976, Skokie (IL). Skokie (IL): College of American Pathologists, 1977:63-73.
62. Stamm D. Reference materials and reference methods in clinical chemistry. J Clin Chem Clin Biochem 1979; 17:283-97.
63. Stamm D. A new concept for quality control of clinical laboratory investigations in the light of clinical requirements and
based on reference method values. J Clin Chem Clin Biochem
1982; 20:817-24.
64. Gilbert RK. Accuracy of clinical laboratories studied by comparison with definitive methods. Am J Clin Pathol 1978; 70
Supplement without no.: 450-70.
65. Hartmann AE, Naito HK, Burnett RW, Welch MJ. Accuracy
of participant results utilized as target values in the CAP chemistry survey program. Arch Pathol Lab Med 1985; 109:894903.
66. Stöckl D, Reinauer H, Thienpont LM, De Leenheer AR Assessment of the accuracy of routine assays for the determination of steroid hormones and thyroxine in serum against isotope dilution mass-spectrometry target values. Comm Lab Med
1992; 1:77-9.
67. Welch MJ, Hertz HS. The how and why of an accuracy base
for proficiency testing programs. Arch Pathol Lab Med 1988;
112:343-6.
68. Ehrmeyer SS, Laessig RH, Leinweber JE, Oryall JJ. 1990
Medicare/CLIA final results for proficiency testing: maximum
intralaboratory performance characteristics (CV arjd bias)
needed to pass. Clin Chem 1990; 36:1736-40.
69. Jansen RTP, Bullock DG, Vassault A, Baadenhuijsen H, De
Leenheer A, Dumont G, de Verdier C-H, Zender R. Betweencountry comparability of clinical chemical results: an international quality assessment survey of 17 analytes in six European
countries through existing national schemes. Ann Clin Biochem 1993; 30:304-14.
Eur J Clin Chem Clin Biochem 1995; 33 (No 3)
Stöckl et al.: Desirable routine analytical goals
70. Fräser CG, Wilkinson SP, Neville RG, Knox JDE, King JF,
Mac Walter RS. Biologie variation of common hematologic
quantities in the elderly. Am J Clin Pathol 1989; 92:465-71.
71. Libeer JC. External Quality Assessment in clinical laboratories, European perspectives, today and tomorrow [dissertation]. Antwerp, Belgium: Univ. of Antwerp, 1993.
72. Ricos C, Arbos MA. Quality goals for hormone testing. Ann
Clin Biochem 1990; 27:353-8.
73. Smith SJ, Cooper GR, Myers GL, Sampson EJ. Biological
variability in concentrations of serum lipids: sources of variation among results from published studies and composite predicted values. Clin Chem 1993; 39:1012-22. ,
74. Fräser CG, Hyltofl Petersen P, Ricos C, Haeckel R. Proposed
quality specifications for the imprecision and inaccuracy of
analytical systems for clinical chemistry. Eur J Clin Chem Clin
Biochem 1992; 30:311-7.
75. Fuentes Arderiu X, Fräser CG. Analytical goals for interference. Ann Clin Biochem 1991; 28:393-5.
76. Vassault A, Grafmeyer D, Naudin CL, Dumont G, Bailly M,
Henny J, et al. Societe Francaise de Biologie Clinique. Protocole de validation de techniques. Ann Biol Clin 1986;
44:686-745.
Eur J Clin Chem Clin Biochem 1995; 33 (No 3)
169
77. Glick MR, Ryder KW, Jackson SA. Graphical comparison of
interferences in clinical chemistry instrumentation. Clin Chem
1986; 32:470-5.
78. Ryder KW, Glick MR. Erroneous laboratory results from hemolyzed, icteric, and lipemic specimens. Clin Chem 1993;
39:175-6.
79. de Verdier C-H, Groth T, Hyltoft Peterson P, editors. Medical
need for quality specifications in clinical laboratories. Upsala
JMedSci 1993; 8:187-491.
80. Stöckl D. Desirable performance priteria for quantitative measurements in medical laboratories based on biological analyte
variation — hindrances to reaching some and reasons to surpass some. Clin Chem 1993; 39:913-4.
81. Hyltoft Petersen P, Fräser CG, Westgard JO, Lytken Larsen
M. Analytical goal-setting for monitoring patients when two
analytical methods are used. Clin Chem 1992; 38:2256-60.
Dr. Dietmar Stöckl
INSTAND e. V.
Johannes-Weyer-Straße l
D-40225 Düsseldorf
Germany
Ruiz et al.: Immunoturbidimetry of digoxin without sample pretreatment
171
Eur J Clin Chem Clin Biochem
1995; 33:171-175
© 1995 Walter de Gruyter & Co.
Berlin · New York
Evaluation of an Immunoturbidimetric Assay of Serum Digoxin
without Sample Pretreatment
By Roberto Ruiz, Luis Barque, Angel G. Soria, Maria A. Cordova and Bianca Asolo
Laboratorio de Bioquimica, Hospital San Millan, Logrono, La Rioja, Spain
(Received May 11 /October 10/December 13, 1994)
Summary: We have evaluated a new turbidimetric immunoassay test (On Line, Roche) for the quantification of
digoxin in human serum without sample pretreatment. The assay, automated on a Cobas-Mira clinical analyser, is
based on the measurement of the changes in absorbance that occur when digoxin-coated microparticles aggregate
with digoxin antibody; the aggregation reaction is partially inhibited by digoxin in the sample.
The calibration curve extends up to 6.4 nmol/1, with a wide measuring range, and was stable at least for 10 days
during the studied period. The intra- and inter-assay coefficients of variation were in all cases lower than 7%. The
detection limit of the assay was 0.256 nmol/1 and the linearity, assessed by measuring serial dilutions of poisoned
patient sera, showed correlation coefficients higher than 0.998. There is no interference from bilirubin (up to 340
μιηοΐ/ΐ) or haemoglobin (up to 7900 mg/1), although a positive interference by lipids was found, which can be
avoided by previous delipidation of these sera. No digoxin-like factors were identified affecting digoxin values in
a concentration higher than that of the sensitivity of the assay in samples of premature infants, pregnant women,
or patients with renal or hepatic failure. Correlation coefficients between the turbidimetric assay and two immunoassays (fluorescence polarization and RJA) were 0.983 and 0.955 respectively, calculated from the assay of 102
and 98 samples. Nevertheless the values obtained by the evaluated assay were proportionally 20% higher, a fact
probably related to differences in calibrators.
Introduction
Digoxin, a glyc sylated steroid-like drug derived from
foxglove, is indicated in the treatment of congestive
heart failure, atrial fibrillation, or atrial flutter (1, 2).
Digoxin exerts inotropic and electrophysiological effects
on the heart, including increase in cardiac output and
diuresis, with decrease in venous pressure, heart size and
volume (3).
Measurement of serum digoxin concentration is necessary because of the narrow therapeutic range of this
drug. Furthermore, the important biological response of
patients, even with similar dosage regimens, often resuits in very different concentrations of digoxin in serum
(4). Concentrations higher than 2.6 nmol/1, are generally
interpreted as toxic in adult patients (5).
graphy (6), methods based on inhibition by digoxin of
sodium-potassium-activated adenosine triphosphatase in
red blood cells (7), radioimmunoassays (8), an enzymemultiplied immunoassay technique (9), an affinity-mediated immunometric assay (10), a fluorescence polarization immunoassay (11), an enzyme immunoassay using
recombinant DNA technology (12) and a radial partition
unmunoassay O3)·
Some of these methods are relatively tedious and have
been largely repiaced by the more practical radioimmunoassay techniques in routine clinical laboratories,
Nevertheless, the use of hazard-prone radionuclides
have spurred the use of other immunological labeis.
^^ methods often Jack Sensitivity5 require sample
Many techniques have been used to determine blood
concentrations of digoxin. They include gas chromato-
pretreatment or phase separation, or require special technology (8).
Eur J Clin Chem Clin Bioohetn 1995; 33 (No 3)
172
Ruiz et ah: Immunoturbidimetry of digoxin without sample pretreatment'
An immunoturbidimetric assay has recently been released by Roche. This assay, when automated on the
Cobas-Mira clinical analyser, measures the serum digoxin concentration, in 12 minutes without sample pretreatment. Such an advance would be greatly welcome
in current emergency, and intensive care units. The goal
of this work is to evaluate this method analytically, and
to study the possible use of this technology in our emergency laboratory.
Materials and Methods
Assay procedure
We used a Cobas-Mira S clinical analyser (Roche, Basel, Switzerland) for all the determinations. All the steps are performed automatically by the instrument. First, 6 μΐ of calibrator, control or serum are added concurrently with 125 μΐ of the antibody solution.
The mixture is incubated for 400 s at 37 °C, and subsequently
30 μΐ of microparticle-containing solution, are pipetted. After a second mixing of the reaction cuvette, the kinetic increments of absorbance are measured at 600 nm during 175 s. When calibrated,
the instrument builds a six-point log-logit curve with the increments of absorbance per minute of the duplicated calibrators,
and stores it in the memory for subsequent recall and comparison
with absorbance values of the samples. Specimens with assay values greater than the highest calibrator, were diluted with the 0.0
nmol/1 calibrator and reassayed.
Samples
Sera from different clinical areas of our hospital were obtained
after centrifiigation of blood samples at 1250 g for 10 min. When
not immediately measured, the sera were stored at 4 °C for no more
than three days or frozen at -20 °C until use (no more than a
month). In the study of digoxin-like immunoreactive factors, samples from pregnant patients, premature infants, and individuals with
renal or hepatic insufficiency were measured. For studying haemoglobin, bilirubin^and Intralipid® interferences (Kabivitrum, Stockholm, Sweden), a serum pool with digoxin was spiked with the
different interferents in accordance with the procedure of Glick
(14). Lipaemic samples were treated with 1,1,2-trichloro-trifluoroethane, a clearing agent from Serva (Ref 36965, Serva, Heidelberg, Germany).
Reagents, calibrators and controls
The reagents are ready to use as supplied (Roche-TDM ONLINE
for Digoxin, Ref 0739863, Hoffmann-La Roche AG, Basel, Switzerland), and were stable at least during 4 months when stored at
4 °C after their use. The kit consists of:
Reagent 1, antibody solution: monoclonal antibodies anti-digoxin
(mouse) in a buffered solution with 7.7 mmol of NaN3 per litre as
preservant. Digitoxin, dihydrodigoxin, digoxigenin and digitoxigenin (84, 162, 64 and 933 nmol/1, respectively) display crossreactivities of 5.0, 2.5, 12.8 and 0.9%, respectively. The assay has
no clinically significant cross-reactivity (< 0.03%) with different
steroids at concentrations lower than 10 mg/1 (oestradiol, oestriol,
progesterone, 17-hydroxyprogesterone, prednisone, hydrocortisone, spironolactone; communication from the manufacturer).
Reagent 2, microparticles: digoxin conjugated particles suspended
in a neutral solution with 7.7 mmol of NaN3 per litre as preservant.
Comparison methods
The results obtained by the present method were compared with
those of a fluorescence polarization immunoassay and a radioimmunoassay, both performed in accordance with the manufacturers' instructions (Ref 9511-60, Abbott TDx, Abbott Laboratories,
USA and Ref CA-1527, Incstar Corp, USA, respectively).
Results
Calibration curve
When the change of absorbance per minute is plotted
vs. digoxin concentration, a classic immunoinhibition
curve is obtained. Figure 1 shows mean ± 1 standard
deviation of 10 calibrations obtained during a period
of 3 months. The calibration curves were stable for at
least 10 days.
Precision
Intra-assay precision of the Roche reagent was determined with the three concentrations of the control by
replicate testing (n = 20) with a single calibration run.
Inter-assay precision (n = 20) was evaluated for each of
these samples over a period of 3 months by using a
The calibrators provided by Roche were used (Ref 0717649).
These consist of 6 liquid vials with digoxin concentrations of 0,
0.64, 1.28, 2.56, 3.84 and 6.4 nmol/1 in native human sera.
The controls (Ref 43286, Roche) were also used in the precision
studies, and were analysed as supplied. These controls are obtained
from native human sera.
0.045 p
c 0.035
ε
i 0.025
Principle of procedure
ο
ο
<
The method is an immunoassay based on the measurement of
changes in absorbance that occur when activated microparticles aggregate. The microparticles, coated with digoxin, aggregate in the
presence of a digoxin antibody solution. When a sample containing
digoxin is introduced, the antibody binds to sample drug and is
therefore no longer available to promote microparticle aggregation.
The aggregation reaction is partially inhibited and an inhibition
curve with respect to concentration is obtained; the maximum rate
of aggregation occurs at the lowest sample digoxin concentration.
0.015
0.005
L
L
0
1
2
3
4
5
6
7
Digoxin [nmol/l]
Fig. l Calibration curve of the digoxin assay. Change of absorbance per minute (mean ± S.D. of 10 calibration curves) versus
standard concentrations.
»;
Eur J Clin Chem Clin Biochem 1995; 33 (No 3)
173
Ruiz et al.: Immunoturbidimetry of digoxin without sample pretreatment
total of 10 separate calibration runs. Results are shown
in table 1.
Linearity and detection limit
Linearity was evaluated by diluting (with 9 g/1 NaCl solution) three different patient samples containing high
digoxin concentrations. The results are expressed in figure 2. The coefficient of linear correlation r, was greater
than 0.998 in every case.
The detection limit of the assay, calculated as the mean
plus two standard deviations of the digoxin values obtained with 20 replicates of the 0 nmol/1 calibrator, was
0.256 nmol/1.
Interference studies
Figure 3 shows interferences associated with addition of
increasing quantities of haemoglobin, bilirubin or Intralipid® to a pool containing 1.98 nmol/1 of digoxin. No
effects were seen in any sample with levels up to 340
μηιοΐ/ΐ for bilirubin and 7900 mg/1 for haemoglobin. A
positive interference was found when Intralipid® emulsion was added.
Tab. 1 Precision data for the present method for serum digoxin.
Digoxin
Intra-assay3
Inter-assayb
20
x, nmol/1
SD, nmol/1
CV, %
x, nmol/1
SD, nmol/1
CV, %
Sample
I
Sample
II
Sample
III
1.91
0.05
2.7
2.92
0.06
2.2
4.28
0.10
2.10
0.14
3.14
0.13
4.70
0.14
6.7
2.4
3.0
4.1
b
ο
υ
α>
oc
0.25
0.50
0.75
1.00
Interfering substance, fraction of
highest concentration
Fig. 3 Interference studies of the immunoturbidimetric assay. The
abscissa shows the concentrations of interfering substance, expressed as a fraction of highest concentration:
D-D haemoglobin 7800 mg/1; Δ-A bilirubin 340 μηιοΐ/ΐ;
Ο—Ο Intralipid® 2%; Π—D Intralipid®-spiked series after treatment
with delipidating agent 1,1,2-trichloro-trifluoroethane.
In order to assess whether the treatment of lipaemic
samples with delipidating agents affects the digoxin
concentration, the Intralipid® spiked samples were measured after delipidation with 1,1,2-trichloro-trifluoroethane. Figure 3 shows that this treatment of lipaemic
sera does not affect digoxin concentrations.
When measuring digoxin in 10 pregnant patients, 10
premature infants, 10 patients with renal insufficiency
and 10 individuals with hepatic failure, the values of
digoxin obtained were in no case higher than the detection limit of the assay. So we were not able to identify
digoxin-like immunoreactive factors.
Correlation
Comparison of results by a non-parametric regression
analysis (15) was carried out with 102 and 98 samples
containing a wide range of concentrations (fig. 4). The
regression equations were y(Roche) = 1.200 x(TDx)
+ 0.030, and y(Roche) =1.185 x(RIA) - 0.157, and
the correlation coefficients r were 0.983 (p < 0.001) and
0.955 (p < 0.001) respectively, indicating a close correlation between the two assays.
n = 20
.£
x
I
Practicability
0.2
Θ.4
0.6
0.8
1.0
Dilution
Fig. 2 Assay linearity of the immunoturbidimetric assay. Digoxin
concentrations of diluted samples from poisoned patients.
Eur J Clin Chem Clin Biochem 1995; 33 (No 3)
Due to the absence of sera pretreatment, the method is
especially suitable for urgent digoxin determinations.
The first result is obtained after 12 minutes, followed by
results at one minute intervals. Although the manufacturer recommends calibration with duplicates, we find
no necessity to duplicate calibrators or samples.
174
Ruiz et al.: Immunoturbidimetry of digoxin without sample pretreatment
=: 7 r
ο
•σ
ο
χ ο
Ο|
!5
ο
1
2
3
4
5
6
Digoxin (TDx) [nmol/l]
Fig. 4 Comparison of results according to a non-parametric regression analysis (15) for digoxin measured by the immunoturbidimetric method under test (y) versus a fluorescence polarization
assay in a TDx analyser:
y = bx + a; b = 1.200, a = 0.030, r = 0.983, N = 102 and versus
0
1
2
3
4
5
6
Digoxin (RIA) [nmol/l]
7
a radioimmunoassay from Incstar:
y = bx + a; b = 1.185, a = -0.157, r = 0.955, N = 98
Solid lines represent the regression lines of the Passing & Bablok
statistical procedure,
Discussion
digoxin-like immunoreactive factor or factors (17-22).
The degree of interference caused by these factors varies
The results presented in this paper indicate that a non-raconsiderably with the immunoassay kit used (1). This
dioactive immunoturbidimetric assay for digoxin can be
variability most likely results from the different bincfihg
accomplished without difficulty, and that such an assay is
affinities of these substances with the different digoxin
rapid, fully automated in a widely used instrument, and
antibodies. In the present assay, albeit with few samples,
requires no previous pretreatment of samples. The method
we found no interference by digoxin-like factors, i. e. no
has an excellent calibration curve (results expressed as ininterference equivalent to a concentration higher than
crement of absorbance per minute), with a wide measurthat of the detection limit of the assay. Nevertheless,
ing range and a long stability, which allows very spaced
given the unknown structures and random presentation
calibrations. The detection limit and precision of the assay
of these substances, subsequent studies will show if clinare better than those of other widely used methods and
ically important interferences by digoxin-like factors afsimilar to those of the radioimmunoassays (8). The differ- fect the method, although the very low sample volume
ences for the mean values of the controls as determined by needed (6 μΐ), in comparison with most of the other asintra- or inter-assay measurements must be attributed to a says (8), as well as the selection of a specific monoclostatistically extreme value on the day that intra-assay nal antibody, suggest these could be lower.
precision was determined.
The comparison with two of the most widely used digoxin
For measurement by particle immunoassay (16), it may assays showed that the evaluated method has high debe necessary to treat turbid or lipaemic samples prior gree of correlation for patient specimens over a wide conto assay with high-speed centrifiigation or delipidating centration range. We found, however, that the methods are
agents, because chylous or denatured particles scatter not interchangeable; the values obtained by the immunoor absorb light, causing non-specific interferences. The turbidimetric method were proportionally 20% higher
present assay suffers from this drawback, which can be than those of the fluorescence polarization or radiornetric
easily avoided by previous treatment of lipaemic sera assays, a fact that can be related to differences in manuwith 1,1,2-trichloro-trifluoroethane, without affecting facturer's calibrators, as found by other authors (13).
digoxin concentration. Neither haemoglobin nor biliru- In summary, the use of this technology in a Roche's
bin interfered with the assay and the method showed Cobas-Mira clinical analyser greatly simplifies the assay
good linearity.
of digoxin in human serum and makes the procedure
practical for routine or emergency applications. NeverThe validity of immunoassays for measurement of di- theless, the proportional differences Obtained by the
goxin has been shown to be compromised in certain evaluated assay suggest that the results should be corclinical populations by the presence of a cross-reacting rected by mean of the software of the instrument.
Bur J Clin Chera Clin Biochem 1995; 33 (No 3)
175
Ruiz et al.: Immunoturbidimetry of digoxin without sample pretreatment
References
1. Cook JD, Koch TR, Cook MS, Knoblock EC. Inaccuracies in
digoxin measurement. CHn Biochem 1988; 21:353—8.
2. Lewis LP. Clinical use of digoxin concentrations. Am J Cardiol
1992;69:97G-107G.
3. Lee KS, Klaus W. The subcellular basis for the mechanism of
inotropic action of cardiac glycosides. Pharmacol Rev 1971;
23:193-261.
4. Moyer TP, Pippenger CE, Blanke RV, Blouin RA. Therapeutic
drug monitoring. In: Tietz NW, editor. Textbook of clinical
chemistry. Philadelphia: Saunders, 1986:1615-69.
5. Bresnahan JF, Vlietstra RE. Digitalis glycosides. Mayo Clin
Proc 1979; 54:675-84.
6. Watson E, Kaiman SM. Assay of digoxin in plasma by gas
chromatography. J Chromotogr 1971; 56:209.
7. Bertler A, Redfors A. An improved method of estimating digoxin in human plasma. Clin Pharmacol Ther 1970; 11:665.
8. Chen I-W, Heminger LA. Digoxin and digitoxin. In: Pesce AJ,
Kaplan LA, editors. Methods in clinical chemistry. Si Louis:
The CV Mosby Co, 1987:897-902.
9. Eriksen PB, Andersen O. Homogeneous enzyme immunoassay
for serum digoxin with use of bichromatic analyzer. Clin Chem
1979; 23:169.
10. Leflar CC, Freytag JW, Powell LM. An automated affinitycolumn mediated, enzyme-linked immunometric assay for digoxin on the DuPont aca discrete clinical analyzer. Clin Chem
1984; 30:1809-11.
11. Porter WH, Haver VM, Bush BA. Effect of protein concentration on the determination of digoxin in serum by fluorescence
polarization immunoassay. Clin Chem 1984; 30:1826-9.
12. Pincus MR, Abrham NZ. Toxicology and therapeutic drug
monitoring. In: Henry JB, editor. Clinical diagnosis and management by laboratory methods. Philadelphia: Saunders,
1991:349-84.
Eur J Clin Chem Clin Biochem 1995; 33 (No 3)
13. Giegel JL, Brotherton MM, Cronin P, D'Aquino M. Radial
partition immunoassay. Clin Chem 1982; 28:1894—8.
14. Click MR, Ryder KW, Jackson S A. Graphical comparisons of
interferences in clinical chemistry instrumentation. Clin Chem
1986; 32:470-5.
15. Passing H, Bablok W. A new biometrical procedure for testing
the equality of measurements from two different analytical
methods. J Clin Chem Clin Biochem 1983; 21:709-20.
16. Masson PL, Cambiaso CL, Collet-Casart D, Magnusson CGM.
Particle counting immunoassay (PACIA). Methods Enzymol
1981; 74:106-39.
17. Doolittle MH, Lincoln K, Graves SW. Unexplained increase
in serum digoxin: a case report. Clin Chem 1994; 40:487-92.
18. Graves SW. Endogeneous digitalis-like factors. Crit Rev Clin
Lab Sei 1986; 23:177-200.
19. Graves SW, Brown BA, Valdes R. Digoxin-like substances
measured in patients with renal impairment. Ann Intern Med
1983; 99:604-8.
20. Valdes R, Graves SW, Brown BA, Landt ML. Endogeneous
substances in newborn infants causing false-positive measurements. J Pediatr 1983; 102:947-50.
21. Graves SW, Valdes R, Brown BA, Knight AB. Endogeneous
immunoreactive digoxin-like substance in human pregnancies.
J Clin Endocrinol Metab 1984; 58:748-52.
22. Nanji AA, Greenway DC. Falsely raised plasma digoxin concentrations in liver disease. Br Med J 1985; 209:432-43.
Dr. Roberto Ruiz
Laboratorio Central
Hospital San Millan
Autonomia de La Rioja 3
E-26004-Logrono
Spain