American Journal of Epidemiology
Copyright C 1999 by The Johns Hopkins University School of Hygiene and Public Health
All rights reserved
Vol. 149, No 3
Printed in U.SA.
Analytical and Biologic Variability in Measures of Hemostasis, Fibrinolysis,
and Inflammation: Assessment and Implications for Epidemiology
Pamela A. Sakkinen,1'2 Elizabeth M. Macy,1 Peter W. Callas,3 Elaine S. Cornell,1 Timothy E. Hayes,4
Lewis H. Kuller,5 and Russell P. Tracy112
An increasing number of cardiovascular epidemiologic studies are measuring non-traditional risk markers of
disease, most of which do not have established biovariability characteristics. When biovariability data have been
reported, they usually represent a short time period, and, in any case, there is little consensus on how the
information should be used. The authors performed a long-term (6-month) repeated measures study on 26
healthy individuals, and, using a nested analysis of variance (ANOVA) approach, report on the analytical (CVA),
intraindividual (CV,), and between individual (CVQ) variability of 12 procoagulant, fibrinolysis, and inflammation
assays, including total cholesterol for comparison. The results suggest acceptable analytical variability (CVA <,
1/2 CV,) for all assays. However, there was a large range of intraindividual variation as a proportion of total
variance (2-78%), and adjusting for intraindividual and between individual variation in bivariate correlations
increased the observed correlation by more than 30 percent for three of these assays. Overall, the assays
showed a significant increase in intraindividual variation over 6 months (p < 0.05). While these findings suggest
that most of these assays have biovariability characteristics similar to cholesterol, there is variation among
assays. Some assays may be better suited to epidemiologic studies, and knowledge of an assay's biovariability
data may be useful in interpreting simple statistics, and in designing multivariate models. Am J Epidemiol
1999;149:261-7.
cardiovascular disease; hemostasis; variability
There is increasing evidence for the role of the
thrombosis risk factors (1, 2) in the chronic progression of atherosclerotic cardiovascular disease (CVD),
as well as in the precipitating acute CVD events. These
factors have been used in univariate and multivariate
analysis of CVD. Little is known, however, about the
analytical and biologic variability of these factors, and
there has been little published about using analytical
and biologic variability in the interpretation of CVD
models.
In epidemiologic studies, intraindividual and analytical variance (a,2 and aA2) diminish measures of correlation and risk, may alter our perception of the physiologic relevance of a variable, and generally
underestimate a univariate measure of association (3).
In case of multivariate models, however, the observed
effects of these variables may be either under- or overestimated, if even only one variable is measured imprecisely (4-6). Although methods exist to adjust relative
risks using multivariate relative risk techniques, these
methods may require assumptions which are not met by
the data. Some investigators advocate focusing on the
interpretation of the observed effect or designing studies to prevent measurement imprecision rather than
adjusting for variation in the analyses imprecisely (4).
There is no consensus on how variability measures
should be used in epidemiologic studies.
A proportionately large population variance (aG2) as
part of the total variation (aT2; GT2 = aG2 + a,2 (intraindividual variation) + OA2 (analytical variation))
enhances a factor's reliability, and generates a more
Received for publication November 14, 1997, and accepted for
publication June 9, 1998.
Abbreviations: CRP, C-reactjve protein; CV, coefficient of variation;
CVD, cardiovascular disease; EUSA, enzyme-linked immunosorbent
assay; F1-2, prothrombin fragment F1-2; FPA, fibrinopeptJde A;
FXIa-a1AT, factor Xla-a)pha-1-antitfypsin; / ( , index of individuality;
PAI-1, plasminogen activator inhibitor-1; PAP, plasmin-alpha-2antiplasmin complex; TAT, thrombin-antithrombin complex; t-PA,
tissue plasminogen activator; t-PA/PAl complex, tissue plasminogen
activator/plasminogen activator inhibrtor-1 complex; R, reliability
factor; ay%, validity coefficient.
1
Department of Pathology, University of Vermont, Burlington, VT.
2
Department of Biochemistry, University of Vermont, Burlington,
VT.
department of Mathematics and Statistics, University of
Vermont, Burlington, VT.
'Maine Medical Center, Portland, ME.
'Department of Epidemiology, University of Pittsburgh, School of
Public Health, Pittsburgh, PA.
Reprint requests to Dr. Russell P. Tracy, Laboratory for Clinical
Biochemistry Research, University of Vermont, 55A S. Park Drive,
Colchester, VT 05446.
261
262
Sakkinen et al.
accurate estimate of the variable-disease association
(7). In addition, an increased oG2 may also improve the
ability of multivariate models to detect correlations
among variables, whereas an increased a,2 may
increase the probability of measuring an extreme value
for an individual. This implies that some factors may
be better suited to epidemiologic study design by
nature of their biologic and analytical variability than
others.
The role of variability in hemostatic risk factors of
CVD was initially reported by Thompson et al. (8).
More recently, others have contributed to addressing
the implications of biologic and analytical variability
on sample size calculations (9, 10), short-term factor
reliability (11, 12), and the effects of selected patient
and sample characteristics (9, 11, 12) on factor biovariability. We present long-term variability data on 12
hemostasis, fibrinolysis, and inflammation factors
(plus cholesterol for comparison). For six of these risk
factors, no variability data exists in the literature. We
suggest that these data may be used to: 1) adjust continuous measures of association between factors in
univariate analyses; 2) evaluate variables for use in
epidemiologic studies based on their characteristic
variation; and 3) estimate the likelihood of observing
relations between CVD and these factors compared
with cholesterol, assuming equivalent physiologic relevance.
MATERIALS AND METHODS
Population
We recruited 26 apparently healthy individuals (16
women and 10 men) who ranged in age from 23 to 72
years. Four of the women were postmenopausal.
Subjects provided information regarding recent medication use and state of general health. Subjects were asked
to abstain from caffeine intake prior to sample collection. Early morning venous blood samples were collected following overnight fast at 3-week intervals over 24
weeks, and samples were analyzed all at one time.
Blood collection and analysis
One citrate tube (0.129 mol/liter of citrate/citric
acid; Becton Dickinson, San Jose, California) and one
SCAT-1 tube (200 KlU/ml Aprotinin, 50 |xM PPACK
(d-phenylalanyl-prolyl-arginyl-chloromethyl ketone)
4.5 mM EDTA, Haematological Technologies, Essex
Junction, Vermont) were obtained following venipuncture. To minimize plasminogen activator inhibitor-1
(PAI-1) release during activation of platelets, blood
collection and handling recommendations were followed as previously described (13).
PAI-1 antigen was measured using enzyme-linked
immunosorbent assay (ELISA) (14). PAI-1 activity
was assayed using the American Diagnostica
(Greenwich, Connecticut) plasmin substrate,
Spectrolyse-PL, to measure PAI-1 inhibition of t-PA
activation of plasminogen to plasmin in the presence
of a fibrin catalyst (15). The assay was performed
according to manufacturer's instructions with the
exceptions that plasminogen and fibrin catalyst were
produced in our laboratory. Plasminogen isolation was
performed by affinity chromatography on lysinesepharose followed by gel filtration using ACA-44.
The fibrin catalyst was made by cyanogen bromide
digestion of fibrinogen purified from plasminogen/
plasmin depleted plasma.
Tissue-plasminogen activator (t-PA) antigen (16)
and t-PA/PAI-1 complex (17) were measured with
ELISA. Plasmin-alpha-2-antiplasmin complex (PAP)
was measured by ELISA developed by Holvoet et al.
(18). The fibrin fragment D-dimer was measured by
ELISA (19). Monoclonal antibodies for the PAI-1 antigen, t-PA antigen, and the t-PA/PAI-1 complex assays
were generous gifts of Dr. Desire Collen at the Center
for Molecular and Vascular Biology at the University
of Leuven, Belgium.
Prothrombin fragment Fl-2 (Fl-2) was measured
with ELISA (Baxter-Dade, Miami, Florida) (20).
Fibrinopeptide A (FPA) was measured on bentoniteextracted, fibrinogen-free plasma by a double
antibody competition radioimmunoassay (BykSangtec Diagnostica) (21).
C-reactive protein (CRP) was determined by ELISA
(antibodies and antigens from Calbiochem, LaJolla,
California) (22). Fibrinogen was measured with the
ST4 instrument (Diagnostica Stago) and bovine
thrombin (Parke-Davis, Lititz, Pennsylvania) by the
general method of Clauss (23).
Factor XIa-alpha-1-antitrypsin (FXIa-alAT) was
measured by ELISA with antigens and antibodies from
Affinity Biologicals, Inc., Hamilton, Ontario.
Thrombin antithrombin complex (TAT) was determined by enzyme immunoassay (Behring Diagnostics,
Inc., Westwood, Massachusetts).
Cholesterol was measured by a Centers for Disease
Control and Prevention-certified enzymatic method, as
described previously (24, 25).
Statistical analysis
Random effects nested analysis of variance
(ANOVA) was performed to separate the sources of
variation (26, 27), defined as sA2 = average variance of
replicate assays (within-run analytical variance), s 2 =
average biologic within-subject variance, and sc2 =
variance of true means among subjects. The coefficient
Am J Epidemiol Vol. 149, No. 3, 1999
Variability in Hemostasis, Fibrinotysis, and Inflammation
of variation (CV) is defined as 100 percent x (standard
deviation/mean), and was calculated for each component of variance (CVA, CV,, and CVG). We used the
standard constant variance model, which assumes that
at differing analyte concentration, intraindividual and
analytical variation remain constant {27).
We calculated the ratio CV/CVG = I,, referred to as
the Index of Individuality, which is a measure of how
individuals vary relative to the population distribution.
The proportion of total variance attributable to
interindividual variation (reliability factor) was also
calculated, ((sG2/sT2); (sT2 = sj + s2 + (sA2)), in addition
to the proportion of variance attributable to withinperson and analytical variation in a similar fashion.
Outlier removal was based on the Cochran test, using
both the ratio of the maximum variance to the sum of
the variances, and the variances of the specimen means
among the individual subjects (28), as suggested by
Fraser and Harris (26). In general, the medical history
provided by the volunteers did not contribute to the
explanation of outlying values. However, in the case of
CRP, all three outlying values were obtained from
individuals who reported influenza symptoms.
Pearson correlation coefficients were adjusted by the
method of Liu et al (29). The actual correlation estimated was defined as rc = rg (1 -r (1 + (s2/ksG2)))m, where
ro = the observed correlation between factor X and a
random variable Z, r. = the actual correlation between
the true mean of factor X and a random variable Z (cor-
rected for attenuation), sf = sample intraindividual variation, sc2 = sample between individual variation, and k =
the number of measurements for each individual.
The validity coefficient p Ti , an estimate of how
much the observed measure differs from the true measure due to its variability, was defined as (1 -r (1 +
(s2/ksc2))yn, where k = the number of measurements
for each individual (29). The minimum number of
measurements for each individual to achieve a validity
coefficient of 0.91 (validity coefficient of cholesterol)
was derived as k = (0.912/(l - 0.912) x (sf/s/)) (29).
The calculation assumes that the analytical variation is
very small compared with intraindividual variation.
SPSS for Windows was used for descriptive analyses (30). A Lotus spreadsheet template was developed
and used for the nested ANOVA calculations.
RESULTS
The median age of the population was 37 years. One
of the women became pregnant during the study, and
was removed from the analyses. Three participants
were smokers.
Descriptive statistics are shown in table 1. Among
the hemostatic factors, PAP, PAI-1 ag, and fibrinogen
had the lowest analytical variation, approaching that of
cholesterol. Although cholesterol had the lowest analytical imprecision among the 13 variables, all of the
variables had an "acceptable" degree of analytical pre-
TABLE 1. Descriptive statistics In a study of 26 apparently hearthy Individuals carried out over 24
weeks at the Laboratory for Clinical and Biochemical Research at the University of Vermont, Burlington,
Vermont, November 1992 to April 1993
Hemostatic factor*
t-PA ag (ng/ml)
PAI-1 ag (ng/ml)
PAI-1 activity (Ill/ml)
t-PA/PAl (ng/ml)
PAP (nM)
D-Dimer (ng/ml)
Fibrinogen (mg/dl)
F1-2(nM)
TAT (ng/liter)
FPA (ng/ml)
FXIa-aiAT(nM)
CRP (jig/ml)
Cholesterol (mg/dl)
Mean
SD»
7.4
3.6
20.5
4.8
1.0
1.3
90.1
56.0
22.1
6.2
1.5
4.0
93.4
233.2
2.1
2.8
5.3
2.3
1.5
208.9
0.7
CV* (control
materials)!
6.3
2.5
21. At
15.9
4.4
6.8
2.9
7.9
0.9
7.1
5.1
4.7
1.6
17.0
5.9
5.6
43.3
2.0§
ANOVA* results
cv A *
CV*
cv 0 *
5.4
3.5
12.2
10.3
1.7
12.2
1.6
6.2
12.5
14.5
6.3
5.2
15.0
47.2
30.3
22.4
19.9
56.4
18.6
24.2
25.0
82.7
26.3
49.8
45.6
70.9
62.0
59.3
29.3
89.5
20.2
28.3
60.5
45.8
184.0
94.0
1.0
8.8
19.8
•SD, standard deviation; CV, coefficient of variation; ANOVA, analysis of variance; CVA, analytical coefficient
of variation; CV,, intraindividual coefficient of variation; CV0, individual coefficient of variation; t-PA, tissueplasminogen activator; ag, antigen; PAI-1, plasminogen activator-1; tPA/PAl, tissue plasminogen
activator/plasminogen activator inhibitor-1 complex; PAP, plasmin-alpha-2-antiplasmin complex; F1-2,
prothrombin fragment F1-2; TAT, thrombin-antithrombin complex; FPA, fibrinopeptide A; FXIa-a1AT, factor Xlaalpha-1- antitrypsin; CRP, C-reactive protein.
t CV calculated from repetitive use of control materials during the anaiyses of the samples in this study.
t CV calculated from samples measured in a separate, but contemporaneous study.
§ CV calculated from monthly mean values measured over one year.
Am J Epidemiol
Vol. 149, No. 3, 1999
263
264
Sakkinen et al.
cision (CVA < 1/2 CV,) (26). The CV determined from
control materials were not consistently smaller or larger than the calculated CVA.
The relative variance components of each variable
are presented in table 2 as the Index of Individuality
(Ij) and the proportion of variation due to the berweenperson (R), intraindividual, and analytical components.
Comparison R values from variability studies in middle-aged persons are given in parentheses (9, 11, 12).
The Ij, a measure of how individuals vary related to the
population distribution, ranged from 0.1 to 1.8, with 11
of 12 measures at 0.9 or less. Therefore, despite greater
CV, and CVA, many of these variables have I,'s similar
to cholesterol (I, = 0.44). Within groups of fibrinolysis,
inflammation, and hemostasis markers, there was a
gradation of the individuality indexes and reliability
factors. The R was essentially a mirror image of the ^
scale, reflecting the small addition of the analytical
variation to the total variation.
In general, the intraindividual variation component
increased for each analyte over the seven timepoints
(table 3), whereas the analytical component of variation was essentially unchanged (data not shown).
Examination of the standardized CV, (obtained by
dividing the subsequent CV, for each analyte by the
initial CV, for each analyte) showed that the mean CV,
across all 13 hemostatic factors at 6 and 9 weeks were
lower than the mean CV, at 24 weeks (paired Mest, p <
0.05); however, there was no difference at >12 weeks
(data not shown).
TABLE 2. Index of Individuality of hemostatic factors, ranked on Individuality Index and reliability
factor (R = s^/s,1) In the present study compared with R values from three other variability studies
(shown In parentheses)
Hemostatic factor*
Components of variation (fl) as a fraction of total*
CV,
Index of Individuality
FXIa-aiAT(nM)
t-PA ag (ng/ml)
t-PA/PAl (ng/ml)
TAT (jig/liter)
Cholesterol (mg/dl)
PAI activity (lU/ml)
CRP (ng/ml)
D-Dimer (ng/ml)
PAI ag (ng/ml)
PAP (nM)
F1-2(nM)
Fibrinogen (mg/dl)
FPA (ng/ml)
0.14
0.33
0.38
0.41
0.44
0.49
0.53
0.63
0.67
0.68
0.85
0.92
1.77
0.98
0.89
0.86
0.82
0.83
0.78
0.77
0.71
0.68
0.63
0.56
0.54
0.24
(0.81 )t(0.97)§
(0.85)4:
(0.86)§
(0.96)§
(0.73)t
(0.72)t
(0.72)f (0.87)§
(0.03)
0.02
0.10
0.12
0.14
0.16
0.19
0.22
0.28
0.31
0.36
0.41
0.46
0.74
<0.01
0.01
0.02
0.04
<0.01
0.03
<0.01
0.01
<0.01
<0.01
0.03
<0.01
0.02
* See table 1 for definitions of abbreviations,
t Chambless et al. (12).
t Nguyen et al. (11).
§ de Maat et al. (9).
TABLE 3. Intraindividual coefficient of variation (CVJ of hemostatic values in a study of 26 apparently
healthy Individuals carried out over 24 weeks at the Laboratory for Clinical and Biochemical Research at
the University of Vermont, Burlington, Vermont, November 1992 to April 1993
No of weeks of follow-up
Hsmostatic factor*
FXIa-aiAT(nM)
t-PA ag (ng/ml)
t-PA ag (ng/ml)
Cholesterol (mg/dl)
TAT (ng/liter)
PAI activity (lU/ml)
CRP Gig/ml)
D-Dimer (ng/ml)
PAI ag (ng/ml)
PAP ag (nM)
F1-2 (nM)
Fibrinogen (mg/dl)
FPA (ng/ml)
1
.
6
9
0.116
0.138
0.186
0.052
0.195
0.262
0.151
0.490
0.443
0.143
0.219
0.162
0.534
0.183
0.129
0.184
0.061
0.233
0.302
0.253
0.462
0.464
0.166
0.223
0.171
0.915
12
0.179
0.147
0.192
0.072
0.250
0.284
0.547
0.513
0.500
0.173
0.239
0.171
0.872
21
15
16
0.201
0.148
0.206
0.087
0.271
0.300
0.517
0.553
0.484
0.195
0.247
0.187
0.855
0.201
0.147
0.217
0.086
0.261
0.294
0.514
0.553
0.464
0.195
0.259
0.184
0.868
0.201
0.150
0.223
0.086
0.257
0.312
0.503
0.530
0.469
0.197
0.251
0.178
0.827
24
0.263
0.151
0.223
0.088
0.250
0.303
0.498
0.564
0.472
0.199
0.242
0.174
0.827
See table 1 for definitions of abbreviations.
Am J Epidemiol
Vol. 149, No. 3, 1999
Variability in Hemostasis, Fibrinolysis, and Inflammation
Observed validity coefficients and the number of
replicate specimens needed to achieve the validity
coefficient level of cholesterol (0.91) are shown in
table 4. In general, only one or two measurements
were necessary to reduce the intraindividual variation
of the factor to {hat of cholesterol. The three factors
which required more than three measurements were
FPA, Fl-2, and fibrinogen, all procoagulant assays.
Table 5 shows adjusted Pearson correlation coefficients for the hemostatic factors based on a hypothetical and arbitrary "observed" correlation of 0.3000. The
percent change ranged from 1.0 to 104 percent (cholesterol: 11.1 percent), and only three factors (all procoagulant) reported a percent change over 30 percent.
DISCUSSION
The major findings of this study are: 1) overall, most
of the examined hemostatic and inflammatory factors
have analytical and intraindividual variation similar to
cholesterol, implying that if pathophysiologic relations
between these factors and CVD exist on the same magnitude as they do for cholesterol, the relations should
be statistically apparent; 2) accurate assessment of the
intraindividual component of variation for some
assays may require long-term evaluation (>12 weeks);
and 3) within the groups of inflammation, fibrinolysis,
and hemostasis, some factors appeared better suited to
epidemiologic study than others, which may help in
selecting factors for analyses, or interpreting results
which include these factors.
Previous studies have compared the biovariability of
factors relative to each other (8, 11), and have sugTABLE 4. Validity coefficients with estimates of number of
replicate specimens needed to achieve validity coefficient
(pn) of 0.91 In a study of 26 apparently healthy Individuals
carried out over 24 weeks at the Laboratory for Clinical and
Biochemical Research at the University of Vermont,
Burlington, Vermont, November 1992 to April 1993
Hemostatic (actor*
Validity coefficient
(PJ
No. of measurements to
achieve p of 0.91
t-PA ag (ng/ml)
PAI ag (ng/ml)
PAI activity (lU/ml)
t-PA/PAl (ng/ml)
PAP (nM)
D-Dlmer (ng/ml)
0.95
0.83
0.90
0.94
0.80
0.85
1
2
2
1
3
2
Fibrinogen (mg/dl)
F1-2(nM)
TAT (u.g/lrter)
FPA (ng/ml)
0.74
0.76
0.92
0.49
4
4
1
15
FXIa-aiAT(nM)
CRP (fig/ml)
0.99
0.88
1
2
Cholesterol (mg/dl)
0.91
1
See table 1 for definitions of abbreviations.
Am J Epidemiol Vol. 149, No. 3, 1999
265
TABLES. Pearson correlation coefficients (r = 0.30),
adjusted* and ranked by percent change In r, In a study of 26
apparently healthy Individuals carried out over 24 weeks at
the Laboratory for Clinical and Biochemical Research at the
University of Vermont, Burlington, Vermont, November 1992
to April 1993
Hemostatic factort
FXIa-aiAT(nM)
t-PA ag (ng/ml)
t-PA/PAl (ng/ml)
TAT (jig/ml)
Cholesterol (mg/dl)
PAI-1 activity (lU/ml)
CRP (ng/ml)
D-Dimer (ng/ml)
PAI (ag) (ng/ml)
PAP (nM)
F1-2(nM)
Fibrinogen (mg/dl)
FPA (ng/ml)
Adjusted correlation
% change
0.3030
0.3158
0.3191
0.3261
0.3296
0.3333
0.3403
0.3529
0.3614
0.3750
0.3947
0.4054
0.6122
1.0
5.3
6.4
8.7
9.9
11.1
13.4
17.6
20.5
25.0
31.6
35.1
104.0
• Pearson correlation coefficients were adjusted by the method
of Liu et al. (29) for inter- and intraindividual variability,
t See table 1 for definitions of abbreviations.
gested an arbitrary limit of acceptable analytical and
intraindividual variation (9). We chose to compare the
biovariability data of our hemostatic factors with
cholesterol because it is a well-established CVD risk
factor. Our results suggest that while there is a range,
most of the examined thrombotic factors have a similar statistical power to detect physiologic relations as
cholesterol, at least in bivariate correlations.
For several factors, our variability results are similar
to those reported in a short-term variability study in
middle-aged persons (11, 12) and in persons with
underlying CVD (9). In this long-term study, FPA and
fibrinogen had higher and lower reliability, respectively. Comparison of the analytical variability data for the
FPA assay indicates the increase in the reliability factor
in our study may be due to less assay variability,
whereas the decrease in the reliability factor of the fibrinogen assay may be due to increased intraindividual
variation. Because of the variability associated with
processing each sample (split samples were used for
the duplicates), and batch-to-batch changes were not
estimated in this study, our analytical variability should
be considered a minimum for our assay methods.
In several cases, the intraindividual coefficient of
variation increased over the first 12 weeks of examination, which suggests that accurate assessment of the
intraindividual component of variation may require
extended observation. The Atherosclerosis Risk in
Communities (ARIC) group (11) has reported that age,
sex, and variable level do not significantly influence
intraindividual variation, and de Maat et al. (9) showed
that the intraindividual variability of fibrinogen was
not increased in patients with underlying disease com-
266
Sakkinen et al.
pared with healthy controls, nor influenced by a polymorphism in the fibrinogen gene. However, it is possible that unexplored genetic polymorphisms may contribute to the length of time a variable must be studied
to accurately assess intraindividual variation.
The gradation of factors within subgroups of hemostatic factors implies that some are better suited to epidemiologic study than others. The methods that we used
to evaluate variables in this study are driven by the relative intraindividual variation and interindividual variation, and do not reflect the analytical variation directly.
In fact, the variables with the lowest analytical variation
in the procoagulant and inflammation groups (PAP, fibrinogen, and CRP) were not the variables with the best
index of individuality, or factors which required the
fewest replicate specimens (t-PA, TAT, and FXIaa l AT). The selection of variables will also be tempered
by the pathophysiology of the epidemiologic questions.
Although the methods introduced to correct for variation in bivariate statistics do not apply to multivariate
models (4-6), they may still be useful in interpretation
or in cautioning interpretation of multivariate models.
Statistics from bivariate correlations may be used in
deciding what factors to include in building multivariate models. Our results indicate that an observed correlation coefficient may be diminished by as much as
one-half of its actual value by failing to account for
intraindividual and analytical variation. This may lead
to underestimating the strength of the true correlation
between two variables.
The major weaknesses of this study are a limited number of subjects, reducing the ability to stratify on factors
such as disease and smoking to assess their role in factor variation and stability. The strengths are a repeated
measures study design, covering an extended period of
time, with appropriate blood collection and storage procedures. We also examined a large number of thrombosis risk factors within different major subgroups of coagulation, fibrinolysis, and inflammatory factors, some of
which have never been examined before, and others
which have not been examined over a long time period.
ACKNOWLEDGMENTS
This work was supported by PHS award RO1 HL46696
(Dr. Tracy) and PHS Training award T32 HLO7594 (Dr.
Sakkinen).
REFERENCES
1. Fuster V, Badimon L, Badimon J, et al. The pathogenesis of
coronary artery disease and the acute coronary syndromes: part
1. N Engl J Med 1992;326:242-50.
2. Fuster V, Badimon L, Badimon J, et al. The pathogenesis of
coronary artery disease and the acute coronary syndromes: part
2. N Engl J Med 1992;326:310-18.
3. Kleinbaum D, Kupper L, Morganstem H. Epidemiologic
research. Belmont, CA: Lifetime Learning Publications, 1982.
4. Phillips A, Smith G. How independent are "independent" effects? Relative risk estimation when correlated exposures are
measured imprecisely. J Clin Epidemiol 1991;44:1223-31.
5. Greenland S. The effect of misclassification in the presence of
covariates. Am J Epidemiol 1980; 112:564-9.
6. Kupper L. Effects of the use of unreliable surrogate variables
on the validity of epidemiologic research studies. Am J
Epidemiol 1984; 120:643-8.
7. Fuller W. Measurement for error models. New York: John
Wiley & Sons, 1987.
8. Thompson S, Martin J, Meade T. Sources of variability in
coagulation factor assays. Thromb Haemost 1987;58:1073—7.
9. de Maat M, de Bart A, Hennis B, et al. Interindividual and
intraindividual variability in plasma fibrinogen, t-PA antigen,
PAI activity, and CRP in healthy, young volunteers and
patients with angina pectoris. Arterioscler Thromb Vase Biol
1996;16:1156-62.
10. Rosenson R, Tangney C, Hafner J. Intraindividual variability
of fibrinogen levels and cardiovascular risk profile.
Arterioscler Thromb 1994; 14:1928-32.
11. Nguyen N, Ghaddar H, Stinson V, et al. ARIC hemostasis
study—IV. Intraindividual variability and reliability of hemostatic factors. Thromb Haemost 1995;73:256-60.
12. Chambless L, McMahon R, Brown S, et al. Short-term intraindividual variability in lipoprotein measurements: the
Atherosclerosis Risk in Communities (ARIC) study. Am J
Epidemiol 1992;136:1069-81.
13. Macy E, Meilahn E, Declerck P, et al. Sample preparation for
plasma measurement of plasminogen activator inhibitor-1 antigen in large population studies. Arch Path Lab Med 1993;
117:67-70.
14. DeClerck P, Alessi M, Verstreken M, et al. Measurement of
plasminogen activator inhibitor 1 (PAI-1) in biological fluids
with a murine monoclonal antibody based enzyme-linked
immunosorbent assay. Blood 1988;71:220-5.
15. Tracy R, Bovill E. Plasminogen and alpha-2-antiplasmin. In:
Williams hematology. Beutler E, Lichtman M, Coller B, et al,
eds. New York: McGraw-Hill, 1995:L106-L107.
16. Holvoet P, Cleemput H, Collen D. Assay of human tissue-type
plasminogen activator (t-PA) with an enzyme-linked
immunosorbent assay (ELISA) based on three murine monoclonal antibodies to t-PA. Thromb Haemostas 1985;54:
684-97.
17. Declerck P, Verstrecken M, Collen D. Measurement of different forms of plasminogen activator inhibitor 1 (PAI-1)
using various monoclonal antibody-based enzyme-linked
immunosorbent assays. Fibrinolysis 1990;4:132-3.
18. Holvoet P, de Boer A, Verstreken M, et al. An enzyme-linked
immunosorbent assay (ELISA) for the measurement of plasmin-alpha-2-antiplasmin complex in human plasma—application to the detection of in vivo activation of the fibrinolytic
system. Thromb Haemostast 1986;56:124-7.
19. Declerck P, Mombaerts P, Holvoet P, et al. Fibrinolytic
response and fibrin fragment D-dimer levels in patients with
deep vein thrombosis. Thromb Haemostast 1987;58:1024-9.
20. Tracy R, Jenny R, Williams B, et al. Active site-specific assays
for enzymes of the coagulation and fibrinolytic pathways. In:
Methods of enzymology. Lorand L, Mann K, eds. San Diego,
CA: Academic Press, 1993:514-24.
21. Tracy R , Bovill E. Methods for determining fibrinopeptide A.
In: Beutler E, Lichtman M, Coller B, et al, eds. Williams
hematology. New York: McGraw-Hill, 1995:L96-L97.
22. Macy E, Hayes T, Tracy R. Variability in the measurement of
C-reactive protein in healthy subjects: implications for reference interval and epidemiological applications. Clin Chem
1997;43:52-8.
23. Clauss A. Gerinnungs-physiologische schnell-methode zur
Am J Epidemiol Vol. 149, No. 3, 1999
Variability in Hemostasis, Fibrinotysis, and Inflammation
bestimmung des Fibrinogens. Acta Haematol 1957:17:
237^16.
24. Geffken D, Keating F, Kennedy M, et al. The measurement of
fibrinogen in population-based research: studies on instrumentation and methodology. Arch Pathol Lab Med 1994;
118:1106-9.
25. Cushman M, Cornell E, Howard P, et al. Laboratory methods
and quality assurance in the Cardiovascular Health Study. Clin
Chem 1995 ;41:264-70.
26. Fraser C, Harris E. Generation and application of data on bio-
Am J Epidemiol
Vol. 149, No. 3, 1999
27.
28.
29.
30.
267
logical variation in clinical chemistry. Crit Rev Clin Lab Sci
1989;27:4O9-37.
Searle S. Linear models. New York: John Wiley & Sons, 1971.
Cochran W. The distribution of the largest of a set of estimated variances as a fraction of their total. Ann Eugen 1941; 11:47.
Liu K, Stamler J, Dyer A, et al. Statistical methods to assess
and minimize the role of intra-individual variability in obscuring the relationship between dietary tipids and serum cholesterol. J Chronic Dis 1978;31:397^tl2.
SPSS/PC+ V2.0. Chicago, IL: SPSS Inc, 1988.