Connie H. Miller, Centers for Disease Control and Prevention, 1600 Clifton Road, Mailstop D-02, Atlanta, GA 30333, USA. Tel.: +1 404 639 2851; fax: +1 404 639 1638; e-mail: firstname.lastname@example.org
Summary. Tests based on three different principles are reported to measure the activity of von Willebrand factor (VWF): ristocetin cofactor (VWF:RCo), collagen binding (VWF:CB), and the so-called ‘activity ELISA’ (VWF:MoAb). We measured these and other diagnostic parameters in a population of 123 randomly selected female study controls, age 18–45 years. Type O subjects had significantly lower levels than non-O subjects in each test. Race differences were seen in all tests except VWF:RCo, with Caucasians having significantly lower levels than African-Americans. ABO differences accounted for 19% of the total variance in VWF:Ag (P < 0.0001) and race for 7% (P < 0.0001), for a total of 26%. Both effects were mediated through VWF:Ag and were independent. VWF:Ag level was the primary determinant of VWF function, accounting for approximately 60% of the variance in VWF:RCo and VWF:CB and 54% of the variance in factor VIII. The ratio VWF:RCo/VWF:Ag differed significantly by race within blood group. The median ratios were 0.97 for type O Caucasians vs. 0.79 for type O African-Americans and 0.94 for non-O Caucasians vs. 0.76 for non-O African-Americans. The ratio VWF:CB/VWF:Ag did not vary. This suggests racial differences in the interaction of VWF with GP1b but not with subendothelium. Alternatively, VWF:RCo may be regulated to maintain a relatively constant plasma level in the presence of excessive VWF:Ag. This heterogeneity within the normal population is partially responsible for the difficulty in defining diagnostic limits for von Willebrand disease.
von Willebrand disease (VWD) is a common disorder of hemostasis caused by qualitative or quantitative defects in von Willebrand factor (VWF). Since no single parameter reliably identifies the disease [1–3], its diagnosis is based on clinical features and a panel of laboratory tests, which may include a test for VWF antigen (VWF:Ag), various tests for VWF activity, a measurement of factor (F)VIII activity, ristocetin-induced platelet aggregation (RIPA), VWF multimers, and bleeding time or other measure of primary hemostasis.
Currently tests based on three different principles are in use for measuring the activity of VWF. The ristocetin cofactor assay (VWF:RCo) is based on the ability of VWF to agglutinate normal platelets in the presence of the antibiotic ristocetin . The collagen binding assay (VWF:CB) measures the binding of VWF to immobilized collagen [5,6]. An ELISA using a monoclonal antibody directed against the A1 domain of VWF, abbreviated here as VWF:MoAb, has been marketed as a measure of VWF activity ; however, its validity for that purpose has been widely questioned [8–10]. All of these tests are influenced by a number of physiological and pathological variables. Genetic factors are reported to account for 66% of the variance of VWF levels, with ABO blood type (ABO) differences a primary contributor . ABO-specific reference ranges for diagnosis have been recommended [12–14]. VWF has been observed to be significantly higher in African-Americans than in Caucasians [15,16]. We have examined a population of normal women to assess the relative contribution of these variables to the level of VWF.
Materials and methods
Subjects for this analysis were 123 women enrolled in a health maintenance organization who were randomly selected to serve as control subjects for a previously described study on women with menorrhagia [17,18]. Blood samples were obtained from these study participants at the Centers for Disease Control and Prevention on one occasion, with informed consent. Seventy-six women (62%) were self-reported as African-American. One control subject with race other than Caucasian or African-American was excluded from this analysis. Forty-five percent of the subjects had ABO blood type O. Those with blood types A, B, or AB were classified as ‘non-O’ for analysis. Age range was 18–45 years. No subject was pregnant. One-third of the subjects were using oral contraceptives. Caucasian and African-American subjects did not differ in ABO blood type distribution, frequency of oral contraceptive use, or age.
For evaluation of the performance of each test, 123 menorrhagia cases from the previously described study [17,18] were used as the test group. Cases did not differ from controls in distribution of ABO blood type, race, or age or in frequency of oral contraceptive use.
Blood was collected into evacuated siliconized glass tubes (Becton Dickinson, Franklin Lakes, NJ, USA) containing 3.2% sodium citrate in a ratio of 1 : 9 with blood. Blood processing was completed within 2 h. Platelet-poor plasma was prepared by centrifugation of whole blood at 1660 g for 20 min at 4 °C followed by repeat centrifugation of the separated plasma at 30 900 g for 20 min at 4 °C. Samples were stored in polypropylene tubes at − 70 °C.
VWF:Ag was measured by ELISA using polyclonal antiserum (Asserachrom VWF; Diagnostica Stago, Inc., Parsippany, NJ, USA). VWF:RCo was measured by agglutination of normal platelets by ristocetin in an optical system (BioData Corp., Hatboro, PA, USA), using ristocetin (American Biochemical and Pharmaceutical Corp., Marlton, NJ, USA) at a final concentration of 1 mg mL−1 and lyophilized platelets (BioData Corp) reconstituted in Tris-buffered saline (0.06 m, pH 7.5). VWF:CB was performed by a commercial method (Gradipore Ltd, Frenchs Forest, New South Wales, Australia) using microtiter plates coated with equine type III collagen and peroxidase-conjugated antihuman VWF antibodies [5,6]. VWF:MoAb (Imubind VWF Activity ELISA; American Diagnostica, Greenwich, CT, USA) was performed as instructed by the manufacturer. FVIII was measured by a one-stage assay (Diagnostica Stago) using silica as activator on an automated analyzer with mechanical end-point determination (STA; Diagnostica Stago). Direct ABO blood typing was performed by Smith Kline Beecham Laboratories (Tucker, GA, USA).
The reference standard for FVIII and VWF:Ag was a lyophilized commercial reference plasma which was standardized against the 3rd International Standard for Factor VIII and von Willebrand Factor in Plasma (3rd IS) (National Institute for Biological Standards and Control, Potters Bar, UK). For VWF:RCo, the reference standard was CAP Reference Plasma (College of American Pathologists, Northfield, IL, USA). Each standard was calibrated in our laboratory against the 3rd IS using our assay methods. Results are expressed in international units (IU). For VWF:CB, the reference plasma provided by the manufacturer was used and subsequently standardized against the 4th International Standard for Factor VIII and von Willebrand Factor in Plasma (National Institute for Biological Standards and Control).
Data for VWF:Ag, VWF:RCo, VWF:CB, VWF:MoAb, and FVIII measurements were transformed to natural logarithms (ln) for analysis in order to achieve normality. Correlation, t-test, and linear regression were performed on transformed data using SAS Version 6 (SAS Institute Inc., Cary, NC, USA). Two-tailed values are reported and a P-value < 0.05 was considered to be significant. VWF:RCo/VWF:Ag and VWF:CB/VWF:Ag ratios were compared by non-parametric tests.
Multiple linear regression was performed by stepwise regression using an F statistic significant at P < 0.05 for entry and for retention in the model. All variables retained their significance when an F statistic with P < 0.01 was used. Curve fitting was performed using Prism (GraphPad Software, Inc., San Diego, CA, USA).
Reference ranges were calculated as 2 SD above and below the geometric mean for a specified group of control subjects, using a two-tailed distribution with 2.5% of the population expected to be in the lower tail. Odds ratios (OR) were calculated from the frequency of factor levels > 2 SD below the mean as the odds of a low result in menorrhagia cases divided by the odds of a low result in controls, and 95% confidence intervals (CI) are reported.
To calculate the probability (P) of one or more abnormal results in three non-independent tests, the following formula was used: P(A or B or C) = P(A) + P(B) + P(C) − P(A and B) –P(A and C) − P(B and C) + P(A and B and C), where A= frequency of low VWF:Ag, B= frequency of low VWF:RCo, and C= frequency of low VWF:CB, ‘low’ referring to a result > 2 SD below the geometric mean.
Effects of ABO blood type, race, and age
Geometric means were calculated for VWF:Ag, VWF:RCo, VWF:CB, VWF:MoAb, and FVIII in control women grouped by ABO (O vs. non-O) and race (Caucasian vs. African-American). Each test showed a significant effect of ABO; all parameters except VWF:RCo also differed by race, with means increasing in the order Caucasian type O, African-American type O, Caucasian non-O, African-American non-O (Fig. 1).
The ratio VWF:CB/VWF:Ag showed no significant difference by ABO or race, since the two components of the ratio are similarly affected by those variables. The ratio VWF:RCo/VWF:Ag differed significantly by race within ABO group. Type O Caucasians had a median ratio of 0.97 vs. 0.79 for type O African-Americans. Non-O Caucasians had a median ratio of 0.94 vs. 0.76 for non-O African-Americans.
No significant effect of age was seen in this population with the limited age range of 18–45 years. There were no significant differences in mean levels between subjects using and not using oral contraceptives.
Relative contribution of variables
To determine whether the ABO and race effects on the VWF activities were mediated through VWF:Ag, those measurements were adjusted for VWF:Ag by linear regression (Table 1), as previously described . Adjustment for VWF:Ag removed the ABO difference in means for FVIII and VWF:CB, while VWF:RCo and VWF:MoAb means remained different. Adjustment for VWF:Ag removed the race differences in VWF:CB, VWF:MoAb, and FVIII. ABO and race effects were examined for independence (Table 2). Adjustment of VWF:Ag for ABO failed to remove the race difference in means; conversely, adjusting for race did not remove the ABO effect. These findings suggest that the actions of the two variables are independent.
Table 1. Effect of adjusting for VWF antigen (VWF:Ag)* on ABO blood type and racial differences
Multiple regression analysis was used to assess the relative contributions of ABO, race, and VWF:Ag level to variation of the activities measured (Table 3). ABO and race accounted for 27% of the variance in VWF:Ag level. VWF:Ag level accounted for similar proportions of the variance of VWF:RCo (61%), VWF:CB (57%), and FVIII (54%). No additionally significant contribution was made by ABO or race to VWF:CB or FVIII. VWF:RCo showed additional effects of ABO and race beyond the VWF:Ag effect, increasing the proportion of the variance accounted for by these three variables to 69%.
Table 3. Proportion of variance attributable to variables affecting von Willebrand factor measurements
Proportion of total variance (P-value)
Stepwise regression was performed requiring an F-statistic with P-value < 0.05 for entry into the model. *Not significant, P > 0.05.
All VWF activity measurements were significantly correlated with VWF:Ag and with each other. Pearson correlation coefficients (r) with VWF:Ag were VWF:RCo = 0.83, VWF:CB = 0.77, and VWF:MoAb = 0.93. Correlation coefficients with VWF:RCo were VWF:MoAb = 0.84 and VWF:CB = 0.74. FVIII was correlated with VWF:Ag at r = 0.77 and with VWF:RCo at r = 0.64. Figure 2 shows the relationship of the two recognized functional assays, VWF:RCo (Fig. 2A) and VWF:CB (Fig. 2B) to VWF:Ag. Figure 3 illustrates the relationship of VWF:MoAb to VWF:Ag (Fig. 3A) and VWF:RCo (Fig. 3B). The correlation of VWF:MoAb with VWF:RCo was significantly lower than its correlation with VWF:Ag (P < 0.05).
The VWF antigen and activity measurements each formed a continuous distribution with positive skewness producing a tail toward the high end of the distribution. Transformation to natural logarithms produced distributions which approximated normality. The four race and ABO group distributions of VWF:Ag could be superimposed on the overall distribution to demonstrate the heterogeneity within the population of control subjects (Fig. 4). For example, the mean VWF:Ag level for type O Caucasians (Fig. 4a) falls in the lower tail of the distribution of VWF:Ag in non-O African-Americans (Fig. 4d).
We reexamined the group of women with menorrhagia from the case–control study from which the control data analyzed in this study were drawn [17,18], in order to illustrate the differences in performance of these tests in a population with history of bleeding. Reference ranges were calculated for each test using the control women grouped by race and ABO. Table 4 shows the frequency of results > 2 SD below the mean in the control women and in the women with menorrhagia (cases) and the ORs, calculated as the odds that a woman with menorrhagia will have a low level compared with the odds of a control having a low level. VWF:RCo, which was low in seven cases and only one control, was the best single discriminator; however, the only OR reaching statistical significance was that based on a requirement of two or more abnormal test results. This plan was devised to avoid classifying subjects as abnormal due to a single result in the lower tail of the distribution. When the lower tail contains 2.5% of the population, three tests could classify up to 7.5% of the population as abnormal, if the tests were independent. These measurements, however, are correlated. Using the observed frequencies of low results singly and in combination in the controls, we calculated the probability of a control subject having a low result in one or more of the three non-independent tests to be 2.4%; the observed rate was 3.3%. Only one control subject (0.8%) met the requirement of two low results, while eight women with menorrhagia met this requirement, giving a significant OR of 8.49 (P = 0.036). Reference ranges based on control subjects grouped by ABO alone and ranges based on ungrouped data with the scheme of two or more abnormal tests also classified one control subject (0.8%) as low but differed on the number of cases. ABO-specific ranges classified six cases as abnormal (OR = 6.26, P = 0.12), and non-specific ranges, seven (OR = 7.36, P = 0.07). Only a requirement of at least two abnormal tests using ABO- and race-specific reference ranges produced an OR that was statistically significant in this small sample.
Table 4. Discrimination of menorrhagia cases from controls using reference ranges based on race and ABO blood type
Tests abnormal (>2 SD below mean)
Controls n = 123
Cases n = 123
Odds ratio (95% CI)
Two or more tests
No ‘gold standard’ exists against which one can evaluate diagnostic tests for VWD. DNA analysis is not useful because gene defects have been identified in only a few cases of type 1 VWD, the form that is most common and most difficult to diagnose . The diagnostic tests available are subject to a number of genetic and physiological variables, three of which we examined: ABO blood type, race, and age.
Among the control women in this study, all VWF parameters measured showed ABO effects, and all except VWF:RCo also showed racial differences. The finding of higher levels of VWF:Ag and FVIII in African-American women confirms previous studies of American populations [15,16]. Similar effects have also been seen in other populations of African descent in Britain , Brazil , and South Africa , suggesting that this is part of the genetic diversity that characterizes racial groups. The race effect is different from the ABO effect in that it does not affect all VWF tests in the same way. VWF:RCo did not show a difference in mean levels by race. Werner et al. also found no racial difference in VWF:RCo in their study of children . These results contrast with those of Kadir et al., who found a racial difference in ‘VWF activity’, using the VWF:MoAb test rather than VWF:RCo. We also found a race difference using that test. Our data support the view that it is not equivalent to VWF:RCo [8–10].
In this study, ABO and race showed independent effects, both mediated primarily through VWF:Ag level. Orstavik and colleagues  in a study of Norwegian twins found genetic factors to account for 66% of the total variance in VWF:Ag. ABO differences accounted for 30% of the genetic variance, or 20% of the total. In our study, ABO accounted for 19% of the total variance in VWF:Ag level, and race accounted for an additional 7% of the variance. The majority of the variability in VWF:Ag remains unexplained. Age has been reported to have a significant effect on the parameters measured [11,15,20], but was not evident in this study with the age range of 18–45. Levels of VWF:Ag and FVIII in Caucasian women in our study were lower than those seen for type O and non-O subjects in a European control population, which included ages 17–73 . These differences may be age-related. Our data more closely match the results of the CARDIA study in the USA . That study included subjects age 23–35 and also showed racial differences. Gender differences in VWF:Ag and FVIII were not seen in two studies with younger populations [11,16], suggesting that our findings apply to both sexes. Additional variables reported to influence VWF levels but not examined in this study include phase of menstrual cycle  and Lewis blood group/Secretor status [25–27]. Other variables contributing to variation may be dependent on the role of VWF as an acute-phase reactant and a marker for endothelial cell damage.
VWF:Ag level accounted for approximately 60% of the variance in the VWF functional activities, VWF:RCo and VWF:CB. VWF:Ag level also accounted for 54% of the variance in FVIII, which is produced by a different gene but depends on binding to VWF for its stability in the circulation. In European populations, subjects with FVIII > 150 IU dL−1 have been shown to have a 5–6-fold greater risk of venous thrombosis than those with FVIII < 100 IU dL−1[23,28]. This effect appears to be independent of the level of VWF:Ag [29,30] and thus must be due to the action of other variables on FVIII level. In this study, ABO appeared to have no additional effect on FVIII beyond that of VWF:Ag. African-Americans have a higher incidence of venous thromboembolism , as do those with ABO type other than O [23,32]. Since the non-O African-American women in our study had a mean FVIII of 132 IU dL−1, a large proportion of them would fall into the category of increased risk based on FVIII level alone. This finding may have significant implications, since risk factors identified in Caucasians, such as FV Leiden, are not common in the African-American population .
The ratio VWF:RCo/VWF:Ag was significantly lower in African-Americans, while VWF:CB/VWF:Ag was not. This may reflect the presence in normal African-Americans of a subset of VWF which functions poorly in binding to platelet membrane glycoprotein Ib (GPIb) but maintains its role in attachment to subendothelium. It is also possible that VWF:RCo is regulated to maintain a relatively constant plasma level of VWF:RCo in the presence of excessive VWF:Ag. We have previously shown this group of African-Americans to have a reduction in RIPA, which measures the function of the individual's platelets, as well as plasma VWF . Twenty-three percent of African-Americans and only 2% of Caucasians had RIPA < 50% (P < 0.0001). There was no difference between cases and controls in the frequency of this defect, and VWF multimers appeared normal. This effect has been previously described [34,35]. Low RIPA did not predict a diagnosis of VWD in these African-American groups; it may serve to compensate for the higher levels of VWF:Ag present.
The mechanism of the ABO blood group differences in VWF is unknown. A study using a combined linkage association method in families has suggested that variation of VWF level results from a functional effect of the ABO genotype rather than simply linkage disequilibrium with a nearby site . ABO blood group antigens are present on VWF in the A1 domain, which contains the binding site for GPIb . Removal of A or B sugars from purified plasma VWF decreases VWF:RCo activity but does not decrease VWF antigenic reactivity or binding to collagen . Type O individuals, however, have lower levels of all three, suggesting that ABO antigens in vivo influence the rate of synthesis or clearance of the entire VWF molecule, rather than its function or multimeric structure. African-Americans of all ABO types have a pattern similar to that seen with removal of A or B antigens, i.e. VWF:RCo activity decreased relative to VWF:Ag and VWF:CB, suggesting that they may have an alteration in other antigens important for VWF:RCo activity.
VWF:RCo has been reported to be the best single test for diagnosis of VWD [3,8,39]. Both VWF:MoAb and VWF:CB have been proposed as replacements for VWF:RCo, since their ELISA methods are more easily performed and more reproducible. Based on our findings and those of others [8–10], VWF:MoAb is not equivalent to VWF:RCo as a measure of VWF activity. VWF:CB also appears to be unacceptable as a substitute for VWF:RCo, because of its lower sensitivity. It is useful, however, in distinguishing between subtypes of VWD and thus remains an important part of the diagnostic panel. These results suggest that VWF:RCo is still the preferred test for measuring VWF function. Since VWF:RCo can now be performed with automated methods , its technical difficulty may become less of an impediment to its use.
In the normal population, VWF antigen and activity measurements form continuous distributions, similar to other continuous traits that are multifactorial, resulting from the effects of several genes as well as environmental modifiers. Superimposing the four distributions of subjects grouped by race and ABO (Fig. 4) illustrates the heterogeneity within the population considered to be ‘normal’ and the difficulty in defining diagnostic limits. Within continuous distributions, setting of diagnostic limits is arbitrary. It is as yet unclear whether bleeding occurs in individuals who have VWF activity levels below a certain threshold or those whose levels fall below the norm for their group. Improvements in diagnosis will result from development of novel tests. Additional technical or statistical manipulations are not likely to improve discrimination substantially using these tests with continuous variation.