Intraindividual variability of von Willebrand factor and the need for repeated testing

Abstract

Background

Diagnosing von Willebrand disease (VWD) is complicated by intraindividual variation of von Willebrand factor (VWF). Current guidelines define VWD as VWF antigen (VWF:Ag) or VWF activity of <0.30 or <0.50 × 10³ IU/L depending on the presence of bleeding symptoms. However, there is no consensus on whether repeated testing is necessary in patients with VWF levels close to the cutoff.

Objectives

This study aimed to examine the intraindividual variation of VWF antigen (VWF:Ag), platelet-binding activity (VWF:GPIbR), and collagen-binding activity (VWF:CB) by use of routine patient data and to define cutoffs for reliably excluding VWD based on a single sample.

Methods

In this cross-sectional study, we extracted patient results of VWF:Ag, VWF:GPIbR, and VWF:CB analyzed at the Department of Clinical Chemistry, Aarhus University Hospital, from January 2013 to January 2024 and calculated total and nonanalytical coefficient of variation (CV_{nonanalytical}).

Results

We found a CV_{nonanalytical} of 22.03% (n = 252), 24.10% (n = 333), and 24.07 (n = 58) for VWF:Ag, VWF:GPIbR, and VWF:CB, respectively. We did not find substantial differences in CV_{nonanalytical} across subgroups based on sex, age, and VWF levels. Furthermore, we estimated a method-specific cutoff to exclude VWD based on a single blood sample at 0.67 × 10³ IU/L, 0.66 × 10³ IU/L, and 0.88 × 10³ IU/L for VWF:Ag, VWF:GPIbR, and VWF:CB, respectively.

Conclusion

We found considerable intraindividual VWF variation in patients referred for VWF testing. Our results confirm that repeated testing is crucial when VWF is low in the normal range to prevent false rejection of a true VWD diagnosis.

Essentials

• •von Willebrand factor (VWF) shows large intraindividual variation, complicating diagnosis.
• •Intraindividual variation of VWF was estimated using routine patient results from 10+ years.
• •We established laboratory-specific cutoffs to exclude von Willebrand disease based on a single sample.
• •Our approach, based on patient data, may be easily adopted by other laboratories.

1. Introduction

von Willebrand factor (VWF) is important for platelet adhesion and aggregation due to its ability to bind to platelets and collagen, respectively. Furthermore, VWF binds and stabilizes coagulation factor (F)VIII [1]. von Willebrand disease (VWD) is the most common inherited bleeding disorder, with a prevalence of up to 1 in 1000 among patients exhibiting bleeding symptoms [2]. Common symptoms include menorrhagia, bruising, mucocutaneous bleeding, epistaxis, bleeding after dental extractions, postsurgical bleeding, and excessive bleeding from wounds. Additionally, VWD may manifest as gastrointestinal bleeding, joint bleeding, gastrointestinal bleeding, bleeding in the central nervous system, or other bleeding-related symptoms [2].

VWD is caused by either a quantitative or qualitative deficiency of VWF. Types 1 and 3 are quantitative deficiencies, with type 1 being a partial VWF deficiency and type 3 being an absolute VWF deficiency. Type 2 encompasses qualitative deficiencies and is further divided into several subtypes. Type 2A arises from a defect in multimerization, whereas types 2B, 2N, and 2M are due to abnormal ligand binding to GPIb, collagen, or FVIII [2].

Since VWD can be due to both quantitative and qualitative defects in VWF, the diagnostic workup for VWD includes several laboratory assays, including VWF antigen (VWF:Ag), VWF collagen-binding activity (VWF:CB), and VWF platelet-binding activity, eg, VWF:GPIbR or VWF:RCo assays, and, in some cases, multimer analysis or VWF:FVIII binding assay. Interpretation of VWF measurements is further complicated by large intraindividual variation due to both genetic and physiologic factors [1]. Furthermore, both preanalytical and analytical variation can enhance the uncertainty on the result. For type 1 and 2 VWD, physicians may often feel that repeated testing is required to confirm the diagnosis [3], especially for VWF levels close to the laboratory’s lower reference limit. However, not all current guidelines include recommendations on whether to perform repeated testing [4,5].

The aim of this study was, therefore, to examine the intraindividual variation of VWF:Ag, VWF:GPIbR, and VWF:CB by use of real-world routine patient samples analyzed in our laboratory to define the cutoff at which VWD can reliably be excluded by a single blood sample.

2. Methods

2.1. Study population and setting

All patient results of VWF:Ag, VWF:GPIbR, and VWF:CB analyzed at the Department of Clinical Biochemistry, Aarhus University Hospital, from January 2013 to January 2024 were extracted from the laboratory information system, including sex and age of the patients.

2.2. Inclusion and exclusion criteria

We included data of patients aged 18 years or older who had had 2 to 4 separate laboratory measurements made in the period and with a minimum of 7 days between each sample to exclude hospitalized patients and patients receiving VWF infusions. A maximum of 5 years was allowed between each sample to minimize age variation. Only results from patients with a C-reactive protein value below our local limit of quantification (4 mg/L) on the day of VWF analysis were included, to avoid possible influence from acute phase reactions. In order to exclude pregnant women, we excluded patients who had undergone the double test (offered as a part of routine antenatal care in Denmark) within a period of 12 months before and 4 months after the test results.

2.3. Laboratory analysis

Analyses included were VWF:Ag, VWF:GPIbR, and VWF:CB. VWF:Ag and VWF:GPIbR were analyzed on ACL Top 550 (ILS Scandinavia). VWF:Ag analysis was made using HemosIL VWF Antigen reagent (Werfen) with an analytical range of 0.09 to 10 × 10³ IU/L. VWF:GPIbR analysis was made using HemosIL VWF Ristocetin Cofactor Activity reagent (Werfen) with an analytical range of 0.07 to 4.80 × 10³ IU/L. VWF:CB was analyzed using the Asserachrom VWF:CB ELISA kit (Stago) with an analytical range of 0.06/0.07 (depending on lowest value of calibrator in each run) to 1.40 × 10³ IU/L.

2.4. Statistical analysis

Data analysis was performed using Microsoft Excel 365. Average plasma levels in subgroups, defined by sex, age, and tertiles were characterized using median, first and third quartiles, and range. Male or female sex was determined via the patient's unique Danish central personal registry (CPR) ID. The age subgroups were defined using the patients’ age at time of their last sample collection.

An SD was calculated for each patient, and all estimates from different subsets of patients were then pooled to obtain a common SD for the subset, according to the Dahlberg formula [6]. The total intraindividual variation (total coefficient of variation [CV_total]) was calculated as the pooled SD divided by mean × 100%. The following equation was used to determine the nonanalytical coefficient of variation (CV_{nonanalytical}) [7]:

$${\text{CV}}_{\text{nonanalytical}}=\sqrt{{{\text{CV}}_{\text{total}}}^2-{{\text{CV}}_{\text{analytical}}}^2}$$

The analytical coefficient of variation (CV_analytical) was calculated based on analysis of internal control material at 2 different concentrations, low and normal (intermediate precision), throughout the inclusion period. CV_analytical varied slightly throughout the study period but was overall estimated to be as follows: VWF:Ag—1.20 × 10³ IU/L: 4% and 0.30 × 10³ IU/L: 8%; VWF:GPIbR—1.03 × 10³ IU/L: 4% and 0.3 × 10³ IU/L: 10%; VWF:CB—0.81 × 10³ IU/L: 6% and 0.42 × 10³ IU/L: 13%. When calculating CV_{nonanalytical}, we used the CV_analytical closest to the median plasma level in the relevant subgroup.

To assess whether test results in the lower range, ie,<0.1 × 10³ IU/L, or total time interval between each patient’s first and last sample significantly impacted CVs, subgroup analyses were performed for all 3 VWF parameters.

3. Results and Discussion

For VWF:Ag, we obtained results from 252 patients (176 women and 76 men). The average number of VWF:Ag analyses performed per patient was 2.48. One patient had a value of <0.09 × 10³ IU/L, which was altered to 0.08 × 10³ IU/L in order to perform statistical analyses. The CV_total and CV_{nonanalytical} for all 252 patients was 22.39% and 22.03%, respectively. The characteristics of each subgroup, including CV_total and CV_{nonanalytical}, are presented in Table 1. Men and patients aged ≥51 years had the highest median plasma levels of 1.23 × 10³ IU/L and 1.17 × 10³ IU/L, respectively. Age group seemed to influence intraindividual variability the most, as CV_{nonanalytical} was 18.16%, 30.83%, and 14.67% for age 18 to 30 years, 31 to 50 years, and ≥51 years, respectively.

Table 1.

von Willebrand factor antigen.

VWF:Ag parameters	Women (n = 176)	Men (n = 76)
Median plasma level, × 103 IU/L (range)	0.79 (0.08-2.29)	1.23 (0.29-3.89)
First quartile	0.61	0.65
Third quartile	1.15	1.97
IQR	0.54	1.32
Age (y), median (range)	40 (18-88)	55.5 (18-88)
CVtotal (%)	20.58	23.59
CVnonanalytical (%)	18.96	23.25
	18-30 y(n= 55)	31-50 y(n= 97)	≥51 y(n= 100)
Nmen	13	21	42
Nwomen	42	57	77
Median plasma level, × 103 IU/L (range)	0.73 (0.16-2.25)	0.77 (0.08-3.55)	1.17 (0.28-3.89)
First quartile	0.60	0.58	0.70
Third quartile	0.92	1.20	1.90
IQR	0.32	0.62	1.20
CVtotal (%)	18.59	31.07	16.71
CVnonanalytical (%)	18.16	30.81	14.67
	First tertile	Second tertile	Third tertile
Nmen	21	15	40
Nwomen	63	69	44
Median plasma level, × 103 IU/L (range)	0.55 (0.08-0.69)	0.83 (0.70-1.16)	1.71 (1.17-3.89)
First quartile	0.41	0.74	1.30
Third quartile	0.61	1.02	2.03
IQR	0.20	0.28	0.73
CVtotal (%)	22.11	15.83	20.74
CVnonanalytical (%)	20.61	15.31	20.35

IQR, interquartile range; CV_total, total intraindividual coefficient of variation; CV_{nonanalytical}, nonanalytical coefficient of variation.

For VWF:GPIbR, we obtained results from 333 patients (259 women and 74 men). The average number of VWF:GPIbR analyses performed per patient was 2.51. Fifteen patients had a value of <0.07 × 10³ IU/L, which was altered to 0.06 × 10³ IU/L in order to perform statistical analyses. The CV_total and CV_{nonanalytical} for all 333 patients was 26.09% and 24.10%, respectively. The characteristics of each subgroup, including CV_total and CV_{nonanalytical}, are presented in Table 2. Similar to VWF:Ag, CV differed mostly among the different age groups: CV_{nonanalytical} was 15.61%, 33.56%, and 22.48% for age 18 to 30 years, 31 to 50 years, and ≥51 years, respectively.

Table 2.

von Willebrand factor GPIbR binding activity.

VWF:GPIbR parameters	Women (n = 259)	Men (n = 74)
Median plasma level, × 103 IU/L (range)	0.64 (0.06-3.72)	0.56 (0.06-3.62)
First quartile	0.44	0.38
Third quartile	0.89	1.08
IQR	0.44	0.70
Age (y), median (range)	35 (18-84)	43.5 (18-87)
CVtotal (%)	25.82	26.51
CVnonanalytical (%)	23.80	24.56
	18-30 y(n= 110)	31-50 y(n= 132)	≥51 y(n= 91)
Nmen	18	25	31
Nwomen	92	107	60
Median plasma level, × 103 IU/L (range)	0.72 (0.06-2.21)	0.56 (0.06-6.62)	0.64 (0.06-3.72)
First quartile	0.50	0.40	0.40
Third quartile	0.97	0.79	1.03
IQR	0.47	0.39	0.63
CVtotal (%)	16.11	35.02	24.60
CVnonanalytical (%)	15.61	33.56	22.48
	First tertile	Second tertile	Third tertile
Nmen	29	20	25
Nwomen	82	91	86
Median plasma level, × 103 IU/L (range)	0.35 (0.06-0.45)	0.62 (0.46-0.80)	1.05 (0.81-3.72)
First quartile	0.17	0.55	0.92
Third quartile	0.44	0.62	1.38
IQR	0.27	0.07	0.46
CVtotal (%)	20.95	19.45	23.96
CVnonanalytical (%)	18.41	16.68	23.63

IQR, interquartile range; CV_total, total intraindividual coefficient of variation; CV_{nonanalytical}, nonanalytical coefficient of variation.

For VWF:CB, we obtained results from 59 patients (45 women and 14 men). The average number of VWF:CB analyses performed per patient was 2.15. One male patient was excluded due to value of >2.00 × 10³ IU/L. One patient had a value of <0.07 × 10³ IU/L, and 2 patients had a value of <0.06 × 10³ IU/L, which were altered to 0.06 × 10³ IU/L and 0.05 × 10³ IU/L, respectively, in order to perform statistical analyses. The CV_total and CV_{nonanalytical} for all 58 included patients was 27.36% and 24.07%, respectively. The characteristics of each subgroup, including CV_total and CV_{nonanalytical}, are presented in Table 3. Once again, CV_{nonanalytical} varied most among the different age groups, as CV_{nonanalytical} was 18.81%, 30.76%, and 22.65% for age 18 to 30 years, 31 to 50 years, and ≥51 years, respectively. CV_{nonanalytical} showed an increasing trend across tertile fractions.

Table 3.

von Willebrand factor collagen-binding activity.

VWF:CB parameters	Women (n = 45)	Men (n = 13)
Median plasma level, × 103 IU/L (range)	0.58 (0.03-1.69)	0.51 (0.06-1.28)
First quartile	0.43	0.44
Third quartile	0.68	0.57
IQR	0.25	0.13
Age (y), median (range)	36 (19-84)	33 (18-74)
CVtotal (%)	26.94	29.00
CVnonanalytical (%)	23.59	25.92
	18-30 y(n= 18)	31-50 y(n= 24)	≥51 y(n= 16)
Nmen	5	6	2
Nwomen	13	18	14
Median plasma level, × 103 IU/L (range)	0.62 (0.03-1.54)	0.55 (0.07-1.28)	0.55 (0.06-1.69)
First quartile	0.46	0.45	0.30
Third quartile	0.68	0.60	0.69
IQR	0.21	0.15	0.39
CVtotal (%)	19.74	33.39	26.12
CVnonanalytical (%)	18.81	30.76	22.65
	First tertile	Second tertile	Third tertile
Nmen	5	4	4
Nwomen	14	16	15
Median plasma level, × 103 IU/L (range)	0.36 (0.03-0.49)	0.56 (0.50-0.62)	0.75 (0.63-1.69)
First quartile	0.10	0.53	0.76
Third quartile	0.42	0.59	0.77
IQR	0.32	0.06	0.01
CVtotal (%)	20.77	22.67	26.15
CVnonanalytical (%)	16.20	18.58	25.45

IQR, interquartile range; CV_total, total intraindividual coefficient of variation; CV_{nonanalytical}, nonanalytical coefficient of variation.

Exclusion of test results <0.1 × 10³ IU/L for all 3 analyses did not lead to substantial alterations in CVs; hence, all test results are included in the final calculations. Furthermore, the effect of the total time interval between each patient’s first and last sample was investigated for each analysis, and no clear association was found (data not shown).

Since we did not find substantial differences in CV_{nonanalytical} across subgroups, we used the overall CV_{nonanalytical} in subsequent analysis. We do note that age seemed to be the most important factor for CV% for all 3 analyses, with the age group 31 to 50 years having the largest CV% of ∼30%. A possible explanation is that the known age-related increase in VWF levels occurs between age 31 and 50 years. A previous study found that, in fact, significant increase in VWF:Ag happened after the age of 40 years, but with the largest increase in the 51 to 60 year age group, which contrasts with our findings [8]. Furthermore, one could argue that this age group might be more heterogenic when it comes to factors such as smoking, overweight, and for women: hormone cycle-dependent variation and menopause.

We also calculated the CV_{nonanalytical} of VWF:Ag and VWF:GPIbR in a cohort of 35 healthy, unmedicated women aged 18 to 35 years included in a previous project [9]. Blood samples were drawn at 3-month intervals and timed to avoid variation due to their menstrual cycle. Of the 35 women, 34 provided at least 2 measurements of VWG:Ag, while 33 provided at least 2 measurements for VWF:GPIbR. In this study, we found a CV_{non-analytical} of 14.4% and 16.75% for VWF:Ag and VWF:GPIbR, respectively.

Several other estimates of the intraindividual variation of VWF have been made. According to the European Federation of Clinical Biochemistry and Laboratory Medicine Biological Variation Database, the median biological variation of VWF:Ag, VWF activity (VWF:GPIbR or VWF:RCo), and VWF:CB is 12.7%, 17.0%, and 25.6%, respectively [10]. These estimates are slightly lower than ours in the cases of VWF:Ag and VWF activity (22.03% and 24.10%, respectively). This is probably largely due to the fact that the European Federation of Clinical Biochemistry and Laboratory Medicine Biological Variation Database estimates are almost exclusively based on studies that included healthy controls [[11], [12], [13]], as is the case with our subcohort of healthy women [9]. In this study, on the contrary, we based our estimates on routine hospital laboratory data. Hence, a larger variation may be expected. Moreover, factors other than biological variation may potentially contribute to the variation observed in our data, including preanalytical variation, a time gap of up to 5 years between measurements, reagent lot variation, and so on.

Current international guidelines recommend that the cutoff for diagnosing VWD is VWF:Ag and/or platelet-dependent VWD activity, for example, VWF:GPIbR, of <0.30 × 10³ IU/L regardless of bleeding symptoms and VWF level of <0.50 × 10³ IU/L in patients with bleeding symptoms. If the lower limit of the laboratory range is <0.50 × 10³, this should be used as the cutoff [4]. However, these recommendations do not take intraindividual variation into account. Therefore, based on our findings, we estimated a cutoff at which VWD could be reliably excluded, with 95% probability, based on a single blood sample. We did so by adding the CV_{nonanalytical} × 1.96 to the lower limit of the normal range in our laboratory (1.96 corresponding to the z-value for the 97.5 percentile point) for each analysis (0.47 × 10³ IU/L, 0.45 × 10³ IU/L, and 0.60 × 10³ IU/L for VWF:Ag, VWF:GPIbR, and VWF:CB, respectively) corresponding to a bleeding phenotype. This yielded cutoffs of 0.67 × 10³ IU/L, 0.66 × 10³ IU/L, and 0.88 × 10³ IU/L for VWF:Ag, VWF:GPIbR, and VWF:CB, respectively.

For VWF:Ag and VWF:GPIbR, we further calculated a cutoff by adding CV_{nonanalytical} × 1.96 to 0.30 × 10³ IU/L (corresponding to a nonbleeding phenotype). This resulted in cutoffs of 0.43 × 10³ IU/L and 0.44 × 10³ IU/L, respectively, which are nearly identical to the lower limits of our laboratory’s reference intervals.

Mehic et al. [3] also performed repeated testing in 277 patients with 655 VWF measurements, where at least 1 measurement included both VWF:Ag and VWF:GPIbR and/or VWF:RCo. They estimated a similar cutoff level for both VWF:Ag and VWF activity of 0.80 × 10³ IU/L, based on the likelihood of a change from normal to pathologic levels [3]. Others have suggested a slightly higher cutoff of 1.00 × 10³ IU/L [14,15]. Doshi et al. [14] proposed such a cutoff in a retrospective pediatric study involving 811 patients evaluated for a suspected bleeding disorder, of whom 180 were diagnosed with VWD [14]. Similarly, Brown et al. set the cutoff at 1.00 × 10³ IU/L in a study focused on acute bleeding where a higher cutoff would be expected compared with nonacute settings [15]. Subsequently, Weyand et al. [16] attempted to validate this cutoff in a cohort of 47 women diagnosed with VWD who had had 2 to 4 measurements of VWF:Ag and VWF:RCo, respectively. This resulted in a negative predictive value of 89% [16].

The focus of this study was on the biochemical criteria for diagnosing VWD. Hence, the lack of clinical data such as bleeding phenotype disabled us to make any conclusions on whether or not VWF levels above the suggested cutoff indeed correlates with absence of significant bleeding symptoms. However, clinicians should reconsider VWD in patients with a very strong clinical suspicion despite normal VWF level at first measurement.

Our CV_analytical were calculated based on quality control data from routine use throughout the study period. This may explain the higher CV_analytical for VWF:GPIbR than previously reported [17] due to preanalytical variation, reagent lot variation, and so on. In general, ristocetin-containing assays may show higher analytical variation than other activity assays, eg, the VWF:GPIbM assay, and laboratories should take this into consideration when choosing their VWF activity assay [17].

Our results leading to method-specific cutoffs of 0.67 × 10³ IU/L, 0.66 × 10³ IU/L, and 0.88 × 10³ IU/L for VWF:Ag, VWF:GPIbR, and VWF:CB, respectively, are comparable with results from other studies. The advantage of our method is that it is easy to adopt by other laboratories since it relies on routine patient data. Indeed, differences between cutoff across studies may be attributed to variations in laboratory methods, different populations, and different methods of determining the cutoff, and cutoff should, therefore, preferably be validated locally. Importantly, our data confirm that the lack of repeated measurement when VWF is found to be low in the normal range may lead to a false rejection of a true VWD diagnosis.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work, the author M.H.K used ChatGPT (version 3.5, OpenAI) in order to improve readability and language. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.