Marcus Lind MD, Department of Medicine, Skellefteå County Hospital, 931 86 Skellefteå, Sweden. (fax: +46-910-771-657; e-mail: Marcus.firstname.lastname@example.org).
Abstract. Lind M, Boman K, Johansson L, Nilsson TK, Slunga Järvholm L, Jansson J-H (Address: Department of Public Health and Clinical Medicine, Umeå University; Department of Clinical Chemistry, Örebro University Hospital). Von Willebrand factor predicts major bleeding and mortality during oral anticoagulant treatment. J Intern Med 2012; 271: 239–246.
Aims. Oral anticoagulation (OAC), predominantly with warfarin, is an effective treatment to prevent thromboembolic events. Serious bleeding is a frequent and feared treatment complication. In this longitudinal cohort study of OAC-treated patients, we aimed to evaluate the relationship between von Willebrand factor (VWF) levels and risk of bleeding complications, cardiovascular mortality and all-cause mortality.
Methods and results. A total of 719 patients receiving warfarin treatment were observed for a mean duration of 4.2 years. All bleeding complications causing hospitalization were registered and classified into clinically relevant bleeding (CRB) and major bleeding. Ischaemic stroke, peripheral arterial embolism, myocardial infarction, and death were also recorded. We identified 113 cases of CRB and 73 of major bleeding. In total, 161 deaths occurred during follow-up with cardiovascular disease identified as the cause of death in 110 patients. Patients in the highest tertile of VWF had a significantly increased risk of bleeding complications: hazard ratio (HR) 2.53 (95% CI 1.41–4.56) for major bleeding and HR 2.19 (95% CI 1.38–3.48) for CRB. VWF, expressed either in tertiles or as a continuous variable, showed a significant association with cardiovascular mortality (HR 1.68, 95% CI 1.40–2.01) and all-cause mortality (HR 1.77, 95% CI 1.52–2.05). In multivariate Cox regression analysis, the findings remained significant after adjusting for age, high-sensitivity C-reactive protein and creatinine.
Conclusions. Patients with high levels of VWF had an increased risk of bleeding complications, cardiovascular mortality and all-cause mortality during OAC treatment. Our findings imply that the use of VWF as a risk marker for thromboembolic events is complicated by the association of VWF with bleeding complications.
Treatment with oral anticoagulation (OAC), predominantly warfarin, is widely used and is highly effective in preventing thromboembolic events . A major side effect that often limits the use of warfarin is bleeding complications, with reported frequencies varying from 3.9 to 13 per 100 treatment-years in elderly populations [2, 3]. Conditions that increase the risk of bleeding complications such as diabetes, previous stroke, increased age, impaired renal function and hypertension also increase the risk of thromboembolic complications . It is often difficult to determine in which patients the benefit of long-term OAC treatment will outweigh the potential risks. Finding biochemical markers that could guide clinicians to decide whether to initiate or withhold OAC treatment would be of importance. It is reasonable to assume that haemostatic factors could be related to risk of bleeding complications.
von Willebrand factor (VWF) is involved in primary haemostasis. A quantitative deficiency of or qualitative defect in VWF is associated with von Willebrand disease . Low plasma levels have also been associated with an increased risk of haemorrhagic stroke . High levels of VWF on the other hand are thought to reflect endothelial dysfunction or damage  and have been associated with increased risk of stroke [8, 9], cardiovascular death in survivors of myocardial infarction , first-ever myocardial infarction [11–13] and overall mortality during OAC treatment . However, the results of a single case–control study suggested that high levels of VWF could also increase the risk of bleeding complications during OAC treatment , but larger longitudinal studies have not been conducted.
The aim of this longitudinal cohort study of OAC-treated patients was to evaluate the relationship between VWF levels and risk of bleeding complications as well as cardiovascular and all-cause mortality.
The study population has previously been described elsewhere . The study cohort was recruited from the OAC clinics at Skellefteå County Hospital, Skellefteå, Sweden and Umeå University Hospital, Umeå, Sweden in June 1996, as shown in Fig. 1. In total, 1204 patients attended one of the two OAC clinics. Of those patients, 957 had a planned treatment time of more than 3 months and were considered to be on long-term treatment and eligible for the study. Consent forms were sent to all patients on long-term treatment; 64 failed to answer and 46 declined to participate. Further patients were excluded because of missing blood samples (n = 102), treatment cessation (n = 15) or death (n = 11). Blood samples were obtained at inclusion for 719 patients (356 from Skellefteå and 363 from Umeå OAC clinics) who comprised the final study cohort. All patients received OAC treatment for at least 2 months prior to blood sampling.
Indications for treatment were available from registers at the OAC clinics. Additional data regarding diabetes, previous peptic ulcer or bleeding peptic ulcer, hypertension, weight and height were available through questionnaires from the Skelleteå OAC clinic. International normalized ratio (INR) values at blood sampling were obtained from the registry at the OAC clinic in Skellefteå.
The study was approved by the Research Ethics Committee of Umeå University.
Blood sampling and laboratory procedures
Venous blood samples were drawn with minimum stasis and collected in standard siliconized plasma tubes containing 0.13 mol L−1 sodium citrate. After centrifugation, the plasma samples were frozen and stored at −70 °C until analysed. Samples from the study cohort were analysed at the same time and location. Laboratory staff had no knowledge of event status.
von Willebrand factor was measured (in kU L−1) using an enzyme-linked immunosorbent assay (DAKO, Denmark). In the present study, the interassay coefficient of variation (CV) at a level of 1.05 kU L−1 was 4.1% (n = 81). High-sensitivity C-reactive protein (hsCRP) was determined with an automated method (IMMULITE Diagnostic Products Corporation, Los Angeles, CA, USA); the interassay CV was <6% in the present study. Creatinine was analysed using a Hitachii 911 multianalyser (Roche, Mannheim, Germany) with an enzymatic method (Crea plus Roche/Boehringer, Mannheim, Germany). INR values were determined at each hospital laboratory.
The date of blood sampling was set as the date of inclusion (from 1 June 1996). Patients were followed longitudinally from 1 June 1996 until death, bleeding, cessation of OAC treatment or 1 January 2002. All medical records from the Departments of Medicine, Surgery, Ear, Nose and Throat, Ophthalmology, Urology, Neurology, Neurosurgery and Orthopaedic Surgery were reviewed during the study period. One patient moved away from the region and was followed to the date of migration. Bleeding complications causing admission to hospital or death, myocardial infarction, ischaemic stroke and peripheral arterial embolism were recorded and classified by a panel of three clinicians (ML, JHJ and LJ). Major bleeding was defined according to the criteria of Schulman and Kearon  as fatal bleeding and/or symptomatic bleeding in a critical area or organ and/or bleeding causing a fall in haemoglobin level of 20 g L−1 or more or leading to transfusion of two or more units of whole blood. Clinically bleeding encompasses major bleeding or overt bleeding that caused admission to hospital or necessitated prolonged hospital care, which did not meet the criteria for major bleeding. All other bleeding episodes were classified as minor and were excluded. The INR at the time of bleeding was determined from hospital records or from the registers of the OAC clinics. Cardiovascular mortality and morbidity included myocardial infarction, stroke, transient ischaemic attack, ruptured aortic aneurysm and peripheral arterial embolism.
The cause of death was registered and classified according to the death certificate. The cause of death could be classified in all except one case. Investigators who were responsible for classifying events were blinded to the biochemical results.
Spearman correlation coefficients were used to evaluate potential relations between variables. The distributions of VWF and hsCRP were skewed and, therefore, natural logarithm transformed with increments of hazard ratio (HR) presented for the standard deviation. The assumption of proportional hazards was verified graphically using Kaplan–Meier survival curves. Statistical analyses were performed with VWF as a continuous variable and by categorizing the data into tertiles, with the lowest tertile as the reference group. Univariate Cox regression analysis was performed on each of the variables to estimate the HR and 95% confidence interval (CI). Multivariate Cox regression analysis was performed to estimate the effects of different determinants when controlling for other factors. As age, hsCRP and creatinine may be related to both VWF levels and outcome, they were considered to be potential confounders and were included in the multivariate analysis if P <0.2 in the univariate analyses. A separate Cox regression model was performed on each of the variables, and patients were censored at the first event for each outcome. In addition, data on hypertension, diabetes, body mass index (BMI), previous peptic ulcer and previous bleeding peptic ulcer were available from the Skellefteå OAC clinic and were tested for association with bleeding with univariate Cox regression analysis. Direct age-adjustment was made using 10-year intervals.
A value of P <0.05 (two-sided) was considered statistically significant. spss version 15.0 (SPSS Inc., Chicago, IL, USA) was used for all statistical analyses. Individuals with missing values (none for VWF, two for hsCRP and 32 for creatinine) were excluded from the statistical analyses.
The follow-up time of the cohort was 3001 treatment-years, and the mean follow-up in individual patients was 4.2 years. During the study, 113 clinically relevant bleeding events and 73 cases of major bleeding occurred. Bleeding location, clinical characteristics at baseline, laboratory results and indications for treatment with OAC are shown in Table 1. A total of 161 patients died; among these patient, 110 deaths were classified as cardiovascular deaths. Nonfatal cardiovascular events were registered in 75 patients, and 136 patients were withdrawn from OAC treatment.
Table 1. Baseline characteristics of the study cohort
Study cohort (n = 719)
INR, international normalized ratio; OAC, oral anticoagulant.
aAvailable for patients monitored at the Skellefteå OAC clinic.
Mean age at inclusion, years
70 ± 11
Female, n (%)
Follow-up time, mean (SD), years
4.2 ± 1.8
von Willebrand factor, kU L−1
1.61 ± 0.59
von Willebrand factor, tertiles kU L−1
C-reactive protein, mean (SD), mg L−1
7.0 ± 14.9
Creatinine, mean (SD), μmol L−1
83 ± 28
Indications for OAC treatment, n (%)
Prosthetic heart valve
Peripheral arterial thromboembolism
Bleeding location (for clinically relevant bleeding), n (%)
von Willebrand factor was significantly correlated with age (r = 0.31), hsCRP (r = 0.29) and creatinine (r = 0.23), but not with sex. INR data at the time of sampling were available from the Skellefteå OAC clinic. Of the patients for whom data were available, 9% had an INR value below 2.0, 5% had an INR above 3.5, and 86% had an INR between 2.0 and 3.0. No significant correlation between VWF and INR at the time of sampling was found (r = 0.043, P =0.42).
The most common sites of clinically relevant bleeding were gastrointestinal (34%), intracranial (16%) and soft tissue (13%). The annual risks of clinically relevant and major bleeding events were 3.9% and 2.4%, respectively. Fatal bleeding occurred in 11 patients (0.4% per treatment-year).
von Willebrand factor as a continuous variable and the highest tertile of VWF as well as age was significantly associated with both clinically relevant and major bleeding in the univariate analyses (Table 2). In a multivariate model including age and VWF, the highest tertile of VWF continued to show significant association with both clinically relevant and major bleeding. Age-adjusted incidence of clinically relevant bleeding was 3.0 per 100 treatment-years in the lowest tertile of VWF and 3.7 and 6.0 per 100 treatment-years in the intermediate and highest tertiles, respectively (Fig. 2). The incidences of major bleeding per 100 treatment-years were 1.6, 2.3 and 4.1 in the lowest, intermediate and highest tertiles of VWF, respectively (Fig. 2).
Table 2. Univariate and multivariate Cox regression analysis showing the hazard ratios (with 95% confidence intervals) for different endpoints
Clinically relevant bleeding (n = 113)
Major bleeding (n = 73)
Cardiovascular mortality (n = 110)
All-cause mortality (n = 161)
hsCRP, high-sensitivity C-reactive protein; VWF, von Willebrand factor.
VWF, hsCRP and creatinine were logarithmically transformed and presented with 1 standard deviation (SD) increment. –, P-value > 0.2 in the univariate analysis and not included in the multivariate analysis.
aMultivariate model included age and VWF.
bMultivariate model included HsCRP, VWF, age and creatinine.
Age per 10 years
VWF per 1 SD
Tertiles of VWF
Low (n = 239)
Intermediate (n = 240)
High (n = 240)
hsCRP per 1 SD
Creatinine per 1 SD
Sex, hsCRP and creatinine showed no significant association with bleeding complications in univariate analyses.
Clinically relevant bleeding and major bleeding occurred at INRs greater than 3.5, at the time of the bleeding event, in 24% and 27% of cases, respectively. Furthermore, after excluding patients with an INR above 3.5 at the time of bleeding, there were 88 cases of clinically relevant bleeding. Multivariate Cox regression analysis continued to show a significant association between bleeding and VWF, both as a continuous variable and for the highest tertile. HR values were similar to those in the total study population. INR at baseline was not associated with bleeding events (P =0.35) in univariate Cox regression analysis.
In patients from the Skellefteå OAC clinic, the incidence of hypertension (45%), diabetes (13.5%), previous peptic ulcer (13.5%) and previous bleeding peptic ulcer (5.9%), as well as the BMI (mean 25.4, SD 3.8 kg m−2), were available (356 subjects) and were not significantly associated with clinically relevant or major bleeding in univariate analyses.
Cardiovascular morbidity and mortality
In univariate analysis, VWF, both as tertiles and as a continuous variable, as well as hsCRP, creatinine and age were significantly associated with cardiovascular mortality. In the multivariate model, we included age, hsCRP, creatinine and VWF. Age and VWF, but not hsCRP or creatinine, continued to show significant associations with cardiovascular mortality. Age-adjusted incidence of bleeding and risk of cardiovascular mortality for tertiles of VWF is shown in Fig. 2.
In univariate analysis, a significant association was shown between VWF and nonfatal cardiovascular events (HR 2.52; 95% CI 1.24–5.10). When adjusted for age in a multivariate model, the association was no longer significant (HR 1.80; 95% CI 0.85–3.82).
von Willebrand factor as a continuous variable and expressed as tertiles as well as hsCRP and age was significantly associated with total mortality both in univariate and multivariate Cox regression analysis. Creatinine was significantly associated with total mortality in univariate, but not in multivariate analysis. All-cause mortality per 100 treatment-years adjusted for age was 2.6, 6.1, and 9.7 in the lowest, middle and highest tertiles of VWF, respectively. After age stratification, patients over 70 years with VWF levels above the median had an HR of 6.6 (95% CI 3.8–11.7) compared with patients ≤70 years with levels of VWF below the median. No significant association was observed between sex and all-cause mortality.
When adjusting for hypertension, diabetes and BMI data, available from the Skellefteå OAC clinic, the association between VWF and both cardiovascular and all-cause mortality remained significant.
In this longitudinal cohort study, we found that a high level of VWF (the highest tertile, i.e. above 1.83 kU L−1) was associated with an increased risk of both major and clinically relevant bleeding events. To our knowledge, this is the first prospective study in which elevated levels of VWF have been found to be significantly associated with bleeding complications during OAC treatment.
We also found that high levels of VWF (the intermediate and highest tertiles, i.e. above 1.39 kU L−1) were associated with cardiovascular death. This is in line with previous findings that VWF is associated with cardiovascular death  as well as stroke [8, 18] and coronary heart disease [11, 12, 19]. In a previous study of OAC-treated patients, we have also shown that VWF level was associated with mortality .
Previously, VWF has been suggested to be a marker of inflammation . In addition, hsCRP has been considered to be a sensitive marker of inflammation and to be associated with cardiovascular events,  irrespective of a relation with VWF levels . In the present study, adjustment for hsCRP did not affect the association between VWF and bleeding complications or mortality, indicating that inflammation alone cannot explain our findings.
Wannamethee et al. found that impaired renal function was related to increased risk of cardiovascular events and to increased levels of endothelial markers such as VWF . A high level of VWF is also known to precede microalbuminuria in patients with diabetes [23, 24]. The results of the present study showed that VWF was related to all-cause mortality, and this relationship was independent of creatinine levels.
It is well known that low levels of VWF are associated with bleeding (i.e von Willebrand disease). The novel finding of the present study is that high levels of VWF were associated with bleeding complications. The mechanism for this association is not clear. Endothelial dysfunction and atherosclerosis are associated with high levels of VWF, but the association between endothelial dysfunction and atherosclerosis and an increased risk of bleeding is not obvious. From previous studies, we know that in patients with aortic stenosis there might be an association between low levels of highest molecular weight multimers of VWF and bleeding at normal levels of VWF antigen  and also that large functional multimers of VWF are necessary for primary haemostasis, especially under conditions of high shear stress . The majority of circulating VWF consists of proteolytic fragments of large multimers . Here, we measured the levels of VWF as antigen and not as multimers or activity; this does not necessarily mirror the activity of VWF in primary haemostasis.
There are limitations to this study. Blood samples were collected only once during OAC treatment and there are conflicting views regarding the possible effects of concomitant OAC treatment on VWF levels [28–30]. ABO blood group status is known to affect the levels of VWF [31, 32], but such data were not available. Additionally, adjustments for other potential confounders were limited because of insufficient data regarding hypertension, diabetes, smoking, weight and height at baseline.
cOur findings indicate that there is an association between elevated levels of VWF and cardiovascular events in patients receiving OAC treatment. Determination of VWF antigen has previously been suggested as a promising method for thromboembolic risk stratification of patients with atrial fibrillation, when deciding whether or not to treat with OAC . However, the utility of VWF is complicated by the finding that high levels of VWF were also associated with an increased risk of bleeding complications during OAC treatment. Thus, the clinical dilemma remains: should patients with high levels of VWF be treated with antithrombotic medications? A prospective randomized trial taking into account both the risk of bleeding and the risk of cardiovascular events would be of value in establishing the roles of VWF and other biochemical markers as predictors in patients receiving OAC treatment. We have recently proposed thrombomodulin as a biomarker associated with bleeding , and it is possible that a combined analysis of VWF and thrombomodulin could add prognostic value, in addition to clinical risk scores such as the CHADS2 score and the HAS-BLED bleeding risk score .
In conclusion, we have demonstrated in this longitudinal cohort study that high levels of VWF are associated with an increased risk of bleeding complications, cardiovascular mortality and all-cause mortality in OAC-treated patients.
This study was supported by generous grants from the County Council of Västerbotten, the Heart Foundation of Northern Sweden, the Foundation for Medical Research in Skellefteå and the Joint Committee of Northern Sweden Health Care Region.