Histo–blood group ABO system transferase plasma levels and risk of future venous thromboembolism: the HUNT study Free

The association between ABO blood groups and risk of venous thromboembolism (VTE) was first reported in the 1960s,^1,2 and more recent studies have consistently shown that individuals with non-O blood groups (ie, A, B, and AB) are at a 2-fold increased risk of VTE.^3,4 The ABO blood group is determined by the combination of 2 alleles at the ABO locus on chromosome 9,⁵ and the gene products of the ABO locus are A- and B-glycosyltransferases (GTA and GTB), encoded by the A and B alleles, respectively, which are jointly called histo–blood group ABO system transferase (BGAT). BGAT catalyzes the addition of N-acetyl-galactosamine or D-galactose, to the common precursor H-antigen, converting it into A- or B-antigens, whereas the O alleles do not encode a functional BGAT, leaving the H-antigen unchanged.⁵ 

A proposed mechanism for increased VTE risk in the non-O blood groups is through the glycosylation by BGAT of von Willebrand factor (VWF),^6,7 a key protein of the hemostatic system associated with VTE.⁸ A greater amount of A- and B-antigens expressed on VWF is associated with higher plasma levels of VWF⁹ and consequently coagulation factor VIII (FVIII), of which VWF is the transport protein, with subsequent increased VTE risk.^8,10 However, the impact of genetically determined ABO blood groups on VTE risk has been shown to be only partly mediated by VWF and FVIII.^11,12 

BGAT exists as an intracellular form located in the endoplasmic reticulum and Golgi apparatus, where BGAT most likely exerts its posttranslational glycosylation activity, and as a soluble form in plasma.^13,14 Despite the fact that the soluble form is reported to likely have no biological function in plasma,¹⁴ O’Donnell et al found a strong correlation between plasma GTA activity and the amount of A antigenic determinant expressed on circulating VWF, and suggested that the intracellular GTA expression levels within endothelial cells would parallel the levels observed in plasma.¹⁵ The A1 allele has been reported to explain ≈80% of the plasma GTA variability.¹⁶ Hence, plasma BGAT may potentially serve as a biomarker and proxy for intracellular GTA/GTB activity. Recently, we reported that plasma BGAT levels reached a Bonferroni-significant association to VTE risk in an unsupervised proteome-wide discovery study.¹⁷ However, the relationships between ABO locus, plasma BGAT levels, and VTE risk remain unsettled. Therefore, using data from a population-based case-cohort study, we aimed to investigate (i) to what extent ABO haplotypes explained the plasma BGAT levels, (ii) whether plasma BGAT levels were associated with risk of future incident VTE and its subtypes, and (iii) to what extent this potential association was attenuated by adjustments for plasma levels of VWF or FVIII.

The study population was recruited from the third survey of the Trøndelag Health Study (HUNT3),¹⁸ a population-based cohort conducted in 2006 to 2008, inviting all individuals aged ≥20 years in Nord-Trøndelag, Norway. A total of 50 800 individuals (54% of those invited) participated. All first lifetime VTE events during the first 5 years of follow-up were recorded. The study was approved by the Regional Committee for Medical and Health Research Ethics, and all participants provided written informed consent.

Baseline information was collected at inclusion through physical examination, validated self-administered questionnaires, and blood samples. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared (kg/m²). Information on arterial cardiovascular disease history (angina pectoris, myocardial infarction, and stroke) and smoking was collected via questionnaires. Nonfasting blood samples were collected from an antecubital vein into EDTA-containing vacutainer tubes. Plasma was prepared by centrifugation (2500×g for 10 minutes, at 6 °C) and frozen at −80°C in the HUNT Biobank. C-reactive protein (CRP) measurement was performed in nonfasting serum samples using latex immunoassay method (Abbott, Clinical Chemistry, IL), as previously described.¹⁹ Estimated glomerular filtration rate (eGFR) was calculated by the Chronic Kidney Disease Epidemiology Collaboration equation²⁰ from serum creatinine measurements.

ABO genotyping was conducted using an Illumina HumanCoreExome array.²¹ For haplotype assessment, we applied the approach by Goumidi et al, identifying 5 single-nucleotide polymorphisms (SNPs) at the ABO locus tagging the main haplotypes.²² A1 haplotype was tagged by rs2519093-T, A2 by rs1053878-A, O1 by rs8176719-delG, O2 by rs41302905-T, and B by rs8176743-T.

We searched for relevant International Classification of Diseases, Tenth Revision (ICD-10), codes (supplemental Table 1 [available on the Blood website]) in hospital discharge diagnosis and autopsy registries at Levanger, Namsos, and St Olav’s University Hospital to identify VTE cases during follow-up (2006-2019).¹⁷ Medical records of each potential VTE case were reviewed, VTE events were adjudicated and recorded when signs and symptoms of lower extremity deep vein thrombosis (DVT) or pulmonary embolism (PE) were confirmed by radiological procedures (ultrasonography, venography, computed tomography pulmonary angiogram, and lung ventilation perfusion scan), and treatment was initiated (unless contraindications were specified). Cases with concomitant DVT and PE were recorded as PE. A VTE was categorized as provoked if ≥1 provoking factor was present in the 3 months preceding the event: surgery, trauma, acute medical condition (acute myocardial infarction, stroke, or severe infection), immobilization (bed rest of ≥3 days, confinement to wheelchair, or plaster cast), active cancer, or other factor specifically described as provoking in the medical record (eg, intravascular catheter).

We used a case-cohort design including all incident VTE cases within the first 5 years of follow-up and a randomly sampled subcohort from the HUNT3 cohort (n = 50 800) (Figure 1). Individuals with known abdominal aortic aneurysm before baseline (n = 100) were excluded. Individuals with a history of VTE (n = 622) or cancer (n = 1908) before baseline and those without plasma samples (n = 1657) were also excluded (see supplemental Table 1 for relevant ICD-10 codes).

Among the 46 513 eligible members, 294 had an incident VTE and were included as cases. The subcohort (n = 1085) was randomly sampled, and weighted for age and sex distributions of VTE cases. A total of 19 subcohort samples were excluded because of missing or insufficient quality plasma samples. The final analytical sample included 294 VTE cases and 1066 subcohort members, with 8 also being cases. A total of 34 subcohort members and 8 VTE cases had missing genetic information. Among the 1318 participants with ABO genotype information, 17 (2 VTE cases and 15 subcohort members) had an ambiguous ABO genotype. For genetic analyses, the sample included 284 VTE cases and 1017 subcohort members, with 8 also being cases.

EDTA plasma samples from the case-cohort were sent to SomaLogic Inc for aptamer-based proteomics using SomaScan v4.1 for semiquantitative measurements of BGAT, FVIII, and VWF.²³ The amino acid sequence used to produce the BGAT aptamer is shown in the supplemental information. The coefficients of variation were reported as 5.5% for FVIII and BGAT, and 34% for VWF.²⁴ The measurements were log₂ transformed and standardized to a mean of 0 and a standard deviation (SD) of 1.

All data processing and statistical analyses were performed with R version 4.1.3 (The R Foundation for Statistical Computing, Vienna, Austria).

We used linear regression to investigate the impact of ABO haplotypes on plasma BGAT levels in the subcohort. Haplotypes were entered into the linear regression models as discrete variables coded as “0,” “1,” or “2,” corresponding to the carrier status of the ABO haplotypes tagged by the 5 SNPs. The O1 haplotype was the reference to estimate the association of A1, A2, B, and O2 haplotypes with BGAT levels under the assumption of additive effects. The regression coefficient (β) and its 95% confidence intervals (CIs) indicate the value of standardized plasma BGAT levels per allele carried. Results from regression models were presented as crude estimates and estimates adjusted for age, sex, and sample batch.

Person-time of follow-up was calculated from the date of inclusion in HUNT3 to the date of an incident VTE, migration, death, or end of follow-up (5 years after inclusion), whichever came first. Quartile cutoffs for plasma BGAT levels were determined in the subcohort. Weighted Cox proportional hazards regression models estimated hazard ratios (HRs) with 95% CIs for VTE according to increasing quartiles of BGAT levels, using the R package “cchs” (Cox model for case-cohort data with stratified subcohort selection).²⁵ The lowest BGAT quartile (quartile 1) served as the reference. Model 1 was adjusted for baseline age, sex, and sample batch. Model 2 was additionally adjusted for baseline BMI, and model 3 was further adjusted for baseline CRP levels and eGFR to address potential confounding due to obesity, inflammation, and renal function. We also investigated the association between quartiles of BGAT levels and VTE subtypes (provoked VTE, unprovoked VTE, DVT, and PE ± DVT).

Given the relatively long follow-up time (5 years for some participants), results based on single baseline measurements of BGAT levels could be influenced by regression dilution bias.²⁶ To address this, weighted Cox regression analysis was performed for overall VTE, considering the time between inclusion and thrombotic events, while keeping all subcohort members in the analyses. The HRs for VTE were estimated at each follow-up time a new VTE event occurred for quartile 4 vs quartile 1 (reference) of BGAT levels, with 5 VTE cases needed for the first estimation. The HRs were adjusted for age, sex, sample batch, BMI, CRP, and eGFR (model 3) and plotted as a function of this maximum time.

To examine the extent to which VWF and FVIII plasma levels mediated potential associations between BGAT and VTE risk, these factors were added as continuous variables to model 3, along with age, sex, sample batch, BMI, CRP, and eGFR. Separate Cox regression analyses were conducted for VWF and FVIII, as their plasma levels are closely related, circulating in a tight noncovalent complex, and can be considered as one entity.²⁷ 

The distribution of baseline characteristics across quartiles of plasma BGAT levels in the subcohort is shown in Table 1. No substantial differences in the baseline characteristics across BGAT quartiles were observed, except for lower median CRP levels in the highest quartile compared with quartiles 1 to 3. The baseline characteristics of the VTE cases and the subcohort members are shown in supplemental Table 2. The mean age and BMI, the median CRP levels, and the proportion of men were slightly higher in VTE cases than in subcohort members, whereas the mean eGFR was slightly lower in cases vs subcohort members.

The characteristics of the VTE cases at diagnosis are shown in Table 2. The mean age was 68 ± 15 years, and 62% of the VTEs were PEs with or without concomitant DVTs. Among the VTE events, 61% were provoked and 19% were related to active cancer.

The association of BGAT plasma levels across ABO haplotypes in the subcohort is shown in Figure 2 and supplemental Table 3. In linear regression models including A1, A2, B, and O2 haplotypes, with O1O1 carriers as the reference, the A1 and A2 haplotypes showed a similar effect on the standardized BGAT level per allele carried (β-coefficients of 1.34 SDs [P = 5.96 × 10⁻²⁸⁴] and 1.40 SDs [P = 7.34 $×$ 10⁻¹⁸⁵], respectively), whereas the effect on BGAT was slightly lower for the B haplotype per allele carried (1.21 SDs [P = 2.91 × 10⁻¹¹⁸]), and for the O2 haplotype, the β-coefficient was negative (−0.02 SDs [P = .816]). The ABO haplotypes explained ≈80% of the plasma BGAT variability (R² = 0.774). Further adjustment for age, sex, and sample batch had a negligible impact on the estimates. Consistent with the regression analysis, when BGAT plasma levels were analyzed across ABO genotypes, the median BGAT level for A1A1, A2A2, and BB genotypes was approximately double the median BGAT level seen for the A1O1, A2O1, and BO1 genotypes (supplemental Figure 1C).

The HRs for VTE across quartiles of plasma BGAT levels are shown in Table 3. In the analysis adjusted for age, sex, and sample batch (model 1), individuals in either the third or fourth quartile had >2-fold increased risk of VTE compared with those with BGAT levels in the lowest quartile (reference), with HRs of 2.58 (95% CI, 1.71-3.89) and 2.12 (95% CI, 1.39-3.22), respectively. The risk estimates for VTE across BGAT quartiles displayed minor differences with additional adjustments for BMI (model 2), and CRP and eGFR (model 3).

To investigate a possible influence of regression dilution on the association between elevated BGAT levels and thrombosis risk, the HRs for the highest vs lowest quartiles of BGAT were plotted as a function of maximum time between blood sampling and the VTE events during the 5 years of follow-up (Figure 3). The HRs for overall VTE in the fully adjusted model stabilized around 2.0 to 2.5 approximately from 1.5 years of follow-up and persisted throughout the study period.

Regarding the VTE subtypes, the association between BGAT levels and thrombosis risk was especially pronounced for unprovoked VTE and DVT (Table 3). The comparison of the highest vs lowest BGAT quartile yielded HRs of 3.71 (95% CI, 1.79-7.67) for unprovoked VTE and 3.28 (95% CI, 1.63-6.59) for DVT in analyses adjusted for age, sex, and sample batch. As for overall VTE, additional adjustments for BMI, CRP, and eGFR did not substantially change risk estimates.

To investigate the impact of plasma levels of VWF and FVIII on the association between BGAT and VTE risk (Figure 4), we further added VWF and FVIII to model 3 in 2 separate regression analyses, because the levels of these factors are known to be closely related. When adding VWF levels to model 3, the HRs for VTE were slightly attenuated. In comparison with the lowest BGAT quartile, the HR for the highest quartile was 2.15 (95% CI, 1.41-3.29) in model 3, and 1.99 (95% CI, 1.29-3.05) when VWF was included in model 3. The attenuation of risk estimates was more pronounced when adding FVIII to model 3, with an HR of 1.44 (95% CI, 0.92-2.24) for the highest vs lowest BGAT quartile.

Because the association between BGAT and VTE risk was particularly strong for unprovoked VTE and DVT, we performed the same analysis for these subtypes. Similar to overall VTE, the risk estimates were especially attenuated with additional adjustment for FVIII, but remained significant for both unprovoked VTE and DVT, with HRs of 2.33 (95% CI, 1.09-5.00) and 2.17 (95% CI, 1.05-4.48) for the highest vs lowest BGAT quartile, respectively.

In this case-cohort study derived from the general population, we found that the ABO haplotypes (A1, A2, B, O1, and O2) explained ≈80% of the plasma variation of BGAT, and that high plasma BGAT levels were associated with 2-fold increased risk of future VTE. Among the VTE subtypes, BGAT levels were more strongly associated with risk of unprovoked VTE and DVT than with risk of provoked VTE and PE. Adjustments for BMI, CRP, and eGFR had minor impact on the risk estimates for VTE and its subtypes. Regression dilution analysis showed that the association between BGAT levels and VTE was relatively stable throughout the study period. Notably, the association between BGAT levels and VTE was only moderately attenuated when VWF or FVIII were introduced into the regression models. Our findings suggest that plasma BGAT levels are mainly determined by the ABO locus, and the robust association between elevated BGAT levels and VTE risk, which was only partially explained by VWF and FVIII, highlights the role of BGAT as the putative biological link between the ABO locus and VTE risk.

The association between ABO blood group and VTE is well established. In large meta-analyses, individuals with non-O blood groups had ≈2-fold higher VTE risk when compared with those with O blood group.^3,4 However, genetically determined ABO blood groups revealed differences in the VTE risk within the non-O groups. Early studies found a higher VTE risk in carriers of A1 and B alleles (eg, A1A1/A1B/BB genotypes) compared with A2 allele carriers (A2A2/A2O1/A2O2 genotypes).^3,4,11 Recently, Goumidi et al supported these findings, showing that the A1 and B haplotypes were associated with an almost 1.8-fold increased risk of VTE compared with the O1 haplotype, whereas the A2 haplotype resulted in a minor increase in thrombosis risk, and the O2 haplotype tended to be slightly protective against VTE.²² Non-O blood groups seem to substantially contribute to the VTE burden at the population level. In a large prospective cohort study of blood donors from Denmark and Sweden, >30% of the VTE events could be attributed to the non-O blood group,²⁸ whereas a Norwegian population-based case-cohort study estimated that almost 19% of the VTE events in the general population were attributed to the non-O blood group.²⁹ 

To the best of our knowledge, this is the first study to explore to what extent major ABO haplotypes determine the variability of plasma BGAT levels. We found that BGAT levels were mainly genetically regulated, as major ABO haplotypes explained almost 80% of the plasma variability of BGAT. Accordingly, previous studies have reported that GTA activity was strongly related to ABO genotypes,¹⁵ and that A1 explained ≈80% of the plasma GTA variability.¹⁶ Taken together, these findings corroborate the major role of the ABO locus as determinant of BGAT levels but also suggest other regulatory mechanisms of plasma BGAT levels. Whether cellular release of BGAT to plasma reflects intracellular and extracellular pathways beyond ABO warrants further investigation.

In this case-cohort study, plasma BGAT levels were associated with risk of future VTE. Putative confounders, such as obesity, chronic inflammation, and reduced kidney function, assessed by BMI, CRP, and eGFR, respectively, had minor influence on the relationship between BGAT levels and VTE risk. With a prospective design, we could determine a clear temporal sequence between the exposure and outcome, thereby avoiding reverse causation. The relationship between BGAT levels and VTE displayed a nonlinear pattern, with an apparent plateau effect, where there was no significant difference in risk estimates derived from the third and fourth quartiles of BGAT levels compared with the reference, as the 95% CI of the estimates overlapped. Still, one might argue that a slightly higher HR in quartile 3 compared with quartile 4 could be due to a higher frequency of carriers of A2 allele in quartile 4 vs quartile 3. In our results, BGAT levels in the A2 haplotype were similar to the levels observed in the A1 haplotype, but the A2 haplotype is reported to be associated with a lower risk of VTE,²² which could have led to an attenuation of the risk estimates for the uppermost quartile. However, the distribution of ABO genotypes among VTE cases in the third and fourth quartiles of BGAT does not seem to explain the observed slight differences in risk estimates between these quartiles (data not shown). Intraindividual fluctuation of modifiable biomarkers tends to underestimate the true association between the exposure and outcome in studies with long follow-up.³⁰ However, we found that high vs low BGAT levels displayed similar risk estimates for VTE throughout the entire follow-up period, indicating minor intraindividual fluctuation of plasma BGAT levels, a finding that suggests that plasma BGAT is robustly associated with risk of future incident VTE.

Data from literature on ABO blood group and VTE have primarily focused on serologically or genetically determined ABO blood type, and not on the levels or activity of BGAT. In a French case-control study, the authors found that the mean activity of GTA and GTB was lower in VTE cases than in controls within the same ABO blood group.¹⁶ The biological mechanisms behind these findings remain to be clarified, but because measurements of glycosyltransferases were performed after the thrombotic event, reverse causation cannot be ruled out.

In subtype analyses, BGAT levels were more strongly associated with unprovoked VTE and DVT than with provoked VTE and PE. In a meta-analysis by Dentali et al, the strength of the association between non-O blood group and VTE was weaker for provoked VTE in comparison with overall and unprovoked VTEs.⁴ Furthermore, some but not all studies have previously shown a higher risk of DVT than for PE in the non-O blood groups compared with the O group.^12,28,31 The mechanism by which BGAT plasma levels would have a differential impact on the risk of DVT and PE is unknown. However, a resemblance might be speculated to the phenomenon known as factor V Leiden (FVL) paradox, where FVL is associated with higher risk of DVT than for PE.³¹ The glycosyltransferases encoded by ABO are reported to be associated with VWF and FVIII levels,¹⁶ and elevated FVIII seems to be more strongly associated with DVT than with PE.^32,33 Because FVIII increases activated protein C resistance,³⁴ the same phenotype of FVL, it is biologically plausible that BGAT may have a stronger effect on DVT than PE risk partially through FVIII.

In the present study, the association between plasma BGAT levels and risk of VTE, albeit attenuated, persisted after adjustment for plasma levels of VWF or FVIII, especially for unprovoked VTE and DVT. Accordingly, the association between the non-O blood group and VTE risk also seemed to be partially independent of VWF and FVIII plasma levels.^11,12 The attenuated relationship between BGAT and VTE risk with adjustment for VWF or FVIII in our study is consistent with the association of the ABO locus and the expression of the ABO(H)-antigens on VWF with plasma levels of VWF and FVIII.^9,15 The presence of ABO(H)-antigens on VWF particularly affects the clearance of the protein, and consequently plasma concentrations of VWF and FVIII.⁶ The ABO(H)-antigens on VWF may also impact its hemostatic function by influencing the susceptibility of VWF to proteolysis by ADAMTS13 (a disintegrin and metalloproteinase with a thrombospondin type 1 motif, member 13).⁶ Nonetheless, according to our results, the association between plasma BGAT levels and VTE cannot by explained solely by a VWF-dependent mechanism. Interestingly, data from a large genome-wide association study revealed ABO as a pleiotropic locus, being associated with circulating levels of >80 proteins.³⁵ Whether the effect of BGAT on VTE risk can be explained by other proteins regulated by the ABO locus beyond VWF and FVIII remains to be unraveled.

The main strengths of this study include the recruitment of participants from a large population-based cohort with validated VTE events, the clear temporal sequence between the exposure (BGAT) and the outcome (VTE), and the follow-up period of 5 years, which is deemed to be clinically relevant for risk assessment. Some limitations related to SomaScan technology merit attention. It was not possible to differentiate between GTA and GTB because the same aptamer targets both glycosyltransferases. It was also not possible to differentiate between GTA1 and GTA2 using the SomaScan technology. Indeed, BGAT levels across A1 and A2 haplotypes (Figure 2) or A1A1 and A2A2 genotypes (supplemental Figure 1) were essentially similar, likely indicating that the aptamer-based assessment has similar affinity for both transferases, and that any potential qualitative differences in GTA1 and GTA2 function would not be detectable in a semiquantitative assay, like SomaScan. In addition, although the O2 allele encodes for a full-length glycosyltransferase, the multiple amino acid substitutions^36,37 may impair the affinity of the SomaScan aptamer to the O2 product, which could explain the BGAT measurement close to 0 in the O2O2 and O1O2 genotypes (supplemental Figure 1). SomaScan is reported to have limited specificity for some proteins, and the aptamers may not be sensitive for changes in protein structure, such as posttranslational modifications.³⁸ However, previous studies have assessed the specificity of protein measurement by SomaScan using mass spectrometry,^39,40 and by identifying genetic variants that regulate protein levels as proxies for target specificity.^41,42 Any limited specificity of protein measurement by SomaScan in our study is expected to result in misclassification of plasma proteins that would be nondifferential with regard to the VTE status, thereby introducing the possibility of underestimation of the true associations between BGAT and VTE. The nondifferential misclassification of protein measurements could also result in an underestimation of the mediation analysis (ie, less attenuated association between BGAT and VTE risk with adjustments for VWF or FVIII). This would be more likely with adjustment for VWF as its coefficient of variation was found to be particularly high (34%) in SomaScan.²⁴ Nonetheless, when the association between BGAT and VTE was adjusted for FVIII, which displayed a considerably lower coefficient of variation in SomaScan (5.5%),²⁴ the association was only moderately attenuated and remained significant for unprovoked VTE and DVT. Plasma biomarkers and BMI were only measured at baseline. Although the follow-up period was relatively short (5 years), we cannot completely rule out that the adjustment for possible modifiable confounders, like BMI, CRP, and eGFR, could have been insufficient due to regression dilution,²⁶ leading to an underestimation of the true effect of the confounders on the association between BGAT and VTE risk, with consequently potential overestimation of the effect of BGAT on VTE risk. The 5 SNP model used to assign ABO haplotypes does not fully capture the complexity of ABO genetics, and using this approach, we might not have addressed rare alleles, reflected by 1.5% of the subcohort members with an ambiguous genotype using our approach. We cannot completely rule out that the diagnosis of a VTE event was missed during follow-up. However, this would be unlikely considering that the Levanger Hospital, Namsos Hospital, and St Olav’s Hospital are the only hospitals that provide diagnostic radiology and treatment for VTE in the region, and that the overall migration rate in the region is reported to be low.^17,43 Caution is warranted when extrapolating our findings to other ethnicities as the great majority of participants were of European ancestry.¹⁸ Finally, our novel findings on the association between BGAT and VTE risk build the foundation for further studies and need to be validated in other populations.

In conclusion, elevated plasma BGAT levels were associated with an increased risk of future incident VTE, particularly DVT and unprovoked VTE. The associations, albeit attenuated, persisted after adjusting for plasma levels of VWF and FVIII. Our findings suggest that BGAT, which comprises both glycosyltransferases A and B, is robustly associated with risk of future VTE and may have a role in the pathogenesis of VTE beyond VWF and FVIII.

The authors acknowledge SomaLogic Operating Co, Inc, as the provider of the proteomic data using the modified aptamer-based SomaScan Assay. SomaScan, SOMAmer, and SomaSignal are trademarks of SomaLogic Operating Co, Inc. The visual abstract illustration was created with BioRender.com: Onsaker A. (2025) https://biorender.com/b26j034.

The Thrombosis Research Center was supported by an independent grant from Stiftelsen Kristian Gerhard Jebsen (SKGJ-MED-012) during the establishment of this project. The SomaScan laboratory work was partly supported by grants R01HL059367 and R01HL155209 from the National Institutes of Health, National Heart, Lung, and Blood Institute.

The Trøndelag Health Study (HUNT) is a collaboration between HUNT Research Centre (Faculty of Medicine and Health Sciences, Norwegian University of Science and Technology), Trøndelag County Council, Central Norway Regional Health Authority, and the Norwegian Institute of Public Health.

Contribution: J.-B.H. conceived and designed the study; J.-B.H., K.H., C.J., A.R.F., and W.T. performed data collection; A.L.O. and K.D.H. performed statistical analysis; A.L.O., A.Y.A., D.-A.T., T.H.N., W.T., W.G., C.J., P.-E.M., K.D.H., A.R.F., K.H., V.M.M., and J.-B.H. performed interpretation of data; A.L.O., V.M.M., and J.-B.H. drafted the manuscript; A.L.O., A.Y.A., D.-A.T., T.H.N., W.T., W.G., C.J., P.-E.M., K.D.H., A.R.F., K.H., V.M.M., and J.-B.H. performed critical revision of the manuscript; and all coauthors reviewed and approved of the final version.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Asbjørn L. Onsaker, Thrombosis Research Group, Department of Clinical Medicine, UiT—the Arctic University of Norway, N-9037 Tromsø, Norway; email: asbjorn.l.onsaker@uit.no.