AUTHORS
Samuel Short, MD, Bjoernar Tuftin2, Stephan Moll, MD3, Laura Raffield4
AFFILIATIONS
1 University of North Carolina, Chapel Hill, North Carolina, United States
2 University of North Carolina, Chapel Hill, North Carolina, United States
3 University of North Carolina, Chapel Hill, North Carolina, United States
4 University of North Carolina, Chapel Hill, North Carolina, United States
ABSTRACT
Background: Venous thromboembolism (VTE) is a multifactorial event driven by situation, comorbid risk factors, and genetic predisposition. Polygenic risk scores (PRSs) are under intense investigation to understand how genetics might influence clinical decision making for VTE prevention. Factor V Leiden (FVL) heterozygosity, the most common inherited thrombophilia, is included in all current VTE PRS instruments but FVL heterozygosity by itself does not influence clinical VTE risk assessment without a strong history of VTE. Assaying additional minor risk alleles could help refine risk status in unaffected FVL heterozygotes or, alternatively, this may not add clinically relevant information. Separately, different PRS instruments often rank an individual's risk inconsistently – a phenomenon known as PRS instability – so it is unclear which (if any) PRS may be clinically helpful. To bridge these gaps, we assessed the stability and predictive utility of multiple published VTE PRS scores among UK Biobank participants with FVL heterozygosity and no prior VTE.
Methods: UK Biobank is a prospective cohort of >500,000 people recruited in 2006-10 who provided health information and biosamples at baseline, including DNA. We included only participants with FVL mutation (rs6025). Exclusion criteria were FVL homozygosity, prothrombin G20210A mutation(s), baseline anticoagulant use, and/or pre-baseline VTE. Incident VTE was defined by ICD codes from UK hospitalization and death records. VTEs were further classified as occurring in a higher-risk context – defined as surgery or hospitalization in the preceding 3 months or cancer within the preceding 1 year – or a lower-risk context without any of these. Five previously published and validated PRS scores were calculated for each participant. Stability was assessed via pairwise Spearman correlations. Predictive utility of each PRS was assessed by comparing a base Cox model adjusted for age, sex, BMI, and smoking to models adding each individual PRS. C-statistics assessed model fit to data. Continuous net reclassification indices (NRI) determined if PRS models correctly up- and down-classified predicted risk among cases and non-cases, respectively. Higher NRI indicates improved model discrimination. Analyses were repeated with events limited to VTE cases in lower-risk clinical contexts.
Results: After exclusions, 19,376 FVL heterozygotes without prior VTE were studied. Median age was 58 (IQR 50-63), 64% were female, median BMI was 27 (IQR 24-30) kg/m2, and 10.0% were current smokers. Over 10.2 years, 287 (1.5%) developed first time incident VTE: 127 (44.3%) in higher-risk and 160 (55.7%) in lower-risk contexts. Pairwise correlations between PRSs were only weak to moderate (Spearman ρ’s 0.30-0.54) with the highest correlation between the two most recently published scores derived by Thibord et al and Truong et al. Only 536 (2.8%) participants had genetically predicted risks corresponding to the highest quartile in all PRSs and 648 (3.3%) had risks corresponding to the lowest quartile in all PRSs. The base clinical model had moderately good fit to VTE incidence (C-statistic 0.65, 95% CI 0.63-0.68). Adding PRSs previously derived by Dehaan et al, Ghouse et al, or Klarin et al did not meaningfully improve model fit (C-statistics 0.66-0.67). However, the Thibord et al and Truong et al PRSs did improve fit (C-statistics [95% CIs] 0.69 [0.66-0.72] and 0.71 [0.68-0.73], respectively; likelihood ratio test p’s <0.001). All PRSs showed modest-to-moderate improvement in risk discrimination over the base model (NRIs 0.21-0.40). The highest NRI was for the Truong et al PRS (NRI 0.40, 95% CI 0.31-0.53) which owed to superior, similar reclassification of VTE risk for both cases and non-cases. Results for analyses restricted to VTE cases in lower-risk contexts were similar to the primary analyses.
Conclusions: Five published VTE PRSs showed limited correlation with one another in a large cohort of unaffected FVL heterozygotes indicating substantial PRS instability. The Thibord and Truong PRSs showed similar, marked improvements in risk prediction compared to clinical risk factors. Results support future study of these two PRSs, or future improved PRS instruments, to improve clinical risk estimation for incident first-time VTE leading to hospitalization among FVL heterozygotes. Replication is desirable in cohorts with available outpatient data.
DISCLOSURES
Samuel Short, MD: No relevant conflicts of interest to declare.