Predicting Atrial Fibrillation After Stroke by Combining Polygenic Risk Scores and Clinical Features.

Abstract

Background: Since treatment with anticoagulants can prevent recurrent strokes, identification of patients at risk for incident AF after stroke is crucial. We aimed to investigate whether the addition of AF polygenic risk scores (PRS) to existing clinical risk predictors could improve prediction of AF after stroke. Methods: Patients diagnosed with ischemic stroke at Massachusetts General Hospital between 2003-2017 were included. Clinical AF risk was estimated using the Re-CHARGE-AF model and genetic risk was estimated using a contemporary AF PRS from 1,093,050 variants. Patients were divided into clinical and genetic risk tertiles. Cox proportional hazards models at different follow-up windows were fit, and C-indices and percentile-based Net Reclassification Index (NRI) were used to determine improvement of clinical risk models with the addition of AF PRS. Results: Of 1004 stroke survivors, 900 (90%) were non-Hispanic White, 413 (41%) were female, and the mean age was 67 (SD 14). Of 1004 survivors, 239 (23.8%) had prevalent AF and 87/765 (11.4%) of the remaining patients developed incident AF during 5 years of follow-up. AF PRS was associated with greater risk of incident AF after stroke (HR 1.16 [95% Confidence Interval (CI) 0.94-1.44] per 1 SD increase), although the association was not statistically significant. PRS improved discrimination in the first month (AUC 0.78 [95% CI 0.70-0.82] vs AUC 0.71 [95% CI 0.60-0.82], p = 0.05), with more modest estimates across longer time windows. Conclusion: Addition of an AF PRS to clinical risk models may improve identification of individuals at risk of AF after stroke, particularly within the first month.