Abstract 14747: MS Gene: Multistate Modeling of Dynamic Lifetime Risk of Coronary Artery Disease Using Electronic Health Records in the UK Biobank

Abstract

Introduction: Estimated lifetime risk for coronary artery disease (CAD) is an essential goal, yet few tools exist to dynamically update predictions with age-dependent effects. Methods: We considered 221,997 individuals (54% female) followed in the UK Biobank (UKB) Electronic Health Record (EHR) and GP (general practice) since adulthood (median entry age 24, IQR 18-37). We built a multi-state model for dynamic risk, MSGene, comprised of 10 time-dependent states based on cardiometabolic conditions. We report annual state transitions to CAD as age-dependent functions of sex, polygenic score (PRS) and smoking. We evaluate prediction relative to the Pooled Cohort Equations (PCE) for 10-year risk and to empiric values for lifetime risk. We report age-specific absolute risk reduction by integrating yearly relative reduction using imputation from randomized statin trials. Results: MSGene improved accuracy (root mean squared error [RMSE] 0.85% vs 6.56%, P<0.001) compared with PCE for 10-year risk, and has RMSE of 2.6% [SEM 0.0067] for lifetime risk ( Figure 1a ) when compared to empirical rates after validation. The relative effect of PRS on transition from health is greater earlier in life ( Figure 1b ). MSGene reinforced that individuals with low short-term risk are at high lifetime predicted risk: a healthy non-smoking male at the 80 th percentile PRS with a 2.4% (10-year) risk at age 40 has a 23.6% lifetime risk ( Figure 1c ). MSGene nimbly models dynamic transitions ( Figure 1D ): for example, for individuals with hypertension, these increase to 5.6% and 35% (SEM 0.02) respectively. Under statin therapy, predicted absolute risk reduction for those at the highest genetic risk declines from 5.9% [5.85-5.94] at age 40 to 2.8% [2.78-2.92] at age 70. Conclusions: We introduce a novel statistical method, MSGene, which accurately describes the age-dependent risk of developing CAD over the lifespan conditional on PRS and updated health status, for adaptive prevention and screening.