Evaluation of the Performance of the RECODe Equation with the Addition of Polygenic Risk Scores for Adverse Cardiovascular Outcomes in Individuals with Type II Diabetes

Abstract

AbstractAims/HypothesisIndividuals with T2D are at an increased risk of developing cardiovascular complications; early identification of individuals can lead to an alteration of the natural history of the disease. Current approaches to risk prediction tailored to individuals with T2D are exemplified by the RECODe algorithms which predict CVD outcomes among individuals with T2D. Recent efforts to improve CVD risk prediction among the general population have included the incorporation of polygenic risk scores (PRS). This paper aims to investigate the utility of the addition of a coronary artery disease (CAD), stroke and heart failure risk score to the current RECODe model for disease stratification.MethodsWe derived PRS using summary statistics for ischemic stroke (IS) from the coronary artery disease (CAD) and heart failure (HF) and tested prediction accuracy in the Penn Medicine Biobank (PMBB). A Cox proportional hazards model was used for time-to-event analyses within our cohort, and we compared model discrimination for the RECODe model with and without a PRS using AUC.ResultsThe RECODe model alone demonstrated an AUC [95% CI] of 0.67 [0.62-0.72] for ASCVD; the addition of the three PRS to the model demonstrated an AUC [95% CI] of 0.66 [0.63-0.70]. A z-test to compare the AUCs of the two models did not demonstrate a detectable difference between the two models (p=0.97)Conclusions/InterpretationIn the present study, we demonstrate that although PRS associate with CVD outcomes independent of traditional risk factors among individuals with T2D, the addition of PRS to contemporary clinical risk models does not specifically improve the predictive performance as compared to the baseline model.Research in ContextEarly identification of individuals with T2D who are at greatest risk of cardiovascular complications can lead to targeted intensive risk-factor modification with the aim of altering the natural history of the disease.Current approaches to risk prediction tailored to individuals with diabetes are exemplified by the RECODe algorithms which predict both individual and composite CVD outcomes among individuals with T2D.We sought to determine if the addition of a polygenic risk score to current clinical risk models improve predictive modeling of adverse cardiovascular events in individuals with type II diabetes.We demonstrate that although PRS associate with CVD outcomes independent of traditional risk factors among individuals with T2D, the addition of PRS to traditional, validated models does not specifically improve the predictive performance as compared to the base model.RECODe demonstrated modest discrimination potential at baseline (AUC = 0.66). As such, the lack of improved risk prediction may reflect the performance of the RECODe equation in our cohort as opposed to lack of PRS utility.Current performance of clinical risk models appears modest. Although PRS doesn’t meaningfully improve performance, there is still substantial opportunity to improve risk prediction.