Abstract P416: An Externally Validated and Highly Sensitive Artificial Intelligence-driven Integrated Genetic-epigenetic DNA Test for Incident Coronary Heart Disease Prediction

Abstract

Despite being largely preventable, every year, coronary heart disease (CHD) affects millions of Americans. Currently, primary prevention of CHD starts with an understanding of several risk factors (e.g. age, gender, diabetes) that are aggregated to derive an incident CHD risk score using risk calculators such as the Framingham Risk Score (FRS) and ASCVD Pooled Cohort Equation (PCE). Unfortunately, this approach involves multiple tests and its lack in sensitivity could result in ineffective treatment recommendations to patients for the prevention of CHD. As an alternative, we have developed and independently validated a single, simpler, AI-driven, digital PCR (dPCR)-based integrated genetic-epigenetic DNA test that can more sensitively estimate the 3-year incident CHD risk for both men and women. Our technology accounts for the inherited and acquired (lifestyle/environment) risks for CHD through three genetic (SNP) and three epigenetic (DNA methylation) biomarkers, respectively, and can be performed on DNA from blood or saliva. The incident CHD risk prediction model was developed using genome-wide DNA methylation and genotype data from the Framingham Heart Study (FHS) Offspring cohort (n=1172 in training set, n=512 in test set) and was validated in an Intermountain Healthcare (IM) cohort (n=80 in validation set, n=79 in test set). The final prediction model is an ensemble of SVM, Random Forest and Logistic Regression models, and its performance in the FHS and IM test sets is summarized in Table 1. The FRS and PCE risk calculators were implemented on all FHS and IM cohort data. A clinically implementable version of this tool as part of its translation into a Laboratory Developed Test incorporates standard Taqman assays for genotyping and custom dPCR assays for DNA methylation quantification. The strong correlation between our dPCR assay values and the Illumina array values are shown in Figure 1, indicating successful dPCR assay translation. We are also extending this tool to include more ethnically diverse cohorts.