Targeting Residual ASCVD Risk by Integrating Genetics and Clinical Data

Patients regularly have myocardial infarctions or strokes, despite being on treatment to lower low-density lipoprotein cholesterol (LDL-C) and blood pressure (BP). This residual risk of atherosclerotic cardiovascular disease (ASCVD) remains a substantial medical burden globally. To understand and control residual ASCVD risk, we need to answer three questions: (1) what are the underlying mechanisms? (2) who is at higher risk? (3) are there new treatment options? Building on our experience, we propose to fill these knowledge gaps: (1) The genetics of residual ASCVD risk remain largely unexplored. Existing genetic studies focus primarily on baseline risk (e.g., risk of CHD) or response to CHD-relevant drugs (e.g., LDL-C change on statins), not the residual ASCVD risk. Our group has previously identified and replicated LPA variants associated with residual CHD events in patients on statin treatment independent of LDL-C change. In Aim 1, we will identify new users of drugs to lower LDL-C or BP in the biobank at Vanderbilt (BioVU) and conduct a GWAS comparing those who have an ASCVD event while on treatment to those without an event. We will conduct replication in eMERGE and All of US (AoU) and construct polygenic risk scores. (2) There is a dearth of generalizable and accurate models to predict residual ASCVD risk. Electronic health records (EHRs) contain granular, longitudinal, real-world data that are inherently medically relevant. Additionally, large programs, such as AoU, collect information on lifestyle (e.g., smoking) and social determinants of health (e.g., insurance and employment status). Recently, we applied machine learning (ML) algorithms for CHD risk prediction on longitudinal EHR data and found that ML outperformed the ACC/AHA risk equation for prediction of ASCVD. The performance was further improved by including additional factors. In Aim 2, we will leverage longitudinal EHRs at Vanderbilt and apply advanced ML method to construct prediction models for residual ASCVD risk. We will then validate the model in AoU and eMERGE, and improve the performance by including genetic, environmental, and social information. (3) There are few drug therapies to reduce ASCVD risk other than ones to lower LDL-C and BP. Recent advances in genetics, informatics, and large real-world clinical datasets provide opportunities for systematic drug repurposing. Our group recently pioneered a new drug-repurposing approach. We first leveraged existing GWAS to impute genetically determined transcriptome signatures (the virtual transcriptome) associated with elevated LDL-C or raised BP. We then screened the transcriptome signature in a drug perturbation database to identify candidate drugs, followed by validating them in real-world clinical data. In Aim 3, we will conduct drug repurposing for residual ASCVD risk. We will use two approaches: (a) impute the genetically determined transcriptome and (b) large-scale Mendelian randomization. We will validate the identified candidate drugs in Medicare and AoU. These studies will have potential high impact for residual ASCVD risk by identifying: (1) underlying mechanisms; (2) patients at higher risk; and (3) novel drugs to reduce residual risk.