Personalizing Cardiovascular Disease Prevention

by Amanda Chase, PhD
March 7, 2022

Cardiovascular disease is currently considered to be a leading cause of death in the United States. A better understanding of cardiovascular disease (CVD) and finding better ways to treat CVD have been topics of research for many years. Despite an early description of fats accumulating on artery walls, it took much longer for there to be a link between fat deposition as a major reason for CVD progression, and longer still for the creation of drugs that could lower cholesterol and reduce the risk of heart attacks from fat accumulation. In general, the drugs developed to lower cholesterol are referred to as “statins”. Statins have since become widely used for primary prevention of cardiovascular disease by lowering cholesterol, and are, in fact, considered to be the cornerstone of CVD prevention. It has been shown that using statins significantly reduces the risk of heart disease, stroke, and death for those with a high-risk of developing CVD.

Despite the advantages of statins, adherence to doctor prescribed statins is at only 50% after one year, and down to only 30% after two years. The question of why adherence is so low is important to inform patient and doctor decision making when determining best treatment options. Using historical, real-world data may begin to explain prior treatment responses and inform personalized decision making that could improve adherence. Co-first authors Ashish Sarraju, Instructor at Stanford, and Andrew Ward, along with senior author Fatima Rodriguez, recently developed an electronic health record (EHR) based machine learning (ML) approach to creating personalized statin recommendations for preventing CVD. Their approach was recently published in Scientific Reports. Using ML allowed the team to identify intermediate risk patients who were recommended a moderate- or low-intensity statin, instead of a high-intensity statin, based on outcomes in similar patients. This information may enhance shared decision-making for patients where the evidence about what statin type and dose to prescribe may be less certain.

The use of ML to help identify the most appropriate statin to help lower cholesterol for patients may serve as a way to optimize shared decision-making by allowing a more personalized discussion with an individual patient. It opens the possibility for having discussions about why starting a certain statin may make the most sense for that patient (e.g., risk factors, non-adherence, etc.). The development of this personalized statin decision-making approach for primary prevention of CVD may be an important step in identifying patients at risk for suboptimal statin responses and may provide a way to study the increasing gap between statin efficacy and effectiveness.

Other Stanford Cardiovascular Institute affiliated authors include Areli Valencia and Latha Palaniappan. Other Stanford authors include Jiang Li and David Scheinker.

Ashish Sarraju, MD

Fatima Rodriguez, MD