Project profile: Debiasing Clinical Care Algorithms: Evaluating PREVENT Equation Across Demographic and Socioeconomic Groups
Status: Active
This project evaluates the American Heart Association’s PREVENT cardiovascular risk equations using large-scale EHR data from Duke Health and national sources. PREVENT demonstrated strong overall performance, with robust discrimination and generalizability across diverse populations and real-world data settings. While performance disparities in performance were observed in some subgroups, findings support PREVENT’s utility as a reliable tool for guiding cardiovascular risk prediction. Ongoing work explores recalibration strategies to enhance fairness and ensure equitable application.
Principal Investigator: Chuan Hong
Paper 1: Evaluation of PREVENT in Duke EHR
Title: Evaluating the PREVENT Cardiovascular Risk Prediction Equations Across Demographic and SDOH Subgroups Using Electronic Health Records from Duke Health
Overview: This study assessed the 5-year performance of the AHA’s PREVENT cardiovascular risk equations using data from over 400,000 patients in the Duke Health EHR, with a particular focus on “relax cohort”, which included patients with missing values to better reflect real-world clinical settings. By evaluating model performance in this broader, more inclusive cohort, we aimed to assess PREVENT’s robustness to incomplete data, which is a common challenge in routine healthcare data. Our results demonstrate that PREVENT maintained strong discrimination across demographic and socioeconomic subgroups, confirming its reliability in ranking risk despite data missingness.
Paper 2: National Evaluation Using Truveta EHR Data
Title: Toward a Deeper Understanding of PREVENT for 10‐Year Atherosclerotic Cardiovascular Risk: Fairness, Local Adaptation, Social Context and Cost Effectiveness
Overview: Using a large, multi-system dataset of over 550,000 adults from Truveta, this study evaluated 10-year ASCVD risk prediction across fairness, local adaptation, SDOH integration, and cost-effectiveness. PREVENT consistently outperformed the Pooled Cohort Equations (PCE) in fairness metrics. Local model adaptation and adding SDOH variables yielded modest performance gains. Cost-effectiveness analysis showed PREVENT provided higher net utility than PCE across varied scenarios.
Paper 3: Cost-Effectiveness Modeling of PREVENT
Title: From Prediction to Prevention: Cost-Effectiveness of PREVENT Equation for Cardiovascular Decision Making
Overview: This paper presents a detailed economic evaluation of PREVENT-guided prevention strategies. Predictive models are increasingly used to guide preventive interventions in healthcare, yet the cost-effectiveness of deploying these models remains underexplored. Model performance alone is insufficient to justify deployment in clinical settings. Cost-effectiveness depends on the interplay between predictive accuracy, economic parameters, and intervention efficacy. The framework provides practical guidance on aligning model performance with operational constraints. Interventions guided by predictive models can generate net benefit when precision is high and cost-benefit conditions are favorable. This approach supports evidence-based decisions regarding the use of AI-driven risk tools in preventive care.
