The PREVENT model showed fairness across most demographic and SDOH subgroups in predicting 10-year atherosclerotic cardiovascular disease risk, and adding SDOH predictors offered minimal incremental benefit, reinforcing the original equations' utility as a reliable and fair tool for general populations.
The 10-year atherosclerotic cardiovascular disease event rate in the study cohort was 1.8%.
The most pronounced calibration disparities were observed between White and Asian participants.
Pronounced disparities in model fairness were observed between private and public insurance groups.
Most subgroups exhibited consistent calibration and Cross Concordance Index values when evaluated with the PREVENT model.
Adding social determinants of health predictors to the PREVENT model had minimal effects on model performance.
This research examined how well a newer heart disease risk prediction tool called PREVENT (Predicting Risk of Cardiovascular Disease Events) works fairly across different groups of people. Using health records from over 554,000 adults across multiple U.S. health systems, the researchers checked whether the model was equally accurate for people of different races, sexes, ages, and social circumstances. Overall, the model performed consistently and fairly across most groups studied. However, two notable exceptions were found. The model showed meaningful differences in accuracy between White and Asian patients, and between people with private versus public insurance. For example, at higher predicted risk levels, Asian patients had lower actual event rates than White patients, suggesting the model may slightly overestimate risk for Asian individuals in that range. Similarly, the model's ranking accuracy differed substantially between insurance groups, which may reflect underlying differences in healthcare access or disease burden captured imperfectly by the model. This research also suggests that adding information about social factors — such as housing, income, or education — did not meaningfully improve the model's predictions. This finding indicates that the current PREVENT model, which was designed to be race-free, already performs well enough for broad clinical use without needing additional social data inputs. The results may help clinicians feel more confident using this tool across diverse patient populations, while also highlighting specific subgroups where further refinement could be beneficial.
Wang H, Tian Z, Bhattacharya R, Niu M, Sang Y, Wojdyla D, et al.. (2026). Toward a Deeper Understanding of Predicting Risk of Cardiovascular Disease Events for 10-Year Atherosclerotic Cardiovascular Risk: Subgroup Fairness and Predictive Value of Social Determinants of Health.. Journal of the American Heart Association. https://doi.org/10.1161/JAHA.125.046186