PREVENT Equations in Young Adults

Cardiovascular disease (CVD) is increasing among young adults. The American Heart Association’s PREVENT (Predicting Risk of Cardiovascular Disease Events) equations estimate risk of CVD, atherosclerotic cardiovascular disease (ASCVD), and heart failure (HF) for primary prevention. Augmented equations additionally include zip code-based social deprivation index (SDI) to address adverse social exposures. We assessed performance and algorithmic fairness of base and SDI-augmented PREVENT equations in young adults aged 30 to 39 years, defining fairness as similar performance across racial and ethnic groups. An exploratory analysis was conducted among young adults aged 20 to 29 years. We included Kaiser Permanente Southern California members aged 20 to 39 years without prior CVD between 2008 and 2009, followed through 2019. We compared 10-year predicted and observed CVD, ASCVD, and HF events for base and SDI-augmented PREVENT models. Performance (Harell’s C, calibration slopes, mean calibration) and fairness (concordance imparity, fair calibration) were estimated by race and ethnicity and age group (30-39 years [primary analysis], 20-29 years [exploratory analysis]). Among 161,202 young adults aged 30 to 39 years (60.0% women; 51.7% Hispanic, 26.9% non-Hispanic White, 12.5% Asian/Pacific Islander, 8.9% non-Hispanic Black), 10-year CVD incidence was 0.7%. Race-specific Harrell’s C-statistics for the base PREVENT CVD model ranged from 0.68 to 0.72, yielding low concordance imparity (0.04; 95% CI: 0.02-0.22) which implies fair discrimination. Mean calibration showed underprediction in non-Hispanic Black participants (0.54; 95% CI: 0.48-0.65) vs other groups (range: 0.96-1.07). In fair calibration testing, prediction errors differed across racial and ethnic groups. Results were similar for ASCVD and HF. Adding SDI did not improve performance or fairness despite disparities across groups. In exploratory analyses among 80,978 individuals aged 20 to 29 years, performance and fairness results were similar. This large, diverse cohort of young adults demonstrates how the PREVENT equations may perform when applied in real-world clinical settings, reflecting the true operational environment faced by large health systems. Applications of PREVENT in clinical patient care, eg, early initiation of preventive strategies, should consider variations in model performance across age, race, and ethnicity.