PREVENT Equations in Young Adults: Fairness, Calibration, and Performance Across Racial and Ethnic Groups

Area of Science:

• Cardiology
• Public Health
• Health Disparities

Background:

• Cardiovascular disease (CVD) risk is rising in young adults.
• The American Heart Association's PREVENT equations estimate CVD, ASCVD, and HF risk.
• Augmented equations incorporate a social deprivation index (SDI) for social exposures.

Purpose of the Study:

• Assess performance and algorithmic fairness of base and SDI-augmented PREVENT equations in young adults (20-39 years).
• Define fairness as similar performance across racial and ethnic groups.
• Conduct primary analysis in adults aged 30-39 and exploratory analysis in adults aged 20-29.

Main Methods:

• Included 161,202 Kaiser Permanente Southern California members aged 20-39 without prior CVD (2008-2019).
• Compared 10-year predicted vs. observed CVD, ASCVD, and HF events using base and SDI-augmented PREVENT models.
• Estimated performance (C-statistic, calibration) and fairness (concordance imparity, fair calibration) by race/ethnicity and age group.

Main Results:

• Base PREVENT CVD model showed fair discrimination (C-statistic 0.68-0.72) with low concordance imparity (0.04).
• Mean calibration revealed underprediction in non-Hispanic Black participants (0.54) compared to other groups (0.96-1.07).
• Prediction errors differed across racial/ethnic groups; SDI augmentation did not enhance performance or fairness.

Conclusions:

• PREVENT equations demonstrate real-world performance in a diverse young adult cohort.
• Model performance varies across age, race, and ethnicity.
• Clinical application of PREVENT for preventive strategies requires consideration of these performance variations.