Evaluating Discrimination of Cardiovascular Disease Risk Prediction Models Among Diverse Socioeconomic Groups: the All of Us Cohort
Abstract Body: Background The American Heart Association Predicting Risk of cardiovascular disease EVENTs (AHA PREVENT) were introduced as an updated method to guide cardiovascular disease (CVD) risk prediction. These equations included new CVD outcomes (e.g., heart failure) and performed well among diverse race groups. Given that many development and validation studies often include individuals of higher socioeconomic backgrounds, the generalizability of PREVENT provides further opportunities to evaluate its predictive ability in other diverse socioeconomic groups. Objective To evaluate the generalizability of PREVENT on an ongoing diverse U.S. cohort by income and educational attainment. Methods We analyzed PREVENT-eligible participants from the All of Us (AoU) Study, an ongoing longitudinal cohort study of individuals living in the U.S. (n = 4,924). We calculated the PREVENT risk scores for 5-years of follow-up for total CVD and atherosclerotic CVD (ASCVD). We then examined the discriminatory ability of PREVENT for 5 years of follow up using Harrell’s C-Statistic for the full sample and by race, educational attainment, and annual income categories. Results PREVENT Total CVD (C-statistic = 0.765), ASCVD (C-statistic = 0.736), and Heart Failure (C-statistic = 0.811) performed adequately in the examined AoU cohort (Table). For income, Total CVD performance was best among people who made $50k-100k (C-statistic = 0.874) and worse among people who made < $50k (C-statistic = 0.706). However, people who made > $100k had the largest uncertainty in equation performance (C-statistic = 0.713, 95% CI: 0.231-0.909). For education, total CVD performance was best among people with a college degree or higher (C-statistic = 0.821) and worse among people with a high school diploma or lower (C-statistic = 0.709). Conclusion The AHA PREVENT equations showed continued robustness when applied to a diverse, ongoing national cohort and when stratified by educational attainment and annual income. While we show that PREVENT has good generalizability, the inconsistency for certain income and educational attainment groups highlights the need to recruit more socioeconomically diverse populations for cohort studies.
Lewis, Ashley
( STANFORD UNIVERSITY
, Stanford
, California
, United States
)
Bacong, Adrian
( Stanford University
, Mountain View
, California
, United States
)
Palaniappan, Latha
( STANFORD UNIVERSITY
, Stanford
, California
, United States
)
Hernandez-boussard, Tina
( STANFORD UNIVERSITY
, Stanford
, California
, United States
)
Author Disclosures:
Ashley Lewis:No Answer
| Adrian Bacong:DO NOT have relevant financial relationships
| Latha Palaniappan:DO NOT have relevant financial relationships
| Tina Hernandez-Boussard:No Answer
