Reevaluating the Use of Race/Ethnicity in the MESA Risk Score

BACKGROUND: Cardiovascular disease risk scores are widespread in research and clinical settings. The MESA (Multi-Ethnic Study of Atherosclerosis) risk score estimates the 10-year risk of coronary heart disease and was novel in its use of coronary artery calcium alongside traditional risk factors, including race/ethnicity. However, including race/ethnicity in a risk score model requires careful consideration. We developed a race-free MESA risk score and compared it with the original. METHODS: We used data from the MESA cohort, a community-based cohort of individuals aged 45 to 84 years who identified as White, Hispanic/Latino, Black, or Chinese and were free of prevalent cardiovascular disease upon study entry, to develop a race-free risk score estimating 10-year coronary heart disease risk. We used the same 2-step regularization procedure as that for the original MESA risk score: (1) least absolute shrinkage and selection operator for variable selection, and (2) ridge regression to prevent overfitting. Neither race/ethnicity nor any interaction term with race/ethnicity was included as a candidate predictor. RESULTS: When excluding race/ethnicity, several interaction terms were retained that were not retained in the original MESA risk score. The area under the receiver operating characteristic curve for the race-free risk score was 0.820 (95% CI, 0.801-0.840), and the difference in area under the receiver operating characteristic curve was not significant (estimate, 0.0032 [95% CI, -0.0008 to 0.0073]). The discrimination slope in the race-free model was 0.101, while that of the original MESA risk score was 0.0937. CONCLUSIONS: We developed a race-free MESA risk score that performs comparably to the MESA risk score in both calibration and discrimination.