Abstract Body (Do not enter title and authors here): Background:
The Multi-Ethnic Study of Atherosclerosis (MESA) Risk Score is used to predict 10-year coronary heart disease (CHD) risk based on traditional risk factors and coronary artery calcium (CAC). Discussions around the potential for equity consequences when using race/ethnicity as a predictor in models for clinical decision-making have prompted other risk scores, such as the PREVENT score, to no longer include race/ethnicity. To this end, we developed a revised score omitting race/ethnicity and compared its performance to the original MESA score.
Methods:
MESA is a longitudinal multi-ethnic observational cohort study of 6,814 participants, who were aged 45 to 84 and did not have prevalent cardiovascular disease upon study entry in 2000-2002. Model development was done using regularized Cox regression. The regularization process had two stages. The first used LASSO regression for variable selection, where we considered age, CAC, total and HDL cholesterol, systolic blood pressure (SBP), sex, diabetes, family history of heart disease, hypertensive medication, current smoking, and lipid-lowering medication use. Interactions were considered between the variables age, sex, and CAC with all other predictors; anti-hypertensive medications with SBP; and lipid-lowering medications with total cholesterol. The second stage involved ridge regression to mitigate over-fitting. We then compared the calibration and discrimination of the model including race/ethnicity to that of the model that did not.
Results:
The updated model excluding race/ethnicity included the same main effects, but several interactions were also selected in the LASSO step. When comparing the performance of the models including race/ethnicity to those excluding it, no significant differences were found. For example, there was no significant difference in AUC between the model excluding race/ethnicity and that including it, with a difference of 0.00323 (95% bootstrap CI: -0.000715 to 0.00717). Harrell’s C-statistic was comparable between the models, at 0.800 in the model without race/ethnicity and 0.797 in that including it. Other metrics revealed a similar picture, where calibration and discrimination did not differ the inclusion of race/ethnicity.
Conclusions:
We developed a MESA risk score model that does not use race/ethnicity as a risk factor that performs similarly to the risk score model including race/ethnicity. This risk score will be made publicly available for clinical CHD risk prediction.