Methods of estimating race/ethnicity using administrative data are increasingly used to examine and target disparities; however, there has been no validation of these methods using clinically relevant outcomes.Objective:
To evaluate the validity of the indirect method of race/ethnicity identification based on place of residence and surname for assessing clinically relevant outcomes.Data Sources:
A total of 2387 participants in the Post-MI Free Rx Event and Economic Evaluation (MI FREEE) trial who had both self-reported and Bayesian Improved Surname Geocoding method (BISG)-estimated race/ethnicity information available.Study Design:
We used tests of interaction to compare differences in the effect of providing full drug coverage for post-MI medications on adherence and rates of major vascular events or revascularization for white and nonwhite patients based upon self-reported and indirect racial/ethnic assignment.Results:
The impact of full coverage on clinical events differed substantially when based upon self-identified race (HR=0.97 for whites, HR=0.65 for nonwhites; interaction P-value=0.05); however, it did not differ among race/ethnicity groups classified using indirect methods (HR=0.87 for white and nonwhites; interaction P-value=0.83). The impact on adherence was the same for self-reported and BISG-estimated race/ethnicity for 2 of the 3 medication classes studied.Conclusions:
Quantitatively and qualitatively different results were obtained when indirectly estimated race/ethnicity was used, suggesting that these techniques may not accurately describe aspects of race/ethnicity related to actual health behaviors.