OBJECTIVES: Commercial plans are not required to collect race or ethnicity from individual members and may rely on data from electronic health records (EHRs) or enrollment files. We developed and validated a method for identifying the best available race/ethnicity for commercial members by leveraging multiple data sources, with the goal of increasing the availability and accuracy of race/ethnicity data, to allow health plans to stratify performance metrics by race/ethnicity groups which informs initiatives to improve health equity.

METHODS: We evaluated sources of race/ethnicity data (enrollment, EHRs, and imputation methods) with verified self-reported data to assess the strength of agreement. Available self-reported race/ethnicity via employment data was used as gold-standard for validation. Primary validation compared member race/ethnicity from each source with the gold-standard. Secondary validation looked at the agreement between standardized composite race/ethnicity, and the gold-standard. Standardized composite race/ethnicity was created for each member based on a prioritization method of Hispanic ethnicity from any source, followed by race from EHRs, enrollment, and finally imputed values. This prioritization method considers all available data and was informed by the primary validation results.

RESULTS: Imputation methods provided race/ethnicity values for >98% of the members, while EHR and enrollment files were available for ~38% and <2% of members, respectively. After using the prioritization method, the standardized composite race/ethnicity was available for all members. The agreement between the standardized composite race/ethnicity compared to gold-standard was very high, with Asian race having the highest TPV and Hispanic/Latino ethnicity having the lowest.

CONCLUSIONS: Race/ethnicity data in EHR and enrollment have better accuracy for commercial members compared to imputation methods, however, imputations can provide race/ethnicity for larger populations. This study demonstrated the feasibility and validity of a prioritization method by identifying best available race/ethnicity across multiple data sources as it increases the availability of race/ethnicity without decreasing data accuracy for population level reporting.