Geographic Variation in Missing Race and Ethnicity Data in Minimum Data Set 3.0

BACKGROUND: Race and ethnicity measures in administrative data can vary geographically. The extent of this challenge in US nursing homes is not well described. OBJECTIVES: To describe geographic variation in missing race and ethnicity data in the Minimum Data Set (MDS) 3.0 and Medicare claims, and to compare discrepancies across data sources. RESEARCH DESIGN: Cross-sectional study. SUBJECTS: Medicare beneficiaries with MDS 3.0 records between 2014 and 2018. The Medicare Beneficiary Summary File provided demographic information. MEASURES: Missingness of MDS race and ethnicity data by state, and misclassification of Medicare race and ethnicity enrollment database (EDB) and Research Triangle Institute (RTI) variables compared with MDS. We calculate the sensitivity, specificity, and positive predictive value of the EDB and RTI variables relative to the MDS. RESULTS: Among 18.1 million nursing home residents pooled across 2014-2018, geographic variation in missing race and ethnicity in the MDS 3.0 ranged from 1.2% to 14.7%. Compared with MDS, misclassification of residents classified as Hispanic in MDS ranged from 48.1% to 89.2% for EDB and 0.5% to 44.8% for RTI. Misclassification of residents classified as Asian American/Pacific Islander in MDS ranged from 29.4% to 77.2% for EDB and 12.7% to 65.4% for RTI. Misclassification of residents classified as Black ranged from 0% to 14.2% for EDB and 0% to 16.2% for RTI. Overall, the RTI variables provided better sensitivity and specificity of race and ethnicity than the EDB. CONCLUSION: Missing race and ethnicity data in the MDS varies geographically, as do discrepancies between MDS and EDB and RTI variables. Thoughtful consideration of these issues is recommended when handling missing MDS race and ethnicity data.