OBJECTIVES: To contrast race and ethnicity data reported in Oracle EHR Real-World Data (OERWD) with a US self-reported survey.

METHODS: Data from OERWD, a de-identified, HIPAA-compliant dataset from 139 US health systems (<2023, n=105,832,841) was linked to data from the US National Health and Wellness Survey (NHWS) (2015-2022, n=292,391), a nationally-representative, self-reported, cross-sectional online survey. Individuals with invalid or inconsistent data for birth, encounter dates, race, or ethnicity within each dataset were removed. Race was categorized as American Indian/Alaskan Native, Asian, Black/African American, Native Hawaiian/Other Pacific Islander, White, Other, and Unknown. A Mixed category was added if multiple races were selected in NHWS. Ethnicity was reported as Non-Hispanic, Hispanic/Latino, and Unknown. Descriptive statistics were derived.

RESULTS: We identified 29,992 individuals, 64.7% were women with an overall mean age at last encounter of 46.2 years (SD=19.6). For race, 72.3% (n=21,683) of patients had concordant responses in both datasets, 9.5% (n=2,852) were discordant, and 18.2% (n=5,457) were Unknown in OERWD. The highest proportions of matched responses were observed among patients identifying as White (78.6%), Black/African American (74.0%) and Asian (51.1%). For all races, except Asian and White, when answers were discordant, most individuals were considered White in OERWD. Regarding ethnicity, 58.0% (n=17,396) of patients had concordant answers, 8.2% (n=2,470) were discordant, and 33.8% (n=10,126) were Unknown in OERWD. The proportion of matching answers increased with age at last encounter from 46.4% (0-14) to 86.0% (75+) for race and 10.3% to 81.5% for ethnicity.

CONCLUSIONS: Our study showed adequate concordance between EHR-recorded race and ethnicity and self-reported data indicating reliability of these variables when reported in EHR. Further research is needed to better understand reasons for underreporting.