3104 — VA Race and Ethnicity Data Consistency and Agreement with Survey Self-Report
Washington DL (VA Greater Los Angeles Healthcare), Lanto A
(VA Greater Los Angeles Healthcare), Mitchell MN
(VA Greater Los Angeles Healthcare), Riopelle D
(VA Greater Los Angeles Healthcare), Sun S
(VA Greater Los Angeles Healthcare)
Since 2003, VA has collected self-reported race and ethnicity data in compliance with a new Office of Management and Budget guideline. Resultant missing race/ethnicity values, and strategies for incorporating Medicare data to improve data completeness, have been examined. However, consistency across time of new VA race/ethnicity measures, and their agreement with survey self-report is unknown.
We linked 2006, 2007, and 2008: Survey of Healthcare Experiences of Patients (SHEP) surveys; VA Outpatient Visit Medical SAS datasets (VA-data); and External Peer Review Program (EPRP) data. SHEP provided self-reported race and ethnicity. VA-data were coded to incorporate race and ethnicity, and method of data collection. SHEP and VA-data both allowed multiple race reporting. For individuals with EPRP data and non-missing self-reported VA-data (n = 413,033), we calculated data consistency as the percentage of visits in which each race/ethnicity category (Hispanic, White, African-American, Asian, Native Hawaiian/Other Pacific Islander [NHOPI], American Indian/Alaskan Native [AIAN]) was coded (e.g., percent visits an individual was Hispanic). For the subgroup with 100% VA-data consistency and one SHEP survey (n = 87,138), we calculated percent agreement between VA-data and SHEP. Independently, for individuals with two SHEP surveys (n = 16,804), we calculated percent agreement between the two survey self-reports.
Individuals in the linked multi-year VA data had 56.6 mean visits (standard deviation 59.7). Race/ethnicity categories were 100% consistent across visits for 97.8% of individuals. This group was 78.8% White, 17.3% African-American, 5.1% Hispanic, 0.5% Asian, 0.7% AIAN, 0.9% NHOPI. Broadening race assignment based on the majority of an individual’s visits increased the number of NHOPIs 19%, AIANs 16%, Asians 5%, Hispanics 4%, African-Americans 2%, and Whites 1%. Concordance between two SHEP surveys was 95% for Whites, 97% for AIANs, and at least 99% for other groups. VA-data/SHEP concordance was similar.
Agreement of self-reported survey and VA-data is comparable to concordance in repeat surveys. Though VA race/ethnicity was highly consistent, race misclassification disproportionately affected smaller groups. Limiting race assignment to 100% consistent records, reduces the available sample from these groups.
Trade-offs exist between assigning race category only when 100% consistent, versus, based on the majority of records. This choice should be guided by the research or policy question.