3034 — Missing Race/Ethnicity Data in the VA Longitudinal Online Research Database: Implications for Health Disparities Researchers

Meghani SH (University of Pennsylvania/Philadelphia VA Medical Center ), Wiedemer NL (Philadelphia VA Medical Center), Harden P (Philadelphia VA Medical Center), Pulman-Mooar S (Philadelphia VA Medical Center), Garvin J (Philadelphia VA Medical Center), Gallagher RM (Philadelphia VA Medical Center)

Objectives:
This study examines the magnitude of missing race/ethnicity data in a sub-sample derived from the VA Longitudinal Online Research (VALOR) Database. VALOR contains 200,000 patient records drawn from the VA electronic medical records (VISTA) containing veterans’ clinical and administrative data. The database is maintained by the Center for Health Equity Research Promotion to facilitate access to data essential for conducting VA-based clinical and health services research. Using a sub-sample of patients extracted using VALOR database, we sought to understand 1) the magnitude of missing race data, and 2) if these data are missing at random, i.e. if there is an association between demographic and service-related variables and missing race data in this sample.

Methods:
The data were from our ongoing study of chronic pain patients (n=819) and the impact of a structured opioid renewal program on pain and treatment outcomes. Using data derived from the VALOR database, we first ran descriptive statistics to identify the percentage of missing or unknown race entries for all 819 veterans (aim 1). To address aim 2, we coded the race variable with missing entries as a binary variable, (1= missing, 0= not missing). A 2-tailed test for continuous (t-test) and categorical (Pearson Chi-Squared) variables was employed to identify the relationship between missing race and demographic, disease, and service-related variables.

Results:
Of the total sample (n=819), race was missing (n=243, 29.7%) or unknown (n=57, 7%) for over one-third of the veterans. Univariate analysis showed that the race variable was missing more often for those without service connection (20.3% v. 9.4%; p < .000) and those with lower comorbidity (1.44 v. 2.42; p < .000) measured by Charlson index.

Implications:
We conclude that race data in the VA administrative database is not missing at random. It appears that veterans with greater number of medical encounters are more likely to have race data recorded.

Impacts:
Racial disparities remain pervasive in the U.S. healthcare system. VA researchers interested in health disparities research using the VA clinical and administrative datasets must be aware of the systematic patterns in the missing race data potentially impacting the validity of findings.