Does concordance between survey responses and administrative records differ by ethnicity for prescription medication?

BACKGROUND Self-reported prescription medication use data is often used to measure differences across ethnic groups, but its accuracy may differ across ethnic groups. OBJECTIVE We compared ethnic groups' self-reported medication use to their administrative records for respondents with diabetes, hypertension, and asthma. METHODS We linked the Canadian Community Health Survey to administrative prescription drug records for 17,191 respondents in British Columbia, Canada. We evaluated the concordance between self-reported medication use and prescription drug records using positive predictive value, negative predictive value, sensitivity, specificity, and kappa statistic for self-identified Whites, Chinese, South Asians, and Southeast Asians/Filipinos. The concordance was calculated using prescription drug records as the reference standard. We also estimated the odds of disagreement (either a false positive or negative) in medication use with logistic regressions for each ethnic group, and compared them using the Blinder-Oaxaca method. RESULTS We found that Chinese had the worst positive predictive value for asthma medication use at 0.41, while South Asians had the worst sensitivity for hypertension medication use at 0.60. The difference in reporting an error between ethnic groups was likely explained by differences in respondent characteristics. Particularly, if White respondents had the same characteristics as South Asians, then White respondents would have had 1.031 (95% CI: 1.020-1.041) higher odds of disagreement for hypertension medication use than with their own characteristics. CONCLUSION Self-reported medication use may be a valid measure of ethnic groups' medication use if ethnic differences in characteristics, like household income are held constant. However, an important determinant of validity for all ethnic groups is whether medications are used routinely, or for a specific episode.