Haywood TW (VA Information Resource Center (VIReC)), Stroupe KT
(Center for Management of Complex Chronic Care), Tarlov E
(VA Information Resource Center (VIReC)), Owens A
(VA Information Resource Center (VIReC)), Weichle T
(Center for Management of Complex Chronic Care), Hynes DM
(VA Information Resource Center (VIReC))
Research into the quality of race/ethnicity data in Medicare records provides important information to researchers who plan to supplement missing VA race and ethnicity data with information obtained from Medicare. We investigated the agreement between VA and Medicare race and ethnicity data among VA patients age 65 years and older.
As part of a larger study, we selected a 10% random sample of patients who used VA acute inpatient or outpatient care during fiscal years 2004 or 2005. Using a deterministic matching procedure, we linked VA and Medicare records for those patients who had race or ethnicity information in VA workload data and were age 65 or older on 11/1/2005, resulting in a sample of 169,018 patients. We compared Medicare race/ethnicity values to VA self-reported race and ethnicity values (considered the gold standard for the purposes of this study).
Among patients self-identified as White or Black in VA data, 97% and 95%, respectively, had concordant information in Medicare records (Kappa 0.84 and 0.93). Of the 9,074 patients identified as Hispanic, Asian, or American Indian in VA data, 25%, 53%, and 24%, respectively, were coded the same in Medicare. Thus, compared to VA data for these minority groups, Medicare data had low sensitivity and low overall rates of agreement (Kappa 0.37, 0.59, 0.30, respectively). VA minorities who were misclassified in Medicare were most often classified as White, except for Asians who were most often classified as “Other.”
Supplementing missing race and ethnicity information in VA records with Medicare data will result in high rates of accurate classification of patients who are either White or Black race. For the small number of non-Black minority VA users, use of Medicare data to supplement VA race data will result in misclassification in a high percentage of cases.
This study provides valuable information for VA research using secondary data since race and ethnicity information are often missing in VA databases and patient race is a critically important variable in most health research. It demonstrates that for elderly veterans Medicare data can be used to supplement VA data and that doing so will result in accurate classification among most White and Black patients. Studies focused on non-Black minorities will need additional means to identify patient race or ethnicity, such as survey data.