Session number: 1056
Abstract title: Use of Surname Matching to Identify Elderly Hispanic Males in Combined Medicare/VA Data Files
Author(s):
RO Morgan - Houston Center for Quality of Care and Utilization Studies
IW Huang - Houston Center for Quality of Care and Utilization Studies
Objectives: Identification of race/ethnicity of veterans in the national VA utilization files is problematic. Within VA datasets, race/ethnic identifiers may be missing for 40% or more of veterans. Although the race/ethnic codes in the Medicare administrative data sets are more complete, individuals of Hispanic origin are still largely under-identified, greatly limiting the use of these administrative data for examining race/ethnic differences. These analyses assessed the utility of a Hispanic surname match for improving the accuracy of race/ethnic codes in combined VA/Medicare data sets.
Methods: We used the U.S. Census Bureau’s Hispanic Surname list combined with the 1998 Medicare Name and Address files and the 1993-1998 VA utilization files for five Florida counties. We compared Medicare race codes and the surname-enhanced codes to self-reported race/ethnic identification obtained from a sample of individuals drawn from the combined VA/Medicare files. Further, using the entire five-county VA/Medicare dataset, we examined the percentages of elderly VA-using and non VA-using males who were identified as ‘Hispanic’ using the Medicare codes versus the surname-enhanced codes. Analyses were limited to males, aged 65 and over.
Results: Forty-five percent of self-identified Hispanic survey respondents (N=116) were correctly identified by the original Medicare codes. In contrast, the surname match alone correctly identified 78%, and 89% were correctly identified by the surname-enhanced Medicare code. The percentage of ‘false positives’ using the Medicare codes alone was 28%, compared to 23% using the surname enhanced codes. For the combined five-county VA/Medicare data, the surname enhanced race codes increased the percentage of VA-users identified as Hispanic from 1.3% to 3.3% (282 versus 618), and the percentage of VA non-users from 9.2% to 18.5% (18,622 versus 31,625). Individuals identified by the surname enhanced codes were significantly younger than those identified by the original Medicare codes alone (p < .0001).
Conclusions: The surname-enhanced race/ethnic codes approximately doubled the numbers of individuals identified as ‘Hispanic’ and dramatically increased the percentage of individuals who were correctly identified. Further, Hispanic samples based on the Medicare codes alone appear to be age-biased.
Impact statement: Use of the Census Bureau’s Hispanic Surname file appears to substantially improve our ability to correctly identify Hispanic individuals from administrative data.