Search | Search by Center | Search by Source | Keywords in Title
Tarlov E, Perrin RA, Zhang Q, Koelling K, Ferreira MR, Hynes DM, Bennett CL. Consequences Of Ignoring VA Data in Studies Of Medicare-Enrolled Elderly with Cancer. Poster session presented at: AcademyHealth Annual Research Meeting; 2007 Jun 1; Orlando, FL.
Research Objective: Health services researchers often rely on Medicare claims data to identify disease-specific cohorts and for information on health care use and costs among the elderly. Veterans, who comprise 25 percent of the U.S. population over age 65, may also receive care within the VA system. Approximately 1.5 million elderly veterans receive VA healthcare. The objective of this study was to understand the impact on research results when VA data are omitted in a database study of Medicare-enrolled elderly with colon cancer. Methods: As part of a national study, a retrospective cohort of incident colon cancer patients who were 66 years old or older and eligible to use both VA and Medicare between 1999 and 2001 was identified from California Cancer Registry data. We examined characteristics of the sub-group who used the VA system exclusively for their healthcare. For the group who used Medicare only or a combination of Medicare and VA healthcare, we compared Charlson scores computed using a combination of Medicare and VA data with scores obtained without the benefit of VA data. Results: The California cohort comprised 633 veterans with Stage I-IV colon cancer. Ninety-seven (15%) of the cohort were users of VA healthcare only and had no Medicare claims data. Compared with the full cohort, the 97 VA-only patients were more likely to be male (98% vs. 93%), African American (26% vs. 16%), younger (47% vs 59% over age 75), and diagnosed at Stage IV of their cancer (28% vs. 18%). This group also had a lower mean Charlson score (0.76, SD 0.99 vs. 0.93, SD 1.11, not including their cancer diagnosis). Among the remaining 536 patients, many received care through both VA and Medicare. Linking VA data with a registry and Medicare linked dataset, we uncovered 3 more colectomies (for a total of 474) and evidence of chemotherapy for 4 more patients (total 176). Charlson scores computed for the 536 using Medicare data only were lower (mean 0.83, SD 1.05 and 20.9% with a score of 2 or higher) than those computed after VA data were added to the dataset (mean 0.96, SD 1.13, and 24.8% with a score of 2 or higher). Conclusions: The 97 patients who used VA healthcare exclusively differed substantially from the full cohort of veterans. Incomplete data on patients who used some VA care resulted in fewer numbers of cases who had had colectomies and received chemotherapy. Charlson scores computed without VA data underestimated true comorbidity. Implications: Researchers’ reliance on Medicare data alone in studies of elderly cohorts will systematically exclude the 18% of elderly veterans who use VA healthcare exclusively, a group that differs in important ways from the general population. Further, research study designs may need to account for missing data on elderly veterans who receive care in both the VA and Medicare systems.