Talk to the Veterans Crisis Line now
U.S. flag
An official website of the United States government

Health Services Research & Development

Go to the ORD website
Go to the QUERI website
HSRD Conference Logo

2023 HSR&D/QUERI National Conference Abstract

Printable View

1037 — Comparison of multi-system data to estimate disease burden for VA-Medicaid dual-enrollees

Lead/Presenter: Patrick O'Mahen,  COIN - Houston
All Authors: O'Mahen PN (Center for Innovation in Quality, Safety and Effectiveness, Houston;), Eck CS IQuESt, Houston Rajan SS University of Texas, Houston; School of Public Health Petersen LA, IQuESt, Houston

Using information from only VA sources to calculate risk scores for VA-Medicaid dual enrollees may understate disease burdens for Veterans who use non-VA medical services. This approach may lead to inadequate risk adjustment in research studies. We seek to quantify differences in scores using VA-only and VA + Medicaid data.

Using enrollment data from the VA’s Corporate Data Warehouse (CDW), we identified the population of VA-enrolled Veterans ages 18-64 years in Priority groups 1-5 during 2011-2016. We matched VA-enrolled patients to the VA Information Resource Center (VIReC) Medicaid Analytic Extract (MAX) and T-MISIS Analytic Files (TAF) data sets using scrambled Social Security numbers Our unit of analysis was person-year. Any Veteran dual-enrolled in the VA and Medicaid for at least one month in a year during the study period was counted as dual-enrolled. We calculated three measures of risk-adjustment: the Centers for Medicaid and Medicare Services 2021 Hierarchical Conditional Categories scores (CMS V21), Charlson and Elixhauser. For each observation, we calculated these three risk scores using different data sources: VA data (VA-only), Medicaid data (Medicaid-only), and both VA and Medicaid data (Combined). To compare scores, we calculated Intraclass Correlations.

There were 686,644 unique dual-enrollees, accounting for 1,832,943 person-years from 2011-2016. VA-only data contained an average of 12.1 diagnoses per Veteran, compared to 6.7 with Medicaid-only data. Combined data had an average of 18.3 diagnoses, with an overlap of 0.5 diagnoses. Across all risk-adjustment measures the VA-only and Medicaid-only scores underestimated disease burden compared with combined scores. For the CMS V21 VA-only data yielded an average score of 0.66, while Medicaid-only data averaged 0.55 and combined scores averaged 0.90. For Charlson, the respective average scores were 0.56, 0.41 and 0.91, while for Elixhauser they were 0.24, 1.30 and 1.46. The Intraclass Correlations indicated low agreement between VA-only and Medicaid-only scores with 0.19 for CMS- V21, 0.20 for Charlson and 0.19 for Elixhauser. Comparing VA-only and Medicaid-only correspondence with combined scores led to mixed results. VA-only scores correlated more closely with both combined scores for V21 and Elixhauser than Medicaid-only (0.71 to 0.59 and 0.70 to 0.63, respectively), while Medicaid-only correlated more strongly in Charlson (0.71 to 0.68). Sensitivity analysis classifying those enrolled in Medicaid for a full year as dual enrollees did not change results.

Analysis shows low agreement between conditions recorded using VA-only and Medicaid-only data. This finding is robust across three methods of risk adjustment. VA data tends to be more closely correlated with combined VA+Medicaid scores than Medicaid-only data, but neither is a reliable substitute for combined data.

Researchers using VA data should be cautious about assuming internal VA risk-adjustment methods adequately measure risk if their cohort contains Medicaid-VA dual-enrollees. Doing so may systematically underestimate disease burden for VA enrollees. Future research needs to determine characteristics of individuals who are likely to utilize both systems, especially given the increase in dual-enrollment after the Affordable Care Act’s Medicaid expansion.