These shifts in utilization after Medicaid expansion may have important implications for VA policymakers. First, more study is needed to know which types of veterans are more likely to use Medicaid instead of VA services—or use both Medicaid and VA services. Our research indicates unsurprisingly that veterans without service-connected disability ratings and eligible for VA services due to low income are more likely to use at least some Medicaid services. Further understanding of who switches will be useful for the VA both tailoring its services to those who prefer VA and for reaching out to specific types of patients who might be better served by staying within the VA system. Finally, VA clinicians and administrators can prioritize improving care coordination for those who chose to use both Medicaid and VA services.
Limitations and Future Directions
Our results should be interpreted within methodological and data limitations. With only 2 states in our sample, NY demonstrably skewed overall results, contributing 1.7 to 3 times more observations than AZ across subanalyses—a challenge also cited by Sommers and colleagues.19 Our veteran groupings were also unable to distinguish those veterans classified as service-connected who may also have qualified by income-eligible criteria (which would tend to understate the size of results) and those veterans who gained and then lost Medicaid coverage in a given year. Our study also faces limitations in generalizability and establishing causality. First, we included only 2 historical state Medicaid expansions, compared with the 38 states and Washington, DC, that have now expanded Medicaid to date under the ACA. Just in the 2 states from our study, we noted significant heterogeneity in the shifts associated with Medicaid expansion, which makes extrapolating specific trends difficult. Differences in underlying health care resources, legislation, and other external factors may limit the applicability of Medicaid expansion in the era of the ACA, as well as the Veterans Choice and MISSION acts. Second, while we leveraged a difference-in-difference analysis using demographically matched, neighboring comparison states, our findings are nevertheless drawn from observational data obviating causality. VA data for other sources of coverage such as private insurance are limited and not included in our study, and MAX datasets vary by quality across states, translating to potential gaps in our study cohort.28 Finally, as in any study using diagnoses, visits addressing care for depression may have been missed if other diagnoses were noted as primary (eg, VA clinicians carrying forward old diagnoses, like PTSD, on the problem list) or nondepression care visits may have been captured if a depression diagnosis was used by default.
Moving forward, our study demonstrates the potential for applying a natural experimental approach to studying dual-eligible veterans at the interface of Medicaid expansion. We focused on changes in VA reliance for the specific condition of depression and, in doing so, invite further inquiry into the impact of state mental health policy on outcomes more proximate to veterans’ outcomes. Clinical indicators, such as rates of antidepressant filling, utilization and duration of psychotherapy, and PHQ-9 scores, can similarly be investigated by natural experimental design. While current limits of administrative data and the siloing of EHRs may pose barriers to some of these avenues of research, multidisciplinary methodologies and data querying innovations such as natural language processing algorithms for clinical notes hold exciting opportunities to bridge the gap between policy and clinical efficacy.
Conclusions
This study applied a difference-in-difference analysis and found that Medicaid expansion is associated with decreases in VA reliance for both inpatient and outpatient services for depression. As additional data are generated from the Medicaid expansions of the ACA, similarly robust methods should be applied to further explore the impacts associated with such policy shifts and open the door to a better understanding of implications at the clinical level.
Acknowledgments
We acknowledge the efforts of Janine Wong, who proofread and formatted the manuscript.