Methods
To investigate the relative changes in veterans’ reliance on the VA for depression care after the 2001 NY and AZ Medicaid expansions We used a retrospective, difference-in-difference analysis. Our comparison pairings, based on prior demographic analyses were as follows: NY with Pennsylvania(PA); AZ with New Mexico and Nevada (NM/NV).19 The time frame of our analysis was 1999 to 2006, with pre- and postexpansion periods defined as 1999 to 2000 and 2001 to 2006, respectively.
Data
We included veterans aged 18 to 64 years, seeking care for depression from 1999 to 2006, who were also VA-enrolled and residing in our states of interest. We counted veterans as enrolled in Medicaid if they were enrolled at least 1 month in a given year.
Using similar methods like those used in prior studies, we selected patients with encounters documenting depression as the primary outpatient or inpatient diagnosis using International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes: 296.2x for a single episode of major depressive disorder, 296.3x for a recurrent episode of MDD, 300.4 for dysthymia, and 311.0 for depression not otherwise specified.18,24 We used data from the Medicaid Analytic eXtract files (MAX) for Medicaid data and the VA Corporate Data Warehouse (CDW) for VA data. We chose 1999 as the first study year because it was the earliest year MAX data were available.
Our final sample included 1833 person-years pre-expansion and 7157 postexpansion in our inpatient analysis, as well as 31,767 person-years pre-expansion and 130,382 postexpansion in our outpatient analysis.
Outcomes and Variables
Our primary outcomes were comparative shifts in VA reliance between expansion and nonexpansion states after Medicaid expansion for both inpatient and outpatient depression care. For each year of study, we calculated a veteran’s VA reliance by aggregating the number of days with depression-related encounters at the VA and dividing by the total number of days with a VA or Medicaid depression-related encounters for the year. To provide context to these shifts in VA reliance, we further analyzed the changes in the proportion of annual VA-Medicaid dual users and annual per capita utilization of depression care across the VA and Medicaid. Changes in the proportion would indicate a relative shift in usage between the VA and Medicaid. Annual per capita changes demonstrate changes in the volume of usage. Understanding how proportion and volume interact is critical to understanding likely ramifications for resource management and cost. For example, a relative shift in the proportion of care toward Medicaid might be explained by a substitution effect of increased Medicaid usage and lower VA per capita usage, or an additive (or complementary) effect, with more Medicaid services coming on top of the current VA services.
We conducted subanalyses by income-eligible and service-connected veterans and adjusted our models for age, non-White race, sex, distances to the nearest inpatient and outpatient VA facilities, and VA Relative Risk Score, which is a measure of disease burden and clinical complexity validated specifically for veterans.25
Statistical Analysis
We used fractional logistic regression to model the adjusted effect of Medicaid expansion on VA reliance for depression care. In parallel, we leveraged ordered logit regression and negative binomial regression models to examine the proportion of VA-Medicaid dual users and the per capita utilization of Medicaid and VA depression care, respectively. To estimate the difference-in-difference effects, we used the interaction term of 2 categorical variables—expansion vs nonexpansion states and pre- vs postexpansion status—as the independent variable. We then calculated the average marginal effects with 95% CIs to estimate the differences in outcomes between expansion and nonexpansion states from pre- to postexpansion periods, as well as year-by-year shifts as a robustness check. We conducted these analyses using Stata MP, version 15.