Estimating the impacts of Medicaid coverage expansions on psychiatric inpatient utilization and expenditures

Estimates of the impacts of Medicaid coverage expansions under the Affordable Care Act (ACA) on psychiatric inpatient utilization and expenditures are needed to inform mental health policy formation.  These impacts have implications for hospitals, especially in relation to psychiatric beds, personnel, and outpatient care linkages.  This manuscript examines the capacity of a unique national data resource, the Medical Expenditure Panel Survey (MEPS), to estimate these impacts, and examines methods for reweighting MEPS data to reduce bias and increase the efficiency of insurance coverage parameter estimates.

The MEPS is a national household survey of the U.S. non-institutionalized civilian population, and is one of the only databases available for obtaining a comprehensive population-level picture of the impacts of ACA insurance expansions on psychiatric inpatient utilization and expenditures. However, MEPS data may underrepresent actual levels of psychiatric inpatient service use.  The MEPS sample may under-represent individuals with behavioral health conditions and MEPS respondents may under-report their use of psychiatric services during personal interviews.

This study used auxiliary information from the State Inpatient Databases (SID), a collection of statewide hospital discharge abstract databases, to create new individual-level weights in the MEPS.  The new weights were used to estimate regression models, in which the dependent variables were total annual psychiatric inpatient bed days and expenditures and covariates included insurance coverage type (Medicaid, Medicare, private, other public, and self-pay), psychiatric diagnosis category (schizophrenia, mood and anxiety disorders, substance use disorders, and other), and demographic characteristics.

Two re-weighting methods were used. The first method is post-stratification, in which psychiatric inpatient utilization totals (i.e., total days) stratified by insurance coverage and diagnosis category were matched to MEPS data, and then used to up-weight underrepresented subgroups. The second method is an empirical likelihood approach (due to Chaudhuri, Handcock, and Rendall, J.R. Statist. Soc. B, 2008), in which moment restrictions from the target population are imposed on the MEPS data. The main advantage of the second method is that it directly incorporates information from the target population about the relationship between covariates (e.g., insurance coverage category) and the dependent variable. Both methods are available in statistical software packages.  

In unadjusted comparisons, MEPS data reflected less than half (45%) of total psychiatric inpatient days. The bias was greater (30% of total days) for Medicaid than it was for Medicare (40% of total days) or private insurance (80% of total days).  In regressions using the original MEPS weights, Medicaid was estimated to increase psychiatric inpatient expenditures 16.2% (t=1.14, p=.371) compared to being uninsured.  By contrast, using MEPS post-stratified weights, Medicaid’s estimated impact was 62.6% (t=6.00, p<.001), which was closer to the SID estimate of 61.1%.  Estimates based on the empirical likelihood approach are expected in early 2014.

These results indicate that MEPS data can be reweighted using auxiliary information to more accurately reflect the impacts of Medicaid coverage on psychiatric inpatient utilization and expenditures.  Data reweighting methods such as post-stratification and empirical likelihood reweighting are increasingly accessible to applied researchers. 

Source of Funding.  NIMH/NIH Common Fund R21-MH096285