To determine whether availability of behavioral health crisis care services is associated with changes in emergency department (ED) utilization. We used longitudinal panel data (2016-2021) on ED utilization from the Healthcare Cost and Utilization Project's State ED Databases and a novel dataset on crisis care services compiled using information from the Substance Abuse and Mental Health Services Administration's National Directories of Mental Health Treatment Facilities. A total of 1002 unique zip codes from Arizona, Florida, Kentucky, Maryland, and Wisconsin were included in our analyses. To estimate the effect of crisis care availability on ED utilization, we used a linear regression model with zip code and year fixed effects and standard errors accounting for clustering at the zip code-level. ED utilization related to mental, behavioral, and neurodevelopmental (MBD) disorders served as our primary outcome. We also examined pregnancy-related ED utilization as a nonequivalent dependent variable to assess residual bias in effect estimates. We extracted data on crisis care services offered by mental health treatment facilities (n = 14,726 facility-years) from the National Directories. MBD-related ED utilization was assessed by applying the Clinical Classification Software Refined from the Healthcare Cost and Utilization Project to the primary ICD-10-CM diagnosis code on each ED encounter (n = 101,360,483). All data were aggregated to the zip code-level (n = 6012 zip-years). The overall rate of MBD-related ED visits between 2016 and 2021 was 1610 annual visits per 100,000 population. Walk-in crisis stabilization services were associated with reduced MBD-related ED utilization (coefficient = -0.028, p = 0.009), but were not significantly associated with changes in pregnancy-related ED utilization. Walk-in crisis stabilization services were associated with reductions in MBD-related ED utilization. Decision-makers looking to reduce MBD-related ED utilization should consider increasing access to this promising alternative model.
Read full abstract