Data and Study Population
We used state-level data on the number of hospital-based psychiatric facilities, residential care settings, CMHCs, partial hospitalization/day treatment settings for the years 2014-2017 for 50 U.S. states and the District of Columbia (DC) from the National Mental Health Services Survey (N-MHSS) [7]. The N-MHSS is an annual survey that collects information from all known facilities providing mental health services in the U.S., including psychiatric hospitals, nonfederal general hospitals with separate psychiatric inpatient units, CMHCs, and partial hospitalization/day treatment facilities. All facilities reported their treatment characteristics during the survey, including settings of care (inpatient, residential, partial hospitalization/day treatment, or outpatient) and the provision of suicide prevention services. One objective of the N-MHSS is to update SAMHSA’s inventory of all known mental health and substance abuse treatment facilities. To our knowledge, this is the only comprehensive source of national data on specialty mental health facilities and their scope of clinical services. The N-MHSS began distinctly identifying federal- and state-licensed CMHCs in 2014. In all four study years, N-MHSS excluded mental health facilities that were 1) Department of Defense military treatment facilities, 2) individual private practitioners or small group practices not licensed as a mental health clinic or center; and 3) facilities in jails or prisons [7]. All licensed psychiatric hospitals, hospitals with inpatient psychiatric units, residential care settings, and CMHCs (including partial hospitalization/day treatment settings) that meet state licensing or certification requirements are eligible for inclusion in the survey. Mental health facilities that have closed since the previous-year survey are excluded. During the study period, response rates were 88.1% of 16,687 eligible facilities in 2014; 91.9% of 14,573 eligible facilities in 2015; 91.1% of 13,983 eligible facilities in 2016; and 93.0% of 13.618 eligible facilities in 2017, with item response rates averages of 96.9%, 97.9%, 97.6%, and 98%, respectively [7].
Data on state-level mental health professionals came from the Occupational Employment Statistics 2014-2017, which produced employment estimates for 415 industry classifications by state. Based on the North American Industry Classification System, occupations related to mental health care include a) psychiatrists, b) psychiatric technicians, c) psychiatric aides, d) clinical, counseling, and school psychologists, e) all other psychologists, f) mental health counselors, h) mental health and substance abuse social workers [19]. Detailed definitions for each professional can be found in the US Bureau of Labor Statistics (19). In particular, psychiatric technicians and aides are certified to have the privilege caring for people who have mental illness.
State-level annual suicide mortality was derived from the Centers for Disease Control and Prevention’s (CDC) Wide-Ranging Online Data for Epidemiologic Research (WONDER). For each state in each year (2014-2017), we considered population size and number of deaths from suicides (intentional self-harm; ICD-10-CM diagnosis codes U03, X60-X71, X72-X74, X75-X84, and Y87.0). All 50 states and the District of Columbia had at least 35 suicide deaths in each year of our study.
Measures
Community Mental Health Centers and Hospital-Based Psychiatric Care Settings
N-MHSS respondents were asked which of the following categories best describe their facility7: 1) psychiatric hospital, 2) separate inpatient psychiatric unit of a general hospital, 3) residential treatment center for children, 4) residential treatment center for adults, 5) other type of residential treatment facility, 6) Veterans Administration Medical Center or other VA health care facility, 7) Community mental health center (CMHC), 8) Outpatient mental health facility, 9) Multi-setting mental health facility (nonhospital residential plus either outpatient and/or partial hospitalization/day treatment). This study focused on CMHCs because these facilities were more likely to accept uninsured or Medicaid-insured patients, to offer suicide prevention services, psychiatric emergency walk-in services, case management, and other specialty practices, compared to other outpatient or multi-setting facilities (Appendix Table 1). N-MHSS respondents received a link of descriptions of each facility type [7]. A CMHC was defined as a facility that provided any of the following services: 1) outpatient services, 2) 24-hour emergency care services, 3) day treatment or other partial hospitalization services, or psychosocial rehabilitation services, and 4) screening for inpatient services to state mental health facilities, and that met applicable licensing or certification requirements for community mental health centers in a state where it is located. Beginning in 2015, a new category, “partial hospitalization/day treatment facility” was added, leading to a separate category for CMHCs to choose, should a CMHC primarily focuses on partial hospitalization/day treatment services [7]. In addition to the self-reported CMHC status, non-hospital mental health facilities that reported providing both outpatient services and day treatment or other partial hospitalization services were also categorized as a CMHC in this study. Using the total number of CMHC/partial hospitalization/day treatment facilities (hereafter called CMHCs) in conjunction with U.S. Census state population estimates, we calculated the number of CMHCs per 100,000 persons in each year-state.
To address changes in the hospital-based inpatient psychiatric supply by state, we considered hospital-based psychiatric services in all regressions. When determining the availability of hospital-based psychiatric settings, we calculated the number of psychiatric hospitals or separate inpatient psychiatric units of a general hospital per 100,000 persons each year.
Mental Health Professional Supply
Overall changes in the supply of individual mental health professionals in each state per year are essential for suicide prevention and for facility provision services, as difficulty in staffing may result in facility closures. Thus, in all models, we also included state-level number of psychiatrists, psychiatric technicians, psychiatric aides, clinical, counseling, and school psychologists, all other psychologists, mental health counselors, or mental health and substance abuse social workers, per 100,000 persons.
Covariates
In accordance with variables described in previous literature [4, 6, 20], we calculated the following covariates using U.S. Census Bureau data to control for relevant population-level characteristics: percent population by age group (less than 15 years old, 15-24 years old, 25-44 years old, 45-64 years old, 65-74 years old, 75 years old or more); percent race/ethnicity (White Non-Hispanic, Black/African American Non-Hispanic, Hispanic, Asian, American Indian and Alaska Native, Native Hawaiian or other Pacific Islander, and multiple race individuals); percent of state residents below 200 percent Federal Poverty Levels (FPL).
Statistical Analysis
We first visually plotted mean state CMHCs rates and mean state suicide rates pooled over four years (2014-2017) to better understand the cross-sectional associations between these variables.
Because state availability of CMHCs and hospital-based psychiatric units varied substantially across states and from 2014 to 2017, we used state-level variations in the timing and size of changes in CMHCs and hospital-based psychiatric units per 100,000 persons to identify the independent associations of changes in the supply of CMHCs and hospital-based psychiatric units with suicide rates, controlling for state-level mental health professionals per 100,000 persons in each year.
We used multivariate generalized linear time series models to analyze changes in state suicide mortality and sequentially included state and year fixed effects, as well as sociodemographic, and socioeconomic characteristics in a series of models. To identify possible multicollinearity between covariates, we used variance inflation factors (VIFs) [21]. The state-level proportion of population below 200 percent FPL was highly collinear (VIF>10) with other socio-economic factors (proportion of population that was unemployed and proportion of individuals younger than 65 without health insurance), as were unemployment rates and uninsured rates among population ages <65. Therefore, we controlled for these factors in separate multivariate models, in addition to unobservable characteristics unique to each year and each state by having year and state fixed effects.
In all models, we weighted observations by state-year population. This also accounts for state heteroscedasticity, as variability in suicide rates may be inversely correlated with state population. Given the time-sequenced nature of the suicide data, all models generated robust standard errors, accounting for intra-state correlated variances across years, to adjust for possible residual autocorrelations.
The final model included four observations for each state, and included the state-year number of CMHCs per 100,000 persons, number of hospitals with psychiatric services per 100,000 persons, number of mental health professionals per 100,000 persons, percent population below 200% FPL, percent population by age group, percent population by race/ethnicity, state and year fixed effects. Finally, to better illustrate within-state association between the availability of CMHCs and suicide rates, we graphed the adjusted suicide mortality rates for each state-year against availability of CMHCs in the analogous state-year, linking the four data points from each state.
All analyses were conducted using SAS (version 9.4), and Stata (version 15); p-values < .05 were considered as statistically significant. The Institutional Review Board at the authors’ University designated this study exempt from review.