The FACT program studied here was carried out from 2000 to 2003 in the central city (pop. 189,000) of a large agricultural county in California. The FACT program was staffed and operated by the county behavioral health department through MIOCRG funding administered by the county sheriff’s office. The program had high fidelity to the ACT model with Dartmouth Assertive Community Treatment Scale (Teague et al. 1998
) scores of 4.5 and 4.6 (range 1–5) during its first and second years of operation. Participants in the FACT program received team-based mental health and substance abuse services, as well as support for housing, employment assistance, benefits applications, and advocacy. The FACT program also had a full-time peer recovery specialist. Each member of the team had primary responsibility for providing and/or coordinating services appropriate for designated consumer participants. Psychiatric and medication services were available to the participants through a part-time psychiatrist and registered nurse.
The full-time probation officer worked with the courts and the participants to establish conditions of probation (e.g., mandated participation in FACT, restitution (if applicable), agreement to avoid substance abuse) that encouraged participation in behavioral health services. She also tracked pre-existing and subsequent arrests, incarcerations, and jail days. If participants were incarcerated during the period of the study, jail staff ensured that the participants’ medications were maintained and that a FACT staff member was notified of any changes.
Participants randomly assigned to the treatment as usual (TAU) condition were not provided any additional services beyond those routinely available in the county-operated public behavioral health system. The range of usual services included psychiatric assessment, psychiatric medications in both outpatient and inpatient general hospital settings, outpatient mental health and substance abuse counseling, and case management.
At the time of FACT enrollment participants were detained in the county jail and were diagnosed by FACT clinicians with a major mental disorder on Axis I of the DSM-IV. Potential participants were excluded if they were ever charged with a serious, violent offense (e.g., murder, forcible rape, aggravated assault) as defined in Penal Code section 667 (n = 33), and/or if they were a “third strike” candidate (n = 46). Participants were not excluded if they had a secondary substance abuse diagnosis or an additional Axis II or Aix III diagnosis.
Jail inmates were first identified by correctional mental health staff through initial screening. Inmates screening positive for history of mental illness, current suicidal ideation, or current/past use of psychotropic medications by corrections staff were then referred to a member of the research team for a full mental health assessment. Assessments were conducted by the FACT team within three working days from the date of referral. During the assessment, individuals were informed of the purpose of the study and consent was obtained. After final determination of eligibility by the team, participants were randomly assigned into the experimental treatment (FACT) or treatment as usual (TAU). Using a table of random numbers, 134 individuals were assigned in blocks of two, to either the FACT group (N = 72) or the TAU comparison group (N = 62). Assignment by random numbers achieves equivalency between-groups (see below) but not necessarily equal group size.
The current study did not rely on the data reported to the California Board of Corrections (CBOC) as part of the MIOCRG evaluation. The statewide evaluation conducted by the CBOC used a randomization design but it failed to follow the logic of an experiment where follow-up observations on subjects are recorded at common intervals following enrollment (exposure time). Rather, MIOCRG sites were required by the CBOC to report outcome data on all enrollees every 6 months (calendar time) regardless of when participants were enrolled. As a result, although person-level data on symptoms and quality of life were collected on all participants, these measures were tied to calendar time and we were not able to convert them to an exposure time format.
Consequently, we relied upon administrative data (see below) obtained from the county jail and behavioral health authorities to carry-out this study. Administrative data had several advantages. First, they were available for all participants independent of the schema employed in data collection for the MIOCRG evaluation. Second, it was possible to array them in a prospective cohort fashion for intervention and control subjects. And third, they allowed us to implement an intent-to-treat design whereby outcomes were observed regardless of active or continued participation in services during the 2-year follow-up period. Furthermore, it seemed reasonable to assume that any inaccuracies in the administrative data would be equally distributed in the intervention and control groups, thus unlikely to bias our results.
Finally, it should be noted that funding for the statewide MIOCRG program was reduced midway through the fourth year of its planned, 5-year lifespan due to budgetary cut-backs by the California state legislature (CABOC 2004
). These budget reductions led to the premature ending of the evaluation. Except as noted below, this did not prevent us from accessing the relevant administrative data for the 2-year follow-up period.
Criminal Justice Involvement
Information on bookings, convictions, and jail days was obtained through administrative data collected by the county jail. Data were obtained for 1 year prior to study enrollment and up to 2 years after enrollment. Information on the cost per day in jail ($100) was obtained from the county sheriff’s office which operated the jail.
Due to premature termination of the MIOCRG funding, hand collection of criminal justice data ceased in study months 13–24, resulting in missing bookings, jail days, and convictions for the last 20 participants (10 FACT and 10 TAU) enrolled in the study. We compared the subset of participants with missing second year data with those who had complete data to see if there were any differences on baseline variables. There were no significant differences at baseline between these two conditions on treatment group assignment, demographic, clinical, or criminal justice variables. Therefore, our criminal justice analyses were based on 62 TAU and 72 FACT participants in the first 12 months, but on a reduced sample of 52 TAU and 62 FACT participants for the second 12 months.
Behavioral Health Services
Administrative data on behavioral health service use on all enrollees 1 year prior to enrollment and 2 years after enrollment were obtained from the California Department of Mental Health (CDMH). These data were routinely reported to CDMH by the county behavioral health authority as part of the state-wide service reporting system. The data elements obtained included hospital admissions, number of days hospitalized, psychiatric crisis contacts, and outpatient services for both mental health and substance abuse. Unit costs (per day or visit, etc.) for each service type are based on the California Medicaid rates (MediCal) and were also obtained from the CDMH. Unit costs were summed separately for inpatient and outpatient services to the person-month level and then aggregated into first and second year totals. No other cost indicators were used for these analyses.
Descriptive statistics are presented on the sample of participants (). Chi-square and t-test statistics were computed to determine differences between groups on these characteristics. Data on criminal justice involvement (number of bookings, jail days), mental health services (hospital days, outpatient mental health visits), and costs (jail, mental health) were analyzed cumulatively for months 1–12 and months 13–24 post-randomization.
Participant characteristics at baseline
Each of these variables represents a count of events, the distributions of which are often characterized by over-dispersion and excess zeros. Depending upon their distributions, these variables were analyzed using count models including negative binomial (outpatient visits, crisis contacts, booking, convictions, and outpatient costs) and zero-inflated negative binomial regression (inpatient days, jail days, inpatient costs, and jails costs). We first tested the fit of various count models including poisson, negative binomial, and zero-inflated versions of both. Zero-inflated models are appropriate in situations where the dependent variable has an excess of zeros (non-events). Conceptually, zero-inflated models estimate two parts. The first part of the model predicts the probability of the behavior occurring. The second part predicts the intensity or amount of the behavior conditional on its occurrence (Cox et al. 2009
). Results for the count portion of the model are presented in the form of incident rate ratios (IRRs). IRRs correspond to the rate (count) of the event among the FACT condition (exp(β0
*x1)) divided by the rate (count) among the TAU condition (exp(β0
)). Odds ratios (OR) are reported for the zero-inflated portion of the model and correspond to the odds of having no instances of the event (jail or hospital days) in the follow-up period for the FACT compared to TAU condition. The IRRs and ORs are obtained by exponentiating the coefficients from either the count portion or logistic portion of the model, respectively. IRRs less than 1.0 indicate that FACT participants had a reduced rate compared to TAU whereas IRRs greater than 1.0 indicate that FACT participants had an increased rate compared to TAU participants. All analyses were conducted in SAS 9.1.3.
The county behavioral health agency approved the informed consent and study protocol. These procedures were also approved by the Institutional Review Board at the CDMH and the University of North Carolina at Chapel Hill.