Assertive community treatment (ACT) is a model of care that adopts a comprehensive, individualized approach for persons with severe mental illness. ACT was developed by Stein and Test in the 1970s to address the revolving-door phenomenon of repeated admissions at a state hospital in Madison, Wisconsin (1), and has subsequently been widely disseminated across the country as a standard-of-care treatment for this population (2). Essential elements of ACT include high-intensity services, multidisciplinary staffing, integration of services, a team approach, low client-to-staff ratios (ideally less than 10:1), service delivery in the community rather than in an office setting, assertive outreach, time-unlimited services, and continuous responsibility for a discrete group of clients (3). Clients on ACT teams generally demonstrate marked functional impairment, frequent inpatient hospitalizations, and high rates of comorbid medical conditions, substance use disorders, and homelessness.
ACT is one of the most widely studied psychosocial treatments for persons with severe mental illness, and the approach consistently has been found to be effective in reducing psychiatric hospitalization and improving housing stability for this population (4–6). Results are less consistent for quality-of-life outcomes, medication adherence, symptom severity, substance use, and vocational functioning (3,7,8). However, a recent study found that ACT had a significant and positive impact on one-year outcomes, including improved global functioning and quality of life, decreased illness severity, and greater likelihood of being employed, being medication adherent, and living independently (9). Interestingly, ACT has not been shown to be effective in reducing arrests and jail time for persons with severe mental illness. Yet, as a group, persons with severe mental illness are jailed more often than they are hospitalized (10). Furthermore, this population has a high rate of reincarceration. In a recent study in the Texas criminal justice system, over 50% of inmates with psychotic illnesses who were incarcerated over a one-year period had at least one previous incarceration over the prior six years (11).
Numerous randomized trials have investigated the efficacy of ACT, yet only nine have examined criminal justice outcomes, and all of those studies had substantial limitations (1,12–19). Specifically, sample sizes were small to moderate (ranging from 61 to 196), ACT interventions were well described but generally did not report fidelity measures, comparison groups generally received treatment as usual (outpatient community mental health center, drop-in center, or traditional case management), and studies were not specifically designed to address criminal justice outcomes. None of the trials demonstrated significant improvement in the ACT group, consistent with Marx and colleagues’ (1) seminal 1973 study reporting a statistically significant increase in the number of jail days for ACT clients during the five-month treatment period. The failure of traditional ACT models to improve criminal justice outcomes stimulated attempts to adapt ACT for clients with a history of forensic involvement, including the development of forensic ACT (FACT) (20). However, a clinical model for FACT has yet to be elucidated (10), and the three randomized trials of FACT that have been published have reported mixed results (21–23).
This study examined outcomes of ACT clients in New York State as part of a larger project to develop interventions to improve ACT discharge outcomes. We are not aware of any large-scale studies that have compared outcomes over time between ACT team clients with and without a history of forensic involvement. Such data may allow better estimation of the value of the standard ACT model for patients with forensic involvement and help to identify unmet needs. In this article, we describe a retrospective cohort study in which we used administrative data from a large ACT population in New York State to examine and explore risk factors predictive of vulnerability to four adverse outcomes of interest for forensic and nonforensic clients in a large state’s ACT population: arrest or incarceration, hospitalization, homelessness, and early discharge from ACT.
New York State operates 79 ACT teams. ACT services are delivered by a multidisciplinary team, with focus on assertive outreach, frequent contacts, around-the-clock coverage, and services delivered in the community. The state operates two models of ACT: a 68-client model (69 teams), and a 48-client rural model (ten teams). Staff-to-patient ratios are high (1:8) and require that staff include a psychiatrist, nurse, family specialist, employment specialist, and a specialist in co-occurring mental and substance use disorders.
The Child and Adult Integrated Reporting System (CAIRS) is a Web-based outcome-monitoring system developed by New York State for high-intensity mental health programs, including ACT (24). The database contains enrollment and outcome data reported by the ACT team every six months and at discharge for all clients. Baseline variables included demographic variables, date of ACT enrollment, date and reason for discharge, homelessness at enrollment and follow-up, hospitalizations at enrollment and follow-up, and forensic variables. At baseline, forensic variables included current forensic status (parole, for example) and number of incarcerations in the past six months and over the client’s lifetime. On follow-up the number of arrests and incarcerations in the past six months were examined.
Approval for this study was obtained from the institutional review board of the New York State Psychiatric Institute.
Inclusion criteria and comparison groups
All clients admitted to ACT between July 1, 2003, and June 30, 2007, with an initial assessment in CAIRS at ACT enrollment (N=4,756) were included in the sample. Seventy-seven ACT enrollees (1.6%) were excluded because of missing forensic status at baseline. We examined all follow-up data for this cohort between July 1, 2003, and June 30, 2008, allowing for a minimum one-year follow-up postenrollment. Clients were divided into three groups according to forensic status and history at ACT enrollment: recent involvement with the criminal justice system (arrests or incarcerations in the past six months or currently on probation or parole), remote involvement with the criminal justice system (more than six months before enrollment), and no prior criminal justice system involvement.
Four outcome variables were dichotomized as any versus none on follow-up: arrest or incarceration, psychiatric hospitalization, homelessness, and discharge from ACT.
Initial group comparisons examined sociodemographic and clinical characteristics at the time of ACT enrollment and rates for each of the four outcomes at one, two, and three years of follow-up. Continuous variables were examined with analysis of variance; categorical variables were examined with chi square tests. Group pairwise comparisons were performed with Tukey’s method for continuous variables, and generalized linear models were used for categorical variables. Within-group time effects over three years of follow-up were examined with the generalized estimating equation method.
Generalized linear multilevel modeling (GLMM) was used to estimate the effects of forensic status and sociodemographic and clinical characteristics determined at enrollment. Effects on outcomes included arrest or incarceration, psychiatric hospitalization, homelessness, and discharge from ACT. Four GLMM models were tested for each outcome: model 1 reviewed the effect of time on each outcome, model 2 added forensic status, model 3 tested the potential cross-level interaction of time and forensic status, and model 4 included all enrollment characteristics and retained any significant cross-level interaction effects.
We conducted a post hoc analysis of the ACT discharge outcome, excluding individuals who were discharged because they became incarcerated.
GLMM models used a binary response distribution with a logit link function and the Newton-Raphson optimization technique with ridging (25,26). Degrees of freedom were calculated with the Kenward Rogers method (26,27). Given the multiple comparisons, we adopted a conservative p value (<.01) in determining statistical significance in enrollment group comparisons. All reported statistical tests were two-tailed. All analyses were conducted with SAS 9.1.
Group differences at enrollment
The three study groups consisted of a total of 4,756 ACT clients. At enrollment, a majority of ACT clients in the study had schizophrenia spectrum diagnoses (69%) and had multiple prior psychiatric hospitalizations (Table 1). Compared with the nonforensic group, the recent group was significantly more likely to be younger (mean age of 36 versus 41 years), male (75% versus 52%), and nonwhite (64% versus 58%); to have lower educational attainment (35% versus 22% without a high school diploma or GED); and to have a recent history of homelessness (29% versus 13%) and problematic substance use (41% versus 23%).
Table 1Characteristics of assertive community treatment clientele, by forensic involvement (N=4,756)
| Add to My POL
(N=792; 17%)||2. Remote
(N=429; 9%)||3. Nonforensic
|Characteristica||Total N||N||%||Total N||N||%||Total N||N||%||Statistical test||df||p||Pairwise comparison|
|Age (M±SD)||792||36±12||429||41±11||3,535||41±13||F=56.55||2 and 4,753||<.001||1<2*, 1<3*|
|High school diploma or GED||695||455||65||380||272||72||2,973||2,310||78||χ2=47.64||2||<.001||1<2*, 1<3***,
|Recent homelessnessd||682||198||29||367||84||23||2,923||381||13||χ2=112.94||2||<.001||1>2*, 1>3***,
|Problematic substance usee||689||283||41||377||134||36||2,808||646||23||χ2=104.49||2||<.001||1>3***, 2>3***|
|Schizophrenia or schizoaffective disorder||792||525||66||428||276||65||3,528||2,485||70||χ2=10.15||2||<.01||1<3*, 2<3*|
|Lifetime psychiatric hospitalization (M±SD)||747||7±11||412||10±14||3,370||6±11||F=28.46||2 and 4,526||<.001||1<2*, 1>3*, 2>3*|
|Global Assessment of Functioning score (M±SD)f||679||46±9||393||46±10||2,972||45±10||F=5.12||2 and 4,041||<.01||1>3*|
|Outpatient commitment||749||153||20||414||59||14||3,391||666||20||χ2=7.64||2||<.05||1>2**, 2<3**|
The recent group had high rates of arrest and incarceration in each of three yearly follow-up periods (33% in year 1 and 27% in years 2 and 3) (Table 2). For each follow-up period, the arrest and incarceration rate was highest in the group with recent forensic involvement, followed by the remote and nonforensic groups. This pattern persisted in the final model, which adjusted for characteristics at enrollment; clients with a recent forensic history had 6.18 greater odds of arrest or incarceration during the third year of follow-up, and clients with a remote history had 2.66 greater odds of arrest or incarceration than those without a forensic history (Table 3). Significant predictors of arrest or incarceration included problematic substance use at enrollment (odds ratio [OR]=2.98), history of homelessness (OR=1.67), and male gender (OR=1.40), with Caucasian race being protective (OR=.72). Cross-level interactions of time and forensic status were nonsignificant.
Table 2Three-year outcomes for assertive community treatment (ACT) clients with recent, remote, or no forensic involvement
| Add to My POL
|Arrested or incarcerateda||Psychiatric hospitalizationb||Homelessnessc||Discharge from ACTd|
Table 3Predictors of three-year outcomes for assertive community treatment (ACT) clients, by generalized linear multilevel modelsa
| Add to My POL
|Arrest or incarceration||Psychiatric hospitalization||Homelessness||Discharge from ACT|
|Characteristic||β||SE||OR||95% CI||p||β||SE||OR||95% CI||p||β||SE||OR||95% CI||p||β||SE||OR||95% CI||p|
|Time (in years)||–.08||.06||.92||.82–1.03||.16||–.34||.03||.71||.66–.76||<.001||–.07||.09||.45||.39–.53||<.001||–.24||.03||.78||.73–.84||<.001|
|Forensic history (reference: none)|
| Time × recentb||–.56||.18||.003|
| Time × remoteb||–.15||.23||.56|
|Sex (reference: female)||.34||.11||1.40||1.12–1.75||.003||–.02||.06||.97||.86–1.11||.71||–.03||.12||.97||.75–1.24||.81||–.04||.05||.96||.86–1.08||.51|
| Homeless (reference: not homeless)||.52||.14||1.67||1.27–2.20||<.001||–.07||.11||.93||.75–1.14||.49||3.17||.14||24.01||18.24–31.58||<.001||.23||.08||1.26||1.06–1.50||.01|
| Assisted outpatient treatment (reference: none)||.07||.13||1.07||.81–1.39||.73||.31||.08||1.36||1.15–1.59||<.001||–.05||.17||.93||.66–1.31||.70||–.15||.07||.86||.75–.99||.04|
| Education level (reference: high school or GED)||–.18||.11||.84||.68–1.02||.08||–.05||.06||.96||.84–1.08||.49||–.30||.12||.74||.58–.94||.014||–.11||.05||.89||.80–.99||.05|
| Substance abuse (reference: none)d||1.09||.11||2.98||2.42–3.56||<.001||.36||.07||1.43||1.23–1.66||<.001||.65||.13||1.97||1.50–2.51||<.001||.12||.06||1.13||.99–1.28||.06|
Rates of psychiatric hospitalization were equally high across groups, ranging from 40% to 42% in year 1. Hospitalization rates did not differ significantly by forensic status at years 1, 2, or 3 and declined over time for all groups (p<.05, Table 2). The multilevel model confirmed this pattern (Table 3): with controls for enrollment characteristics, each year was associated with significantly lower odds of hospitalization (OR=.71). Assisted outpatient treatment status (OR=1.36) and substance abuse problems (OR=1.43) at enrollment were associated with significantly increased odds of hospitalization during follow-up for all groups (Table 3). No significant cross-level interactions of time by forensic status were found.
Rates of homelessness were highest in the first year after enrollment for all three groups and decreased significantly (p<.001) with each year of ACT treatment ( Table 2). Differences in rates of homelessness among groups were most marked in the first year after enrollment (27% for recent, 17% remote, and 9% nonforensic) and became nonsignificant by the third year.
The multilevel model (Table 3) suggested a significant time-by-forensic status interaction, such that clients with a recent forensic history had a greater decline in homelessness over time, compared with those without a forensic history (p<.003). Those in the remote and nonforensic groups declined at similar rates (p<.56). Upon further investigation of the cross-level interaction, the patterns reported in Table 2 persisted after analyses controlled for client characteristics at enrollment. The odds of homelessness were greatest in the first compared with the third year of follow-up for all groups (recent history, OR=9.96, 95% confidence interval [CI]=4.96–20.00; remote history, OR=5.0, CI=1.98–12.61; and nonforensic, OR=4.09, 95% CI=2.75–6.07). Furthermore, the odds of homelessness were significantly higher in the first year after enrollment in ACT for clients with a recent forensic history compared with those without a forensic history (OR=1.91, CI=1.33–2.74), but there was no difference in the odds of homelessness between those with a recent history and those with a remote history (OR=1.56, CI=.93–2.61). After the first year, however, there were no differences among the three groups with regard to homelessness after analyses controlled for client characteristics at enrollment.
Rates of discharge from ACT within the first year after enrollment were significantly higher in the recent (24%) compared with the remote (17%) and nonforensic (17%) groups (Table 2). The odds of discharge from ACT significantly decreased over time (OR=.78, 95% CI=.73–.84), and those with a recent forensic history were significantly more likely than those without a forensic history to be discharged (OR=1.35) after analyses controlled for characteristics at enrollment (Table 3). Clients with a remote forensic history were as likely to be discharged as those without a forensic history. No significant time-by-forensic history interactions were noted. Across groups, discharged clients were more likely to have a history of homelessness at enrollment (OR=1.26) and less likely to receive assisted outpatient treatment (OR=.86) or to have a high school diploma or GED (OR=.89).
Discharges from ACT related to incarceration
Incarcerations accounted for 147 (8%) of the total discharges during the study period, 75 (51%) of which occurred in the first year. Incarceration as the reason for discharge was most common for the group with recent forensic involvement, representing 24% of discharges (92 of 383), followed by the remote group with 11% (18 of 161) and the nonforensic group with 3% (37 of 1,329). For the recent group, incarceration accounted for a decreasing proportion of discharges over time, representing 29% of discharges (54 of 184) in the first year, 22% (29 of 130) in the second year, and 14% (nine of 64) in the third year. Removing discharges due to incarceration from the analysis resulted in the recent group’s no longer being significantly different from the nonforensic group on the discharge outcome (OR=1.16, CI=.98–1.36). Although individuals receiving assisted outpatient treatment were less likely to be discharged from ACT for all causes across groups, assisted outpatient treatment was no longer protective for discharge after we eliminated discharges resulting from incarceration (OR=.89, CI=.77–1.03).
This article presents findings from a large statewide administrative database of ACT teams’ enrollment and outcome data. Recidivism rates were high among individuals with recent forensic histories, with 33% rearrested or incarcerated within one year of enrollment. Although rates may have been significantly higher had these individuals been receiving less intensive treatments, their high risk of recidivism is noteworthy. Indeed, ACT clients with a forensic history at enrollment, compared with those without a forensic history, had higher rates of arrest or incarceration even after three years of ACT services. Arrest or incarceration was the only outcome that did not significantly improve over time, even after we corrected for baseline characteristics. This finding is consistent with a compelling body of literature that indicates that criminal history is among the strongest predictors of forensic recidivism (28).
ACT clients with a recent forensic history were more likely to be discharged from ACT within the first year of enrollment, which may raise concerns that ACT teams were not able to engage these clients as effectively as those without a forensic history. This finding persisted after we controlled for baseline characteristics and consequently cannot be fully explained by the higher rates of substance abuse and homelessness observed in this population. However, when we excluded ACT discharges resulting from incarceration, the clients with a recent forensic history were no longer significantly different from nonforensic clients on this engagement outcome. These findings suggest that ACT teams can engage forensic clients as well as nonforensic clients and that recidivism is a primary barrier to ongoing access to ACT services for the forensic population.
ACT was initially designed to reduce rates and duration of psychiatric hospitalization for individuals with severe mental illness. Hospitalization rates were equally high in the first year of ACT (40%–42%) for forensic and nonforensic clients alike and decreased similarly for all groups over time, with no significant differences among groups at any time point. This suggests that any impact of ACT on hospitalization is the same for forensic clients and nonforensic clients on this primary ACT outcome.
Rates of homelessness decreased significantly for all groups of ACT clients over time, and by the end of the three-year study period, the rate of homelessness for the forensic groups was not significantly different from that of the nonforensic group. This finding suggests that ACT may have been effective in increasing housing stability among individuals with severe mental illness, including those with forensic involvement. Individuals with a recent forensic history had higher rates of homelessness at baseline and in the first year with ACT (27% versus 9%). This effect persisted after analyses controlled for baseline characteristics, but it was no longer significant after year 2 with ACT. Indeed, homelessness is the single outcome on which there was a significantly greater improvement for forensic clients than for nonforensic clients. These findings underscore the need for housing services for this population and suggest that ACT services can support positive housing outcomes for the forensic population.
ACT clients with forensic histories, particularly those with active and recent correctional system involvement, have multiple risk factors for poor outcomes. Forensic clients admitted to ACT teams had higher rates of active substance use and recent homelessness, more lifetime hospitalizations, and lower rates of high school graduation or equivalent education than individuals admitted to ACT without a forensic history. In this study, forensic clients were particularly vulnerable to adverse outcomes in the first year of ACT, including arrest or incarceration, homelessness, and early discharge from ACT, even with controls for baseline characteristics. Consequently, it may be valuable to incorporate clinical approaches that focus attention on this vulnerable period, including identifying and addressing modifiable risk factors and targeting recidivism.
One clinical approach may be to adapt critical time intervention (CTI) strategies for ACT. CTI is an evidence-based model of homelessness prevention for clients with severe mental illness that focuses on the vulnerable transition period from institutional settings to the community (29,30). Draine and Herman (31) have been evaluating the feasibility of extending CTI into correctional settings. CTI supports engagement by having CTI workers establish relationships with clients during their institutional stays. Indeed, various FACT programs described in the literature, including the Thresholds FACT team and Project Link, both of which won American Psychiatric Association Gold Awards, begin to work with clients while they are still incarcerated to develop an alliance with them before their release, which is a vulnerable transition period (32,33). The question of whether CTI could be incorporated into ACT is worthy of investigation.
Two modifiable risk factors, substance abuse and homelessness, were associated with increased recidivism and homelessness at follow-up. The integrated model of care delivered by ACT teams can support identification and mitigation of these modifiable risks. ACT teams have substance abuse specialists on staff, and integrated services for substance use disorders are components of high-fidelity ACT teams.
ACT clinical and program models for delivering housing services and supports are less well specified; for example, there are no ACT fidelity items dedicated to housing, although several items focus on integrated treatment for co-occurring mental disorders and substance use disorders (34). One approach to incorporating housing services is the Housing First model (35). This specialized ACT program engages individuals with severe mental illness by focusing on what clients typically identify as their first priority (housing) through the provision of independent, permanent housing without prerequisites. This contrasts with more traditional models that require individuals first to demonstrate psychiatric stability, abstinence from drugs and alcohol, and “housing readiness” by progressing through a series of supervised short-term living arrangements. The high baseline levels of homelessness in the forensic population, the multiple risk factors for poor outcomes, and the marked improvement in housing outcomes observed in this study suggest that this is an important area for future research.
It is important to note that this study examined outcomes for ACT clients on the basis of time since last arrest or incarceration at enrollment, with the assumption that those with a more recent history are at higher risk. We did not examine degree of recidivism within the forensic cohorts at baseline. Therefore, the recent forensic cohort represented individuals who were first-time offenders and individuals who were highly recidivistic. The remote cohort was a similarly diverse group, because inclusion was based on only six months without an incarceration. Examination of ACT outcomes for forensically involved clients in regard to their degree or their pattern or type of recidivism is an area for future study.
Our study had several limitations. First, we did not have a comparison group to examine outcomes compared with treatment as usual (for example, outpatient treatment at a community mental health center). Consequently, we cannot on the basis of this study comment on whether ACT was more helpful for clients than standard care. Although clients on ACT teams had attenuating risks of homelessness and hospitalization over time, we cannot rule out the possibility that this finding represents regression to the mean. In addition, the highest-risk clients may have been discharged from ACT teams at higher rates, whereas those who remained on the team may have been less vulnerable to adverse outcomes. Second, our data were derived from state administrative data reported by treating clinicians. Criminal justice databases may have been a more reliable source for arrest or incarceration data. However, because of the frequency and in vivo nature of ACT services, it is likely that ACT staff members are more aware than treatment providers in other settings about adverse outcomes such as episodes of homelessness, arrest or incarceration, and hospitalization. Finally, our risk of spurious associations was increased given the multiple comparisons. To minimize type I error we used Tukey’s method and a more conservative level of significance for comparing enrollment characteristics.
This study offers several important findings. First, there was significant risk of arrest or incarceration among clients with forensic histories on ACT teams and a significant risk of homelessness and early discharge from ACT among those with a recent history of criminal justice involvement. This finding underscores the need for ACT teams to incorporate interventions targeted at improving forensic outcomes in this population. Doing so will require clinical interventions to mitigate modifiable risk factors such as homelessness and substance abuse. Alternatively, systems with sufficient numbers of forensically involved patients in need of ACT services might consider the recommendation of Morrissey and colleagues (10) to develop a clinical model for FACT that incorporates interventions designed to target criminal behaviors and jail recidivism. Second, there were no significant differences in rates of psychiatric hospitalization between forensic and nonforensic clients on ACT teams, suggesting that ACT is no less effective for this population with respect to reducing the frequency of psychiatric hospitalizations. Finally, although rates of psychiatric hospitalization, homelessness, and ACT discharge declined over time for all clients, the initial year of ACT enrollment was a particularly vulnerable period for ACT clients, especially for clients with a recent forensic history. This finding highlights the importance of developing strategies to improve engagement and reduce adverse outcomes in the first year of ACT.
This work was supported by the New York State Office of Mental Health, the New York State Psychiatric Institute, the Columbia University Psychiatric Residency Program, and Bristol-Myers Squibb Foundation. Dr. Finnerty’s time on this project was supported by the New York State Office of Mental Health; she is the principal investigator on the ACT Transitions grant from Bristol-Myers Squibb Foundation. Dr. Manuel’s time was supported by the Bristol-Myers Squibb Foundation. The authors thank Veronica Hackethal, M.D., for assistance in manuscript preparation.
Dr. Appelbaum reports an equity interest in COVR, Inc., for violence prediction software. The other authors report no competing interests.