Opening Doors to Recovery: Recidivism and Recovery Among Persons With Serious Mental Illnesses and Repeated Hospitalizations

Institutional recidivism—the cycling in and out of hospitals and jails or prisons among persons with serious mental illnesses—is a major problem for community mental health systems. Recidivism is undoubtedly partly driven by fragmented or even inaccessible community services, lack of engagement by local stakeholders who could be partners in community support after hospitalization, and frequent police contacts and poor communication between mental health providers and the police. Upon discharge or release, such individuals often lack needed supports to successfully reintegrate into their community and face barriers to locating and navigating community services (1). Specialized programs that address these posthospitalization or postdetention problems are needed in order to reduce institutional recidivism and facilitate recovery.

The Georgia chapter of the National Alliance on Mental Illness (NAMI)—in conjunction with diverse local partners—developed the Opening Doors to Recovery (ODR) service model to both prevent institutional recidivism (inpatient psychiatric rehospitalization and arrests and incarcerations) and promote recovery (2). ODR was designed to align closely with the ten principles of recovery put forth by the Substance Abuse and Mental Health Services Administration (3). ODR promotes hope, is person centered (for example, a “meaningful day” is defined by using the client’s own definition), offers peer support, seeks to improve relationships in the family and with others, and capitalizes on participants’ strengths. ODR also assists in reconnecting the participant to the community, setting goals, finding and engaging in meaningful activities, and improving environmental factors, for example, by helping to secure safe housing and improve family and community engagement in the participants’ “circles of support.”

ODR has five key components. First, a mobile team of nontraditional community navigation specialists (CNSs) provides case management and recovery support. The team comprises a professional CNS, who is a licensed mental health professional (for example, a social worker); a peer CNS, who is a certified peer specialist; and a family CNS (4), who is a family member of someone with a serious mental illness with lived experience navigating the complex mental health system. As such, ODR builds on successes of both peer-led programs—such as the Wellness Recovery Action Plan approach (5) and the Health and Recovery Peer Program (6)—and family-focused services, such as NAMI’s Family-to-Family (7). Furthermore, the CNSs’ roles resemble those of community health workers in primary care (8,9) and patient navigators in cancer care (10,11). For example, CNSs are able to engage participants on a personal, nonclinical level; meet the participant in a home, community, or other nonclinical settings; share their own lived experience; and provide concrete assistance as part of navigation, for example, by transporting participants to services. The intensity of CNSs’ involvement with a participant and his or her circles of support varies by need, but at least one CNS meets weekly with the participant during face-to-face visits.

Second, the three CNSs provide community navigation, striving to become intimately embedded in the community through relationships with myriad service providers and community leaders and constantly “mapping” all local services and facilities that might be useful to a client’s recovery. In this way, CNSs serve as a catalyst for engaging, educating, and energizing the community to accept shared responsibility for supporting recovery. Community navigators assess participants’ strengths and needs, facilitate collaboration between participants and care providers, identify supports in the community, and engage in service planning (12). Facilitating an understanding of available resources and how to access them empowers persons with disabilities, including mental disabilities (13).

Third, CNSs continuously focus on four recovery domains: ensuring adequate treatment, finding safe housing, developing a meaningful day, and using technology to support recovery. Fourth, relationships with local partners underpin the ODR model. This “collaborative fusion” process engages diverse agencies and organizations that commit to support ODR, assist CNSs, and aid participants in their recovery. A meeting of these local partners takes place bimonthly. Collaboration among these partners is not just necessary to initially implement ODR—it is part of the ODR model per se.

Fifth, a highly innovative aspect of ODR is a novel linkage between local police officers and CNSs that aims to prevent incarceration through prebooking jail diversion when appropriate. The linkage consists of four steps. First, upon enrollment, participants give special consent for including a very brief disclosure that they are in the ODR program in a registry in the state’s criminal justice information system. Second, if an officer conducts a routine background check during an encounter with a participant, the officer receives an automated electronic message identifying the person as part of ODR and is asked to call a toll-free number that connects to the local mental health system. Third, the call taker immediately contacts one of the participant’s CNSs (whoever is on call). Fourth, CNSs work with the officer on the phone or at the scene to resolve the situation without arrest when possible and appropriate.

This study focused on effects of ODR on participants’ institutional recidivism and trajectories of recovery outcomes during a year of involvement with the ODR demonstration project. First, we relied on administrative data from two state agencies to assess number of hospitalizations, number of days hospitalized, and number of arrests during the year of participation in ODR compared with the prior year. Second, we assessed changes in recovery measures over time from baseline, while accounting for participants’ intensity of involvement with CNSs and baseline symptom severity.

Methods

Setting and Sample

The study took place in 34 counties in southeast Georgia, in conjunction with three community service boards (quasigovernmental public mental health agencies). Two teams of CNSs were employed by Gateway Behavioral Health Services in Savannah and Brunswick, and one each by Pineland Behavioral Health/Developmental Disabilities in Statesboro and Unison Behavioral Health in Waycross.

All participants were enrolled upon discharge after a stay of two or more nights at Georgia Regional Hospital at Savannah or one of three crisis stabilization units in the region and after having had an additional prior stay of two or more nights within the previous six months (thus establishing a history of inpatient psychiatric recidivism). Other inclusion criteria were 18–65 years of age, English speaking, diagnosis of a psychotic or mood disorder, discharge to the catchment area of one of the three community service boards, and capacity to give informed consent. Exclusion criteria included known or suspected intellectual disability or dementia or the presence of a serious general medical condition that could interfere with research participation. Research diagnoses were made by the mood disorders, psychotic disorders, and alcohol and drug abuse and dependence modules of the Mini-International Neuropsychiatric Interview (14); the interviews were conducted by trained research assessors, and the results were reviewed by a board-certified psychiatrist.

Procedures and Measures

Data were collected from four sources after investigators obtained institutional review board approval and informed consent from participants. First, after participants had been enrolled in ODR for 12 months, data on number of hospitalizations and number of days hospitalized in Georgia’s state psychiatric hospitals in the past two years were obtained from the Georgia Department of Behavioral Health and Developmental Disabilities. Hospitalizations in other facilities, such as private psychiatric hospitals or general medical hospitals—or admissions to a crisis stabilization unit—are not included.

Second, lifetime arrest data (including arrests in the past two years) for participants who had been enrolled in ODR for 12 months were provided by the Georgia Crime Information Center, a division of the Georgia Bureau of Investigation. These data pertain only to fingerprintable arrests in the State of Georgia. Furthermore, these arrest counts include arrests for charges that may have ultimately been dismissed, deferred, or acquitted. Arrests occurring in other states are not included. Comparing official arrest records from 12 months before and 12 months after enrollment has been used previously to determine the effects of an intervention (15).

Third, two trained research assessors interviewed participants at baseline, four months, eight months, and 12 months. Interviews lasted two to three hours at each time point, with the first conducted at the hospital or crisis stabilization unit at discharge and all others in the participant’s home or in community settings. We examined three primary recovery measures by using self-report data collected by the research assessors. Community functioning (for example, successfully managing money and day-to-day tasks, getting along with other people, and following recommendations of mental health staff) was measured with the 22-item Multnomah Community Ability Scale–Patient Version (MCAS-P) (16–20). Possible scores range from 22 to 110, with higher scores indicating better community adjustment. Internal consistency (Cronbach’s alpha) at the four time points ranged from .78 to .87.

Mental health recovery was measured with the 30-item Mental Health Recovery Measure (MHRM) (21,22), which includes such items as “I believe in myself,” “I socialize and make friends,” and “I am making progress towards my goals.” Possible scores range from 0 to 120, with higher scores representing better mental health recovery. Internal consistency ranged from α=.93 to α=.95. Quality of life was measured with the 32-item Quality of Life Inventory (QOLI) (23,24), which assesses domains such as health, self-esteem, work, creativity, friends, family, and community. Possible scores range from −96 to 96, with higher scores indicating better quality of life. Cronbach’s alphas (computed for the 16 satisfaction items without the 16 importance items used to weight the satisfaction items) were .82–.85.

Fourth, we examined five secondary recovery measures with data provided by the participants’ professional CNS four months, eight months, and 12 months after enrollment. Functioning was measured with the 20-item Daily Living Activities (DLA) Scale (25). Possible scores range from 20 to 140, with higher scores indicating better daily living skills (Cronbach’s α=.93–.97). For the remaining four measures, the ODR constructs of adequate treatment, meaningful days, safe housing, and use of technology were rated with four scales, each ranging from 1 to 5, with 1 indicating emerging, 2, making progress, 3, meeting benchmarks, 4, moving toward recovery, and 5, independent. Detailed definitions were provided for each score.

Total positive, negative, and general psychopathology symptom severity was measured with the researcher-rated 30-item Positive and Negative Syndrome Scale (PANSS) (26). Possible scores range from 30 to 210, with higher scores indicating greater symptom severity. Interrater reliability of PANSS subscale scores was examined by calculating intraclass correlation coefficients (ICCs) by using a two-way mixed (judges fixed) effects analysis of variance model in which the two research assessors were the fixed effect and 12 to 14 target ratings were the random effect (27). ICCs in the first year of the study (two assessors and 12 participants) were .98 for the positive symptoms subscale, .94 for the negative symptoms subscale, and .87 for the general psychopathology symptoms subscale. In the second year of the study, ICCs (two assessors and 14 participants) were .97, .79, and .75, respectively.

To derive a measure of intensity of involvement with CNSs, participants were classified as heavily involved versus less involved by using the following criteria. Being heavily involved meant that a participant had involvement with a CNS team throughout the year, on the basis of the research team’s knowledge and all available information from the CNSs, as well as documented consistent contact with at least one of the three CNSs throughout the year, as indicated by the CNSs’ data collection packets (collected at four, eight, and 12 months). Participants who met both of these criteria were classified as heavily involved; otherwise, they were considered less involved.

Data Analyses

Differences between the two time periods (with participants thus serving as their own historical controls) were examined by modeling counts with repeated-measures Poisson regression, a generalized linear mixed model with Poisson distribution and random intercept. To test hypotheses pertaining to recovery measures, we used linear mixed models, fit with time as a factor, to estimate each time point separately and thus did not assume any pattern of change over time. Evidence for significant changes from baseline and possible linear trends was tested by using customized hypothesis tests of the model coefficients. Missing data were handled in mixed models (available case analysis) by invoking the missing-at-random assumption, which does not require the data to be missing completely at random but assumes that the model is valid as long as the missingness is not related to the outcome, given the predictors in the model.

Because the study was not a randomized trial, we attempted to control for baseline symptom severity by including baseline PANSS score as a covariate in all analyses. We also considered intensity of involvement with CNSs as a possible predictor or moderator because we were interested in the possible effects of intensity of involvement on change over time. The intensity of involvement × time interaction was tested for all outcomes but was not significant, so given the limited sample size, we removed the interaction from consideration. All analyses were performed by using IBM SPSS Statistics 19.

Results

One hundred participants were enrolled in the study one to three days prior to discharge, and 22 others who were initially referred to ODR were screened for participation and were excluded from the study. The 22 persons who were referred but not enrolled did not differ significantly from the 100 enrolled participants in terms of age, gender, race (Caucasian or African American), referral source (the regional state hospital or a crisis stabilization unit), or diagnosis at hospital discharge (psychotic or mood disorder).

We restricted the sample from 100 to 72 because we wanted to include only those with an adequate level of exposure to ODR who were actually “eligible” to be rehospitalized or arrested in Georgia. Including other individuals could bias findings away from the null hypothesis because those who relocated to another state or were incarcerated long-term would appear to have fewer hospitalizations and fewer days hospitalized in state psychiatric hospitals in Georgia. Specific reasons for excluding participants included dropping out or being transferred out within six months of enrollment (for example, because of a referral to assertive community treatment) (N=18), relocating to another state within six months of enrollment (N=5), and being incarcerated for most of the year (N=5). No significant differences were found between the 28 participants who were excluded from the main analyses and the 72 participants who were included in terms of age, gender, race, baseline symptom severity, or baseline global functioning.

The sociodemographic and basic clinical characteristics of the 100 enrolled participants, as well as the 72 participants included in the analyses, are given in Table 1. Among the 100 enrolled participants, 68 completed in-person, researcher-administered follow-up assessments at four months, 60 at eight months, and 62 at 12 months. Comparing the 32 who could not be interviewed with the 68 who were interviewed at four months revealed no significant differences in terms of age, gender, race, diagnostic category, length of index hospitalization, number of days hospitalized in the year before enrolling in ODR, number of lifetime arrests, baseline symptom severity (positive, negative, manic, and depressive symptoms), or baseline global functioning and community ability.

CNSs provided data for 63 of the 100 enrolled participants at four months, 63 at eight months, and 65 at 12 months.

Recidivism (Hospitalizations and Arrests)

The number of hospitalizations and hospital days differed significantly in the years before and after enrollment in ODR (Table 2). Although a prior history of arrest was not an inclusion criterion for study participation, the participants (N=72) had a mean±SD of 5.6±9.2 lifetime arrests (range 0–60; median=2.0; mode=0), with 54 (75%) having had at least one lifetime arrest. The number of arrests differed slightly before and after enrollment in ODR, but the difference was not significant (Table 2). Because some participants were periodically ineligible for arrest because they were in the hospital, we double-checked this result by computing the number of days eligible for arrest, given that the number of arrests might otherwise appear to be equal or even greater during the year while enrolled in ODR simply because participants were hospitalized less. The arrest rate (arrests per nonhospitalized day) before and after ODR enrollment differed slightly, although not significantly (Table 2).

Recovery Trajectories

Significant linear trends indicating improvement were evident for all three primary measures (Table 3), although it should be noted that these recovery measures were somewhat redundant; for example, the mean intercorrelation between the MCAS-P, MHRM, and QOLI at the four-month assessment was .69. Notably, community ability improved rather quickly (within four months), whereas mental health recovery and quality of life took longer to improve, achieving significance only at 12 months postenrollment.

As shown in Table 4, significant linear trends indicating improvement were evident for all secondary recovery measures except use of technology (meaningful day was significant only at a trend level). Of note, intercorrelations among the ODR measures of adequate treatment, meaningful day, safe housing, and use of technology ranged from r=.37 to r=.62.

Discussion

We observed a significant reduction in the number of hospitalizations and a substantial, clinically meaningful, and significant reduction in the number of days hospitalized during the year of community navigation compared with the previous year. We did not find a significant difference in the number of arrests. Including a recent (past five years, for example) history of arrest or incarceration as an inclusion criterion—or implementing ODR specifically for persons with serious mental illnesses after release from jails or prisons rather than hospitals—might “enrich” the sample sufficiently to demonstrate a significant effect. For example, Chintakrindi and colleagues (15) recently documented a significant reduction in criminal justice recidivism among individuals with mental illnesses who received transitional case management. However, participants were drawn from a criminal justice setting and had an established history of multiple misdemeanors (average of 27.2 lifetime arrests and 3.6 in the 12 months before enrollment), in contrast to our study, in which patients were drawn from an inpatient psychiatric setting and there were no selection criteria for past arrests (average of 5.6 lifetime arrests and .49 in the 12 months before enrollment). Furthermore, the fifth component of ODR is designed to intervene and prevent incarceration when possible, although it might not prevent arrest.

A host of recovery measures significantly improved as participants took part in ODR. Furthermore, strides in recovery were apparent across the 12 months, indicating trajectories of improvement throughout the follow-up period and not just immediately following hospital discharge. Taken together, these data suggest that the ODR model is effective at preventing psychiatric inpatient recidivism and promoting recovery. Given the multifaceted and community-engaged nature of the intervention, the mechanisms of change are likely complex and involve both individual-level drivers (including better access to treatment services and enhanced hope and self-efficacy) as well as community-level mediators (including engagement of key community partners to build “circles of support”).

Evidence-based community support programs remain inadequate, despite the successes of approaches such as assertive community treatment (ACT) (28) and critical time intervention (29), which are not available in many communities. ODR is distinct from ACT and other commonly used models in public-sector community mental health agencies because it has a unique composition of team members, is supported by a broad group of community partners, employs a novel linkage between police and the CNS team, and was designed around recovery tenets. ODR has a distinctive thumbprint in that it is uniquely professionally involved as well as peer and family driven (given that only peers and family members can share the lived experience that is so highly relevant to recovery). Perhaps most important, ODR is based on a navigation model. Navigation programs address barriers facing individuals with chronic medical conditions during the recovery process, especially individuals in underserved populations (30,31).

Several methodological limitations are noteworthy. First, because our primary goal in this demonstration project was to show feasibility of implementation, we decided to enroll all eligible participants instead of allowing random assignment to a comparator group, which obviously would have yielded more persuasive results than having participants serve as their own historical controls. As such, we cannot definitively attribute the positive effects specifically to ODR. However, the continued trajectory of improvement throughout the follow-up period (as opposed to only four months) might indicate true improvements in recovery domains beyond hospital-release effects. We are currently conducting a randomized, controlled trial of ODR that involves a larger sample and random assignment to ODR versus traditional forms of case management.

Second, because ODR is composed of at least five core components—and we could not gather data on specific components separately—it is difficult to know which components are most important in reducing recidivism and promoting recovery. Third, although consistency across the four teams was a major focus during initial implementation (for example, weekly conference calls included all CNSs across the teams), we had no actual data on intervention fidelity or adherence to the five core components. Fourth, using administrative data, although highly accurate, incurred certain limitations; for example, the hospital data cover only admissions to state psychiatric hospitals in Georgia, and the criminal justice data pertain only to fingerprintable arrests in Georgia.

Conclusions

Public mental health systems need new, innovative approaches for addressing the problems of institutional recidivism and poor recovery. ODR appears to help persons with serious mental illnesses stay out of the hospital; as such, they may be better equipped to embrace recovery. Given the promise of the modern recovery paradigm for individuals with serious mental illnesses, ODR and other innovative models are deserving of testing—and, ultimately, dissemination once proven effective—in diverse community mental health settings. This study and future studies of ODR and similar interventions might have policy implications in an era of health care reform and integration of primary care and mental health care, given the similarities between the role of the CNS and the community navigation functions of other care navigation programs.