Evaluation of a Police–Mental Health Co-response Team Relative to Traditional Police Response in Indianapolis

Police are often the first and only agency to respond to mental health crises (1), often serving as “street corner psychiatrists” (2) to deescalate situations and navigate community-based services. Police officers often settle contacts informally by mediating disputes or making referrals; however, when officers determine that a situation involving a person with mental illness may escalate without intervention or believe community-based behavioral health treatment or diversion services are not available, hospitalization or incarceration can occur (3, 4). Consequently, persons with mental illnesses frequently become entangled with the criminal legal system rather than engaged in needed treatment (3, 5–7). As many as one in four people with mental disorders have been arrested (8).

Criminal justice stakeholders have developed interventions along the sequential intercept model (SIM) to address overrepresentation of this population in correctional settings (9). The SIM describes six intercepts (0–5) at which a person could be diverted from the criminal justice system; the earlier a person is diverted, the less likely that person is to penetrate the system. One such prebooking program is crisis intervention team (CIT) training, an approach that incorporates some evidence-based components at intercept 1 to divert or address the needs of persons with mental illnesses at an initial police encounter (10, 11). In the CIT model, officers complete 40 hours of training in diagnoses of mental disorders, treatment, deescalation tactics, community resources, and techniques to safely and respectfully interact with persons with mental illness (12, 13). Findings show improved attitudes, increased treatment referrals, and less use of force by officers with CIT training (13–16).

Co-responding police–mental health teams are another police-based diversion strategy (17). Also called a co-response team (CRT), this model partners a police officer with a social or medical service provider, such as a mental health clinician, licensed social worker, or medical professional (paramedic, nurse, or psychiatrist). Some CRTs provide first response following calls for service (18–21), some provide follow-up after initial police encounters (22, 23), and some provide both (24–28). CRTs share common goals with CIT, including a focus on diversion and increasing treatment access; many of these programs use CIT-trained police officers. However, there is limited research on the effectiveness of CRTs in achieving these goals. Most studies have been descriptive and focused on program implementation, staff perceptions, time on scene for street officers, and emergency hospital utilization (19, 20, 24, 25, 27–30). One retrospective study found that a CRT response was less likely to result in psychiatric hospitalization but had no impact on arrest (27). Few contemporary studies have compared the CRT with police response as usual (27, 31, 32). Furthermore, no studies have examined longer-term emergency medical services (EMS) and criminal justice outcomes or how follow-up care may affect outcomes.

We conducted a prospective, quasi-experimental evaluation of a CRT consisting of a first-response unit and a follow-up behavioral health unit (BHU). We matched 5 months of CRT responses to police treatment-as-usual responses for behavioral health–related calls for service. Our research objectives were to examine the effect of a CRT response on the likelihood of an immediate arrest and emergency psychiatric detention relative to treatment as usual, the effect of a CRT response on subsequent arrest and EMS outcomes over 6- and 12-month follow-ups relative to treatment as usual, and the sensitivity of outcomes to BHU follow-up.

Methods

Setting

The CRT model was piloted in one Indianapolis Metropolitan Police Department (IMPD) district between August 1 and December 31, 2017 (30). The CRT consisted of a CIT-trained police officer, a master’s-level mental health clinician from a community mental health network, and a local EMS paramedic. The CRT is a self-dispatching unit that responds to relevant 911 calls for service heard over police dispatch radio. Relevant calls include those that involve a person with suspected mental or behavioral health or substance use issues, indicate a need for a CIT-trained officer, involve frequent utilizers of emergency services, or report a person with suicidal or homicidal ideation. Over 2 months, CRT members jointly completed training on behavioral health, crisis deescalation, use of force, and legal implications of interagency collaboration, among other topics. Participating officers were also CIT trained. The extensive training curriculum was developed by police and mental health leadership.

The BHU is a team consisting of a CIT-trained officer and a mental health clinician that contacts individuals following emergency behavioral health incidents within 48 hours to support them in obtaining and maintaining medication regimens; connect them with treatment, recovery, or other community services; provide limited case management services; and provide appointment transportation. Police refer all persons admitted to the hospital under emergency detention for BHU follow-up; they also refer those with behavioral health symptoms who did not meet criteria for emergency detention. According to Indiana statute (Section 12-26-4), law enforcement may place residents under emergency detention for 24 hours if an officer has reason to believe that the individual is mentally ill, dangerous to self or others, or gravely disabled and in immediate need of hospitalization and treatment. BHU follow-up services may be phone or in-person visits. During the study period, the BHU team provided follow-up on some CRT cases as well as on traditional police responses. EMS was the only other emergency response option outside the CRT and police, but EMS personnel were not trained in CIT and could provide only hospital transportation.

Overall Approach and Data Sources

We conducted a prospective, quasi-experimental study of outcomes following CRT or treatment-as-usual response to behavioral health calls for service. This study received approval from the Indiana University Institutional Review Board (IRB 190198344). Researchers developed a secure record-keeping system for CRT members to input service call, demographic, and response decision information following completed responses during the pilot period. This system captured all CRT responses, representing 318 calls for service over the study period. Each CRT response was manually matched to a treatment-as-usual case in a separate Indianapolis police district. The district for treatment as usual was selected in consultation with police and mental health providers on the basis of similar rates of calls involving mental health crises. Emergency calls were eligible for treatment as usual if they met established CRT eligibility criteria, including mental or emotional, substance abuse, self-harm, check-the-welfare, and related emergencies. Manual cell matching narrowed the pool of plausible treatment-as-usual cases. For each CRT record, researchers first queried the police records database for similar call types that occurred in other police districts within 2 days of the incident of study. Next, the queried cases were narrowed down by race, sex, and person’s date of birth for a 1-to-1 match with each CRT case.

We linked cases using individual first name, last name, and date of birth to Indianapolis EMS data and jail data from the Marion County sheriff’s office to determine counts of all EMS contacts and jail booking events prior to and following the incident of study. Following data linkage between the two study conditions, we removed two cases in which the person in crisis was reported as dead on arrival, five cases that represented duplicated individuals across both study conditions, and one case in which age information was missing. Our final sample consisted of 313 CRT responses and 315 treatment-as-usual responses (N=628). CRT responses were characterized by a self-dispatch to a behavioral health call for service (59%, N=186) in which the CRT co-responded with the IMPD (90%, N=280), EMS (40%, N=126), and Indianapolis Fire Department (18%, N=57). In 76% of encounters (N=235), the CRT relieved these agencies, and in 5% (N=15) of encounters, the CRT was the sole responding unit.

Participants

Prior to data weighting for the analysis, the number of participants was 628; participants received either a CRT or treatment-as-usual response to an eligible crisis event. Participants’ mean±SD age was 36.9±14.14. Most were male (N=383, 61%) and White (N=353, 56%); 262 participants (42%) were Black, and 13 (2%) were from other racial groups. In the year prior to dispatch, participants had a mean of 0.63±1.64 and 1.58±4.09 EMS contacts.

Variables

Covariates.

Covariates included age (years), race (Black, White, or other), gender (male or female), and crisis call type. Call types included “assist,” usually meaning a call from a fire department, EMS, or other law enforcement agency for backup; “check the welfare,” usually meaning a call to check on a person who seems to be in trouble; or “person in progress,” usually meaning a call by an active witness reporting a crime or other activity occurring in the moment. We also measured number of jail bookings (count) and EMS events (count) in the year prior to initial dispatch.

Study condition.

Study condition was defined by a CRT or treatment-as-usual response. To understand sensitivity of outcomes to BHU contact, we conducted comparisons by using alternatively defined study conditions. First, we conducted another two-group comparison between all CRT participants (N=313) and treatment-as-usual participants who did not have a BHU follow up (N=229). Second, we conducted a three-group comparison between CRT participants with any BHU follow-up during the study period (N=238) relative to CRT participants with no BHU follow-up (N=75) and to treatment-as-usual participants with no BHU follow-up (N=229). Third, we conducted a three-group comparison between CRT participants with BHU follow-up within 48 hours (N=208) relative to CRT participants with no BHU follow-up within 48 hours (N=105) and treatment-as-usual participants with no BHU follow-up within 48 hours (N=229).

Outcomes.

First, we measured two immediate outcomes following the initial CRT or treatment-as-usual response. Immediate jail booking (yes or no) indicated whether an individual was admitted into the county jail within 24 hours of the examined CRT or treatment-as-usual 911 call. The 24-hour lag was included to account for persons who may have been taken to a health care facility before being arrested and booked into jail. Emergency detention (yes or no) indicated whether an individual was involuntarily admitted for hospitalization during the emergency response to the examined 911 call.

Second, we assessed the presence and frequency of subsequent EMS contacts and jail bookings at 6 months and 12 months following the 911 call. Any EMS contact (yes or no) measured whether a person experienced a subsequent EMS encounter at 6- or 12-month follow-up. EMS contacts (count) measured the total number of EMS encounters for an individual at 6- and 12-month follow-ups. Any booking (yes or no) measured whether a person was booked into jail at 6- or 12-month follow-up, and bookings (count) measured the number of times an individual was booked into jail at 6- and 12-month follow-up.

Analytic Procedure

Because we could not manually match CRT and treatment-as-usual conditions on all characteristics, we additionally conducted propensity score matching. Full matching is a flexible matching procedure that is ideal for smaller and unbalanced sample sizes because it matches all treatment and control cases on a 1:k ratio (33). For two-group comparisons, we conducted propensity score matching by using MatchIt in R (34). We conducted each matching procedure twice for both two-group comparisons (i.e., CRT versus treatment as usual and CRT versus treatment as usual with no BHU follow-up). All covariates were included in each matching procedure. To evaluate propensity scores, we assessed absolute standardized mean difference (ASMD) before and after matching. We used an ASMD value of 0.15 to indicate covariate imbalance, below accepted threshold values for imbalance (35, 36). To further examine sensitivity of outcomes to BHU follow-up, we conducted three-group propensity score matching by using Twang in R (37). Three-group propensity score models were conducted by using an average treatment effect (38). Two iterations of this procedure were conducted to examine the impact of any BHU contact (before or after 48 hours) among CRT participants or BHU contact within 48 hours.

Following matching, we estimated weighted logistic regression models for emergency detention and booking outcomes, as well as dichotomous measures of arrests and EMS contacts. Negative binomial models were used for remaining count outcomes (i.e., EMS contacts and arrests). Where relevant, we report predicted probabilities by using average marginal effects. For two-group comparisons, we additionally report unweighted chi-square analyses to examine differences in outcomes by study condition for Black and White participants. All analyses were conducted with Stata, version 16.

Results

Descriptive

Table 1 presents descriptive statistics for the sample overall and by study condition following weighting. After weighting, participants’ mean age was 37.10; 52% (N=327) were White, and 61% (N=382) were male. In the weighted sample, participants had an average of 1.84 EMS responses and 0.63 jail bookings in the year prior to the examined 911 call.

Multivariable Models: Short-Term Outcomes

Emergency detention.

Results of a weighted logistic regression model showed no difference in likelihood of emergency detention between CRT participants (26%, 95% confidence interval [CI]=20.7%–30.4%) and treatment-as-usual participants (25%, 95% CI=18.1–32.1). Supplemental analyses (full models available upon request) showed that emergency detention was unrelated to likelihood of an EMS contact at either 6- or 12-month follow-up and did not moderate the effect of study condition on these outcomes. Unweighted chi-square analyses by racial group showed that this trend was consistent for both Black and White participants.

Immediate jail booking.

Results of a weighted logistic regression model showed a significant between-group difference in likelihood of an initial jail booking (odds ratio [OR]=0.48, 95% CI=0.25–0.92, p=0.027). Specifically, CRT participants were less likely to have been booked into jail within 24 hours of the initial response (5%, 95% CI=2.4–7.2), compared with the treatment-as-usual group (9%, 95% CI=6.2–12.8). Supplemental analyses showed that immediate jail booking increased the likelihood of a subsequent jail booking across both follow-ups (OR=3.75–4.07, p<0.001 for all). However, immediate jail booking did not moderate the effect of study condition on likelihood of subsequent booking (p≥0.063 for all). Unweighted chi-square analyses by racial group showed that this trend was driven by Black participants, who had lower rates of immediate jail booking (3%) in the CRT condition relative to the treatment-as-usual condition (15%; p=0.001). There was no significant effect among White participants.

Multivariable Models: Long-Term Outcomes

CRT versus treatment as usual.

Results of multivariable models comparing outcomes between CRT and treatment-as-usual participants are presented in Table 2. No differences were found in likelihood of any arrest at either the 6- or 12-month follow-ups. However, unweighted chi-square analyses by racial group showed that Black participants had lower rates of rearrest at 12-month follow-up in the CRT condition (25%), compared with Black participants in the treatment-as-usual condition (37%; p=0.029). There were no similar effects among White participants or for either group at 6-month follow-up. Similarly, we found no differences in the overall number of arrests across either follow-up period.

However, CRT participants were more likely than treatment-as-usual participants to have any EMS contact at 6- and 12-month follow-ups (OR=1.71–1.85, p≤0.015 for all). At 6-month follow-up, 38% (95% CI=32.7%–43.4%) of CRT participants had at least one EMS encounter, compared with 26% (95% CI=19.3%–33.6%) of treatment-as-usual participants. At 12-month follow-up, 46% (95% CI=40.7%–51.8%) of CRT participants had an EMS encounter, compared with 32% (95% CI=24.3%–39.3%) of treatment-as-usual participants. Unweighted chi-square analyses by racial group showed that these trends were likely driven by White participants, who had higher rates of any EMS contact in the CRT condition at 6- and 12-month follow-ups (41.9% and 49.4%, respectively), compared with Whites in the treatment-as-usual condition (18.2% and 23.2%, respectively; p<0.001 for all). No similar effect of CRT on any EMS contact was noted among Black participants at either follow-up. No differences in the number of EMS encounters at 6- or 12-month follow-up were found in weighted models.

CRT versus treatment as usual without BHU.

Table 3 presents comparisons of CRT participants and treatment-as-usual participants who did not receive follow-up services from the BHU. No differences in arrest likelihood were noted at either the 6- or 12-month follow-up, nor was there a difference in number of arrests. However, CRT participants were more likely than treatment-as-usual participants to have any EMS contact at 6- and 12-month follow-ups (OR=1.94–2.37, p≤0.012 for all). At 6-month follow-up, 38% (95% CI=32.6%–43.3%) of the CRT group had at least one EMS encounter, compared with 24% (95% CI=15.7%–32.4%) of the treatment-as-usual group. At 12 months, 46% (95% CI=40.6%–51.6%) of the CRT group and 27% (95% CI=18.4%–35.1%) of the treatment-as-usual group had an EMS encounter. No between-group differences were found in the number of EMS encounters.

Sensitivity analyses.

First, we examined differences between CRT participants with BHU follow-up within 48 hours and CRT participants without 48-hour BHU follow-up and treatment-as-usual participants with no BHU contact. As shown in Table 4, we found no significant difference in any arrest or in the number of arrests among the three groups, regardless of BHU follow-up. CRT participants both with and without BHU follow-up within 48 hours were more likely than treatment-as-usual participants with no BHU contact to have any EMS contact at the 6- and 12-month follow-ups (p≤0.029 for all).

Second, we examined differences between CRT participants with any BHU follow-up (within or not within 48 hours) compared with CRT participants with no BHU follow-up and treatment-as-usual participants with no BHU follow-up (Table 5). We found no differences in the likelihood of any arrest or in the number of arrests. However, CRT participants who received BHU follow-up were more likely to have any EMS contact at the 6- and 12-month follow-ups, compared with the treatment-as-usual participants with no BHU follow-up (p≤0.001 for all). In addition, CRT participants with BHU follow-up had a higher number of EMS contacts, compared with treatment-as-usual participants at 12-month follow-up (p=0.026).

Discussion

We conducted a prospective, quasi-experimental investigation to examine the effectiveness of a CRT model that self-dispatches to mental health crisis calls for service and provides 48-hour follow-up. Our findings suggest that CRT response led to a lower likelihood of arrest at the initial incident but a higher likelihood of subsequent EMS involvement. These differences could have reflected issues of selection bias. Specifically, the CRT may have responded to places or persons where they perceived the most need (i.e., where CRT members recognized a frequent utilizer of emergency services). Preliminary, exploratory analysis suggested that these trends differed by race, such that Black individuals were significantly less likely to be arrested at the initial event, compared with Black participants in the treatment-as-usual condition. In contrast, higher rates of EMS utilization appeared to be driven solely by White individuals. Importantly, although we adjusted for prior EMS encounters, we were unable to discern and match cases based on other characteristics (e.g., behavioral health call for service) of those encounters. Following matching, we found little evidence that the CRT was effective in reducing subsequent jail bookings or EMS contacts, compared with treatment as usual. However, some preliminary evidence suggested that the CRT was effective in reducing reincarceration risk among Black individuals specifically.

It is promising that Indianapolis’s CRT resulted in reduced likelihood of an immediate arrest, compared with the traditional police response, particularly for Black individuals. However, given the absence of positive long-term effects on justice system or EMS outcomes, our findings suggest areas of further inquiry. Specifically, it would be important to seek further information on the types of charges for which those in the CRT and treatment-as-usual groups were booked into jail and filed with the local prosecutor. Our findings suggest the potential for the CRT model to reduce longer-term criminal justice involvement among Black individuals in crisis. However, further work is needed to examine the cumulative effects of diversion on overall case processing.

Furthermore, the mechanism through which Indianapolis’s CRT response resulted in higher rates of subsequent EMS contact, particularly among White individuals, is difficult to identify. In fact, both the CRT and the treatment-as-usual groups had increases in EMS contacts following the examined 911 call, although EMS responses among CRT participants increased more dramatically, from an average of less than one encounter in the year prior to an average of more than two encounters in the year after. Overutilization of EMS services by persons with a mental illness may be driving unintended outcomes (39–42). Perhaps EMS involvement in the Indianapolis CRT resulted in individuals’ perceiving EMS as an effective means of managing a crisis and thus increased utilization of subsequent services. Future studies should aim to eliminate potential CRT response bias through randomization of calls for service.

Despite the presence of the BHU unit, which functions to connect individuals to treatment and relieve need for emergency services, BHU follow-up did not mitigate the rate of subsequent EMS events. Given limited treatment data, we do not know whether the CRT or BHU successfully linked persons in crisis to behavioral health services or the extent to which individuals engaged in treatment. Future studies should include measures of treatment engagement to understand whether this factor can mitigate EMS utilization rates. Additional follow-up services should also be tracked and their impact considered. Notably, an implementation study of Indianapolis’s CRT highlighted a lack of treatment availability in the community as a barrier (30). Other CRT studies have noted patient-, family-, and systems-level barriers to engaging persons with mental and behavioral illness in treatment (18–27, 29). Without adequate treatment engagement, CRTs are unlikely to have a long-term impact on subsequent emergency services utilization.

Conclusions

CRT models continue to proliferate, with limited evidence of their effectiveness. Our findings suggest CRTs may reduce short-term incarceration risk but may not have a positive impact on long-term outcomes. Future evaluations of CRT models should aim to employ rigorous research designs and measure treatment engagement following CRT response. More broadly, given variations in CRT models, achieving a model of program fidelity will require studies of program implementation to document specific features of CRTs and examine their effectiveness in improving client outcomes.