A systematic review of co-responder models of police mental health ‘street’ triage

All studies had to meet three criteria detailed in Table 1.

On the 30th April 2018, we carried out searches on the following databases: Ovid MEDLINE, Embase, PsycINFO, EBSCO CINAHL, Scopus, Thompson Reuters Web of Science Core Collection, The Cochrane Library, and ProQuest National Criminal Justice Reference Service Abstracts. We also searched grey literature via ProQuest Dissertations & Theses, EThoS, and OpenGrey. Finally, to obtain other grey literature we entered our search terms in to Google, limiting our search to the first 100 PDF results for each search term. A search strategy was developed for MEDLINE (see Additional file 1) and adapted for the other databases. The following free text terms and phrases were used to search the title and abstract fields to retrieve relevant literature: street triag*, policing or police and triag*, “mental health” and triag*, “liaison and diversion”, “speciali#ed mental health response*”, “co-responder”, coresponder*, “crisis intervention team*”, policing or police and “crisis team*”, “mental health” and “crisis team*”, policing or police and “mental health”, policing or police and liaison, “section 136”, or “psychiatric emergency response team*”.

As there is no consistent terminology for police mental health triage, these search strategies did not include MeSH terms. We mitigated this limitation by performing a second round of searches with a basic MeSH query to combine these sets of terms: (exp Mental Disorders/ or Triage/) and exp. POLICE/; Triage/ and exp. Mental Disorders/; Emergency Services, Psychiatric/ and exp. POLICE/. We inspected reference lists and forward citations of all included studies for further relevant literature.

Two teams of reviewers independently screened titles and abstracts (DP, AK, FD, AH). Any disagreements on eligibility were resolved through discussion with a third reviewer (SP). Following the title and abstract review, two reviewers conducted a full text review of eligible articles (DP, AK). We did not quality assess articles due to the broad range of methodologies included.

We conducted pilot searches in preparation for this systematic review, and during these it became clear that there were likely to be few, if any, randomised controlled trials of street triage. Therefore, we decided to conduct a narrative synthesis of our results. We extracted data from eligible studies using a standardised data extraction form. We contacted authors to request necessary information on triage model design or results data if it was missing from the article. This included data on: authors and publication type, study design, sample size and demographics, type of intervention, days and hours of operation, details of how the service was operated and staff roles, and details of outcomes recorded. We then grouped studies with similar measures and outcomes together in themes. We divided eligible articles into research studies and non-academic reports and report on them separately.

There were 19 different triage models described in the 26 articles (Table 2). Twelve used a ride-along model, where police officer and mental health worker attend the incident in the same vehicle. Five were services with both ride-along and control room support, in which the mental health worker assists the officers remotely via telephone or police radio. Four of these described a service which used the control room as their main method of response, only using a mobile unit for particularly severe incidents. Two services used telephone triage as the method of response. We were unable to determine the method of one study due to a lack of adequate description, and two studies were national surveys which did not request details of the triage model.

We were able to identify how incidents were reported to the triage team for 15 models. Two took calls from emergency control rooms (e.g. 911, or a police control room), four took calls directly from police officers in the field, eight took calls from both emergency lines and police officers, and one had a direct line that emergency services, members of the public, and relevant agencies could call. Of the 15 models that reported details of the vehicles used in triage calls, nine used marked police cars, three an unmarked car, and one used an ambulance alongside an unmarked car (two models were telephone only models). Nine models described a system whereby the mobile unit was dispatched only when a normal police unit had already responded and determined the incident was safe, while one described the mobile unit acting as a first response to an incident and six used a combination of both methods of response. Data on type of response was missing from three models and the two remaining articles were the national surveys which did not collect data on the model of triage.

Twelve models had triage services that were operated 7 days a week, but only the telephone support models were run 24 h a day. Three models had less than 7 days a week services. Times of operation varied greatly (see Table 2), but most services were contracted to cover evening and night-time hours only.

Many of the included studies did not look at effectiveness or outcome. Those that did took a varied approach and we report on a variety of aspects of effectiveness as a result.

Nine of the eligible documents were evaluations of a local street triage service [37‐45], one was a report on a service for strategic review [41], and one was a report on a pilot scheme of nine services [42]. Eight were from the UK, two from the USA, and one from Australia. Reports differed in both methodology and methodological quality. Eight articles measured changes to S136 before and after the introduction of a service (seven local and one wider evaluation) [37‐42, 46, 47]. Five local evaluations reported a decrease in the use of S136 [37, 38, 40, 41, 46] and two an increase [39, 42]. The report of nine evaluations found seven services with a decrease in S136 and two with an increase in S136 [47]. Five documents investigated detentions in police custody, with four finding a reduction in the use of custody [38, 39, 41, 42], and one finding no difference (although the use of custody was negligible both before and after introduction of the service) [38].

Nine documents reported on qualitative findings (eight local street triage services and the evaluation of nine sites described above) [37, 39‐44, 47]. Four interviewed police, three interviewed mental health staff, two interviewed service users, and one interviewed carers. In the report of the nine pilot sites, each site conducted different interviews with different methodologies and different samples. Results were similar to those of the academic studies. Service users found the triage service less distressing than standard police responses [37, 39]. Police officers valued the service and found it reduced the time spent at incidents [37, 38, 42, 43, 47]. Areas identified for improvement included longer operational times [37, 43, 47], and more resources [43, 47].

Few studies measured the demographic and clinical characteristics of triage service users or their reason for referral. From the data reported, those who are middle-aged and male are slightly overrepresented, as are those with psychotic illnesses and previous mental health history. The data we identified in both the academic and non-academic literature suggest that the majority of service users are already know to mental health teams, and between 10 and 20% have multiple referrals to the triage team.

Whilst psychiatric inpatient beds have declined over the last 60 years, the provision of forensic beds and the proportion of the mentally ill in prison has increased in both the US and Europe [5, 49]. Deinstitutionalisation without appropriate community treatment may have led to increased criminalisation of the mentally ill [50]. Police triage services are designed to divert those with mental illness away from police custody, but providing appropriate support prior to a crisis point may reduce the need for an intervention by emergency services [51]. There remains, however, a group of service users who come into contact with triage who have no previous mental health service use or known psychopathology. Measuring who is referred, their history leading up to referral, and the nature of the referral may improve our understanding of this vulnerable population. This may in turn allow us to be clear about the circumstances in which the use of the police-triage resource is appropriate and those in which it can be avoided. This could have significant implications for service use and most importantly for patient experience and outcome.

Many of the papers included in this review did not adequately describe their model of triage. Less than half reported operational hours and only two-thirds described their method of response to incidents. When contacting the authors, many of these details were available, which suggests issues of how triage is reported in the literature rather than poor model specification.

There was marked variation in how co-response models were operationalised. There were differences in times and days of operation, whether the unit was a first or second-response option, whether the police officer and mental health worker were co-located, whether a mobile unit was dispatched or not, and the mode of transportation to the incident (marked or unmarked vehicles). There was also limited, if any, information on other mental health provision in the study area.

Local context will always inform model implementation, and rightly so. There must, however, be efforts to understand which components of the model are most effective and most acceptable to service users, and equally which are not. Where good practice is identified, such as the use of unmarked cars and non-uniformed police officers, these can be rigorously tested.

The current heterogeneity in model implementation creates a problem of circularity, whereby service providers cannot follow good practice guidelines because there are none. At the same time, commissioners cannot establish good practice because each model is idiosyncratic and cannot easily be compared.

This may in part be due to the lack of precise reporting of co-response models. Reporting aspects such as the location and composition of members of the team, hours of operation, referral processes, all potential response methods, and mode of transportation should all be reported as a minimum dataset. In Table 3, we suggest a framework for reporting triage models in future studies to enable model comparison.

There is a striking lack of evidence for the effectiveness of co-response models of triage despite the substantial investment these services have received. We did not identify any randomised control trials. We identified one controlled before-and-after study, which found better engagement with mental health services following the intervention, but did not measure other important clinical follow-up data such as hospitalisation [26].

There would be significant challenges conducting a randomised trial of street triage, such as lack of blinding, consenting participants, and maintaining model fidelity. Whilst these difficulties are not insurmountable, other designs (such as longitudinal data with propensity score matching) may be more feasible. For instance, a quasi-experimental design such as propensity score matching could avoid the difficulty of gaining consent prior to randomisation by retrospectively matching those who have received a triage intervention with a sample who have not. This would incur less cost than an appropriate cluster randomised controlled trial, but would remain only an approximation of randomisation and would require careful covariate selection.

This lack of evidence needs to be urgently addressed as we currently have little understanding of both the immediate and the long-term clinical and functional outcomes following a triage intervention, nor any identification of potential adverse events resulting from it.

This is the first systematic review dedicated solely to investigating co-response street triage. We used a search strategy which included broad criteria and included many different databases and grey literature searches to ensure we captured all studies of street triage. We excluded reviews, book chapters, editorials, and non-English language articles which may have resulted in us missing some relevant literature.

We were unable to assess risk of bias due to the broad range of methodologies used in our included studies, and we decided to include this heterogeneous mix of studies rather than focus on any single methodology due to the sparse literature.