Police and mental health clinician co-responder models in crisis care: process-level effects are consistent, downstream outcomes are not

Introduction

Police officers act as the default first responders to mental health crises in most jurisdictions, which imposes a substantial burden on policing and has been linked to perceived criminalisation of people with complex mental health needs.¹ Administrative data illustrate the scale of demand: in Queensland, Australia, mental health-related calls to police rose from 32,040 in 2016 to 50,755 in 2021, and officers spent an average of 3.13 hours per such call in 2021.² Estimates of the share of police incidents carrying a mental health component vary widely, from as little as 2% to nearly 50% in United Kingdom sources, so definition and recording practice may account for much of that spread.⁴ A cross-sectional survey of frontline staff in Aotearoa New Zealand, covering 57 police officers, 29 paramedics and 33 mental health professionals, found that 89% of police officers and 62% of ambulance staff considered access to mental health expertise difficult, and police and ambulance respondents reported feeling undertrained for this work.³

Two families of intervention have developed in response. Crisis intervention team programmes give specialist mental health training to police officers, whereas co-responder or street triage programmes place a mental health clinician alongside officers, either in a vehicle, at a police station, or by telephone from a control room.^1,4 Terminology is inconsistent across countries, and the same label may denote a ride-along second-response unit in one site and a remote advisory service in another.⁴ The aims stated for these services are typically to reduce detention in police custody, to limit distress during the encounter, and to improve access to treatment afterwards.⁴

Here we review what the available research reports about co-responder models specifically. We consider incident-level process outcomes such as detention and hospital transport, downstream arrest and justice involvement, the clinical activities that may mediate any effect, the determinants of implementation, and what service users say they prefer. Much of the material addresses co-response directly; the remainder supplies mechanism, workforce context or adjacent-population evidence, and is labelled as such where it is used. Because the designs available are largely observational, the appraisal weighs how each study was built as closely as what it reported.

Reviews of incident-level process outcomes

The most comprehensive synthesis confined to co-response is a systematic review by Puntis and colleagues, which screened 11,553 records and included 26 articles reporting 23 studies from Australia, Canada, the United Kingdom and the United States.⁴ No randomised controlled trial was identified, and the authors did not quality-assess included papers because of the breadth of designs. Their narrative synthesis concluded that street triage might reduce the number of people taken to a place of safety under section 136 of the Mental Health Act 1983, or reduce the use of police custody in jurisdictions without that power.⁴ The included work was mostly cross-sectional in character, with mixed-methods and qualitative studies predominating alongside three before-and-after studies, two retrospective case-note reviews and a single health economics paper, and only 23% of articles measured data at more than one time point, over follow-up of 3 to 24 months.⁴

A rapid evidence synthesis funded by the English National Institute for Health Research searched eleven bibliographic databases from inception to November 2017 and included five systematic reviews together with eight primary studies reporting quantitative data and eight reporting qualitative data.⁵ Most interventions involved officers working in partnership with mental health professionals, and staff generally valued them. The synthesis reported some positive effects on procedures, such as rates of detention, and on resource use, but the results were not entirely consistent and several important outcomes went unmeasured. Most of the evidence was judged at risk of multiple biases arising from design flaws or unreported methods, and all primary research came from England, so generalisation beyond that setting appears uncertain.⁵

Two later syntheses point the same way, with similar caveats. The rapid review by Marcus and Stergiopoulos, described in the featured study box, found co-responder models superior to police-only models on the outcomes reported, although the evidence was mixed and the underlying studies were mostly uncontrolled.¹ A scoping review following the JBI methodological framework extracted data from 29 papers and reported that co-response teams improved interagency collaboration and service coordination and reduced jail bookings and emergency department transport, while evidence on long-term cost-effectiveness remained limited.⁶ That review characterised co-response as comprising specialised call handling, a joint police and mental health team, and a community hub for follow-up, and reported that service users felt better understood and better communicated with than in police-only responses, while officers gained confidence and further crisis training.⁶

Diversion at the incident compared with arrest over time

The strongest quantitative signal for co-response concerns what happens at the incident itself. An analysis of 10,904 behavioural health crisis incident records from the Massachusetts Department of Mental Health jail diversion programme database, covering May to December 2023, used logistic regression together with hybrid machine learning to identify predictors of emergency department diversion.⁷ Co-response clinicians achieved the highest diversion rates, ahead of crisis intervention team-trained officers, and community-based assessments appeared to confer a modest additional improvement. Evening and overnight incidents, police referrals, severe psychiatric presentations and substance use were each associated with lower probability of diversion, demographic variables also predicted outcomes, and the authors concluded that expanded community mental health resources were needed, particularly during off-hours.⁷

Effects measured over a longer horizon appear weaker. An exploratory quasi-experimental study in Michigan compared 474 recipients of co-response, mobile crisis or office-based crisis services with 690 people who received law enforcement-only responses to mental health calls across five jurisdictions in 2021.⁸ Inverse probability of treatment weighting balanced groups on age, sex, jurisdiction and prior-year arrest, and a weighted Poisson regression modelled arrests over the eleven months following the first thirty days after the index crisis. All three crisis models showed fewer post-year arrests than their local comparison groups, but only mobile crisis reached statistical significance, with an incidence rate ratio of 0.548; the authors placed their co-response result within an existing null arrest literature.⁸

Service design may account for part of that contrast. The two mobile crisis teams in the Michigan study responded around the clock to requests from homes, schools, law enforcement and other community members, whereas the co-response units operated ten-hour shifts on four days each week and drew referrals from law enforcement and dispatch.⁸ Race was excluded as a weighting covariate because it was missing in 10.5% of cases and the missingness was systematically related to programme type, so residual confounding cannot be ruled out.⁸ Larger administrative comparisons are feasible without randomisation: one Canadian study included in the Puntis review used 4,607 records to describe triage service users and 18,969 records to compare street triage incidents with police-only incidents.⁴

These findings are not necessarily in conflict, because they measure different things: immediate disposition at a single encounter versus cumulative arrests over roughly a year. A commentary on equity in crisis services describes the same pattern from another evaluation, in which a co-responder team composed of a clinician and an officer reduced short-term incarceration risk but not longer-term justice involvement, with initial findings suggesting a significant reduction in incarceration among Black recipients.⁹ The same commentary cites unpublished Arizona data indicating that Medicaid beneficiaries seen by mobile crisis teams and crisis facilities were more likely to be booked into jail within 30 days of a crisis episode, which runs counter to the Michigan result and remains unpublished.⁹

The clinician's role on scene

Qualitative work has begun to specify what embedded clinicians actually do, which is the plausible mechanism for any process-level effect. A study of the Boston Police Department co-response programme analysed clinical notes from 4,111 co-response encounters recorded between July 2019 and March 2022, using an inductive approach combining grounded theory and thematic analysis.¹⁰ The resulting framework had three domains: interaction with police personnel, people in crisis, family members and health and social service providers; assessment of symptoms, history and risk indicators; and provision of clinical support, assistance to police, and general assistance. The authors present these as candidate mechanisms rather than demonstrated causes of outcome change.¹⁰

Comparative focus group work suggests that the disciplines involved frame the same scene differently. A descriptive qualitative study conducted in the United States between 2022 and 2023 ran four focus groups: 12 police officers with a mean of 21.3 years of experience, 13 co-response clinicians with a mean of 13.8 years, 6 sexual assault nurse examiners and 4 correctional nurses.¹¹ Safety and de-escalation emerged as shared priorities, while nurses described a broader assessment encompassing physical health, substance involvement and injury; the authors proposed a nursing role within first-response multidisciplinary teams.¹¹ That report cites estimates that 10% to 20% of emergency calls to police involve a behavioural or mental health incident, which may indicate the potential reach of any first-response redesign.¹¹ The same focus group study appears twice in the retrieved literature, which should not be read as independent replication.¹²

Evidence assembled in the background of a realist review protocol points in a similar direction, although at second hand and without appraisal of the underlying papers. A Canadian study is reported there as finding that co-responses led to more peaceful resolutions than traditional police responses, and a United States study as finding fewer arrests and physical confrontations under co-response.¹⁶ De-escalation by an accompanying clinician is the mechanism most often invoked in that literature, though it has rarely been measured directly.¹⁶

Whether these mechanisms operate outside large cities has been examined once. An evaluation of a co-responder team in a geographically isolated mid-sized city in northwestern Ontario drew on programme documents, interviews with frontline and leadership staff, and ride-along site visits, analysed within an extended Donabedian framework, yielding 12 themes and 11 subthemes.¹³ The programme appeared to operate as intended through de-escalation and improved quality of care, while the data also identified thin complementary community services as a constraint. The authors interpreted this as evidence that the model may transfer beyond large urban centres.¹³

Model heterogeneity and implementation determinants

Variation between programmes is the principal obstacle to pooling results. Puntis and colleagues identified 19 distinct triage models across the 23 studies they included, differing in hours and days of operation, staffing, vehicle, and whether the team attended as a first or second response.⁴ Some models involved telephone support only, with no clinician at the scene, while others placed a clinician in a marked or unmarked police car for part of the day.⁴ Comparisons that treat co-response as a single intervention therefore risk averaging across substantially different services.

Two Australian implementation studies converge on similar determinants. Semi-structured interviews with 39 participants in 2016, comprising first responders and community service providers familiar with the Cairns Mental Health Co-Responder Project, identified team characteristics, senior and executive support, collaborative governance and co-location of the team within a mental health setting as essential elements.¹⁴ In that project the team paired a mental health nurse holding crisis intervention competencies with a specially trained police officer; perceived challenges included initial concerns about client confidentiality, absence of an evaluation plan and inadequate resourcing.¹⁴

A post hoc implementation determinant evaluation of the Brisbane co-responder units interviewed 30 police and mental health staff and their managers, coding responses against the Consolidated Framework for Implementation Research.² Those units served a health service catchment of about 1.2 million people and used a ride-along second-response design, in which a senior mental health clinician and a police officer travelled in an unmarked vehicle from mid-afternoon until midnight, seven days a week, with the closing hours of each shift given over to documentation.² Participants across all groups regarded the units as a substantial improvement on usual police management of mental health crisis cases. Enablers included a formal information-sharing agreement signed in 2017 and compatibility with existing services; barriers included organisational cultural differences, notably police risk-aversion to suicidality against a health focus on least-restrictive practice, extensive documentation requirements and limited police mental health training.²

Comparable frictions recur elsewhere, which suggests that they are structural rather than local. The JBI scoping review reported role conflict, staffing and data-sharing difficulties as persistent challenges across the co-response programmes it mapped.⁶ In the New Zealand survey, only 36% of mental health staff described good processes for accessing inter-agency support, and police and ambulance respondents reported distrust and miscommunication with mental health services.³

Service-user preference, equity and young people

Preference data complicate the case for co-response. Interviews with 50 mental health care clients with serious mental illness and a history of misdemeanour arrest, enrolled in a randomised controlled trial of a police-mental health linkage system, together with 18 of their family members and friends, asked participants to rank four crisis response options.¹⁵ Respondents selected a non-police response first and a crisis intervention team response last, citing past negative interactions with police and the value of trained responders, while also raising concerns about safety and the limitations of a response without police.¹⁵ This study did not evaluate co-response outcomes and speaks to acceptability rather than effectiveness. A related concern is that safety-driven practice may itself deter help-seeking: the realist review protocol notes reports that encouraging a person to leave home so that police powers under section 136 can be used may foster distrust and discourage later contact with services.¹⁶

Equity remains the least evidenced domain. The rapid evidence synthesis noted that health equity was scarcely discussed in the street triage literature,⁵ and a commentary on crisis services argues that deficiencies in sociodemographic data infrastructure make baseline inequities difficult to measure at all.⁹ The same commentary identifies diagnostic overshadowing, whereby diagnoses are assigned at different rates to subgroups for non-clinical reasons, as a source of confounding that may compromise control groups and outcome data in evaluations of coercive interventions.⁹

For children and young people, no completed evaluation of co-response was identified among the sources reviewed here; the available document is a protocol. A registered realist review protocol proposes to develop a programme theory of how police and mental health practitioners co-respond to young people in crisis, searching six databases to June 2024 alongside grey literature and practitioner workshops.¹⁶ Its background notes that youth mental health care plans developed by Greater Manchester Police officers attending emergency calls rose by 14.2% between 2021 and 2022, and characterises the international co-response literature on young people as predominantly Australian, Canadian, New Zealand and American.¹⁶ Any application of adult co-response findings to adolescents is therefore extrapolation, and the direction of any difference remains unresolved.

Conclusions and outlook

What appears reasonably established is narrow. Across the systematic syntheses and the scoping review, pairing officers with clinicians is associated with lower use of police custody, fewer detentions to a place of safety and fewer transports to emergency departments than police-only responses.^4–6 In the largest incident-level dataset, co-response clinicians achieved higher diversion rates than crisis intervention team-trained officers.⁷ Qualitative evaluations consistently describe improved interagency collaboration and staff endorsement, though such accounts may favour the views of programme advocates.^2,13

What remains open is broader. No randomised trial of co-response has been identified in the reviews covering literature to 2020.^1,4 Effects on arrest measured over a year have been null in the one multi-site weighted comparison available, in which only mobile crisis reached significance.⁸ Long-term cost-effectiveness is largely unevaluated,⁶ equity-stratified outcomes are rarely reported,^5,9 and the people most affected rank police-involving models below non-police alternatives.¹⁵

Several design features would move the field. Evaluations with concurrent comparators, ideally contrasting co-response against specialist officer training and against non-police models in the same jurisdictions, would address the confounding that pre-post designs cannot.^5,8 Standardised reporting of model characteristics, including hours, staffing and whether the team responds first or second, would make pooling meaningful given the 19 model variants already documented.⁴ Future studies should measure outcomes beyond detention counts, including linkage to ongoing care, symptom change, cost across police, health and social services, and disparities between subgroups.^5,9