When a call reaches 911 about a person in a behavioral health crisis, the most consequential public safety decision may occur before any responder arrives. A call-taker and dispatcher must determine whether the situation requires police, emergency medical care, a civilian crisis team, a clinician paired with an officer, or a combination of these that may change as the risk becomes clearer. If the only reliably available option is a patrol car, then police become the default not because every call requires law enforcement, but because the response system has not built enough alternatives.

A major 2026 Urban Institute evaluation of Denver’s Support Team Assisted Response program and its police–clinician Co-Responder Program gives cities a better way to frame this debate. The findings do not prove that civilian responders should replace police across the board. They show that communities benefit when emergency systems maintain multiple response pathways and match them to the circumstances. The public question is no longer whether every crisis requires police involvement or whether police should be removed from crisis response. It is whether government can send the right team, at the right level of risk, with enough capacity and follow-up to change what happens next.

Denver built two different tools for two different kinds of calls

Denver’s STAR program sends an unarmed clinical van team, typically a paramedic or emergency medical technician working with a behavioral health professional, to eligible calls in place of police. Responders assess needs, de-escalate the immediate situation, and can refer people to a network of community partners for continuing assistance. The program began as a downtown pilot in 2020 and expanded citywide in 2022.

The Co-Responder Program operates differently. It pairs mental health clinicians with police officers in situations where law enforcement presence remains necessary or where risk is less certain. The clinician brings assessment, de-escalation, and service navigation skills; the officer retains responsibility for immediate safety and lawful enforcement. These models overlap, but they are not interchangeable. One expands the range of calls that can be handled without police. The other improves the quality of calls where police involvement is appropriate.

That distinction matters because “alternative response” is too often discussed as if cities must choose a single universal model. They should not. A person quietly disoriented at a transit station presents a different response problem from a person making credible threats with a weapon. A welfare check, a suicide attempt, an intoxication call, a trespassing complaint, and a violent disturbance may all contain behavioral-health elements, but they do not create identical safety conditions. Mature systems build a continuum rather than forcing disparate incidents into the same vehicle.

The new evidence is encouraging

The Urban Institute examined Denver data from June 2020 through September 2024. During that period, STAR and Co-Responder staff each handled more than 20,000 calls for service. Denver’s 911 system identified nearly 50,000 calls as potentially eligible for STAR, but limited capacity meant police responded to 57 percent of those calls, either alone or with another public-safety agency. STAR vans responded without police to 21 percent and alongside police to an additional 6 percent. In total, STAR vans handled 23,114 calls originating through 911.

Using matched comparison groups and regression analysis, the researchers found that people who received a first documented STAR clinical encounter were less likely than comparable people who received a police response to have any police contact or arrest in the following year. The estimated effects were two to three times larger among people with prior experiences of homelessness. People receiving co-responder clinical encounters were also less likely to be arrested in the following year, with reductions in police contacts and bookings. Both groups were more likely to have later clinical encounters, suggesting that subsequent crises were more often routed back toward services rather than exclusively toward conventional police response.

These are important results, but they are not a license for promotional certainty. The study was not a randomized trial. Data limitations meant some people in the police comparison groups may not have been eligible for STAR or directly comparable to those receiving clinical responses. The analysis included only documented clinical encounters rather than all contacts, and the researchers could not assess housing, health, employment, referral completion, or service retention. The City and County of Denver and the Caring for Denver Foundation funded the evaluation, although the report states that funders did not determine the findings.

The responsible conclusion is therefore substantial but bounded: Denver has credible evidence that differentiated crisis response is associated with reduced criminal justice involvement, particularly for people already cycling through public systems. The next stage is not declaring victory. It is improving data, measuring health and housing outcomes, testing service follow-through, and ensuring enough teams are available so dispatch decisions reflect need rather than scarcity.

Capacity is not an administrative detail; it determines who responds

Denver’s experience exposes the gap between creating an alternative and making it operationally real. In 2024, dispatchers designated roughly 1,000 to 1,500 calls each month as STAR-eligible. Yet most eligible calls during the study period still received a police response. That does not necessarily mean dispatchers rejected the civilian model or officers captured work that belonged elsewhere. It often means no STAR unit was available when the call required a response.

The Urban Institute estimated that nine or ten concurrent STAR vans per day, averaging approximately five responses per unit, could meet the current level of identified demand. In 2023, the full program cost about $4.4 million, including $532,000 in capital spending for vans. Researchers estimated approximately $237 per van response and $470 per response when the community-partner network and other program components were included. STAR encounters were more resource-intensive than a basic police contact because they delivered clinical assessment, de-escalation, referrals, and follow-up services that a conventional police contact is not structured to provide. Some costs were offset by reductions in later arrests and bookings, but the strongest case for STAR is not that care is automatically cheaper. It is that a more appropriate response may produce better safety and human outcomes.

Cities should stop announcing small pilots without publishing a capacity plan. A civilian crisis program operating only during limited hours, in a small geographic area, or with too few teams may generate good individual encounters while leaving the overall dispatch pattern largely unchanged. The meaningful measure is not whether a city owns a crisis van. It is the percentage of eligible calls for which the intended responder was actually available.

Dispatch is the brain of a diversified safety system

A city can hire excellent clinicians and still fail if its call-taking and dispatch protocols cannot reliably identify appropriate calls. Callers may use ambiguous language, describe symptoms rather than risk, or demand police because they do not know another response exists. Dispatchers must make decisions with incomplete information and be able to upgrade or downgrade a response as new facts emerge.

The solution is structured triage supported by shared training, clear eligibility rules, supervisory review, and real-time communication among 911, 988, police, fire, emergency medical services, and civilian crisis teams. The federal 911 program explains that call centers use protocols to determine the right responders and location. SAMHSA’s current crisis-care framework adds a complementary continuum: someone to contact, someone to respond, and a safe place for help. Community safety depends on connecting those systems rather than building 911 and 988 as parallel networks that transfer responsibility back and forth.

Every jurisdiction should define at least three response levels. Low-risk, noncriminal calls should be eligible for civilian mobile response. Calls with uncertain safety conditions or a documented need for officer presence should receive a co-response. Calls involving weapons, active violence, or immediate threats should receive a police-led response with clinical support available when the scene is secure. These categories need room for professional judgment, but the default should be based on assessed risk rather than institutional tradition.

A responder is not a crisis system

The most important limitation in the Denver evaluation may be the absence of robust information about what happened after referral. A compassionate encounter can prevent an unnecessary arrest, but it does not by itself secure housing, stabilize medication, provide detoxification, repair family relationships, or ensure that someone can obtain ongoing care. If mobile teams repeatedly return to the same person without access to treatment, housing, transportation, benefits, and a safe place for stabilization, the community has created a more humane revolving door rather than a complete system.

Denver’s model recognizes this through its Community Partner Network and facilities such as the city’s Assessment, Intake, and Diversion Center, which accepts voluntary referrals from STAR and other outreach teams. Other cities should build the same connective tissue before scaling response volume. Mobile teams need warm handoffs, same-day appointments, transportation, data-sharing rules, and authority to solve practical barriers. Follow-up should begin before the team leaves the scene and continue long enough to determine whether the referral became a service.

Community safety needs a public crisis-response dashboard

Cities should report more than the number of calls handled. A useful dashboard would show the number of potentially eligible calls, the number of responses actually dispatched, reasons the preferred team was unavailable, response times, police backup requests, injuries, transports, involuntary interventions, arrests, repeat contacts, referral completion, housing and health outcomes, and costs across the full response continuum. Results should be disaggregated by neighborhood and demographic group without exposing individual clients.

Equity belongs in that dashboard. Urban’s Denver analysis found that Hispanic and Latino residents appeared significantly underrepresented among STAR and co-responder clients compared with the city population. The study could not determine why. Possible explanations include language access, geographic deployment, call-taking practices, community knowledge, or differences in who contacts emergency systems. The finding should trigger investigation rather than assumption. An alternative that is not equally reachable can reproduce disparities even when each individual encounter is well handled.

Safety metrics should protect both responders and clients. Programs must document when civilian teams request police assistance, why scenes are reclassified, and whether eligibility rules exclude calls that could be safely diverted. Transparency can prevent two opposite failures: sending unarmed workers into conditions beyond their mandate and using hypothetical danger to keep nearly every call inside conventional policing.

The goal is the right response, not an ideological victory

Denver’s evidence points toward a practical community-safety settlement. Police remain essential for violence, credible threats, investigations, and enforcement. Civilian clinicians and medical personnel are often better suited for low-risk behavioral health and welfare calls. Co-responder teams occupy the important ground between those situations. Dispatchers need the training and authority to choose among them, and every pathway requires enough staffing to be available when needed.

This approach should appeal across political lines. It reserves sworn officers for work that requires their legal authority and safety capabilities. It reduces unnecessary criminal-justice contact. It gives residents in crisis a response that more closely matches their needs. It also replaces slogans with an operational question that the public can measure: when someone calls for help, did the system send the right team—and did that response create a safer next day, not merely a quieter next hour?

Selected sources

• Urban Institute: Aligning Crisis Response with Community Needs
• Urban Institute: Full 2026 Denver STAR and Co-Responder evaluation
• SAMHSA: National Behavioral Health Crisis Care Guidance
• SAMHSA: 2025 National Guidelines for a Behavioral Health Coordinated System of Crisis Care
• SAMHSA: 988 and mobile crisis response frequently asked questions
• National 911 Program: Frequently asked questions about 911
• City and County of Denver: Assessment, Intake, and Diversion Center