Social Science & Medicine published an RCT of an Indianapolis program that partners police officers with mental health clinicians in responding to behavioral health emergencies. Quick take: The study findings (no significant impacts) are unreliable due to study limitations.
Program & Study Design:
The study randomized 686 behavioral health 911 calls to receive (1) co-response from police officer + clinician (treatment) vs (2) police-as-usual (control). The program aimed to divert persons in crisis from arrest & hospitalization, & connect them with needed services.
Findings:
Over a 1y follow-up, the study found none of the hope-for impacts (e.g., no discernible decrease in subsequent EMS/911 calls, emergency department visits or jail bookings; no increase in behavioral health treatment). But I think these results aren't reliable due to study limitations.
The most important limitation was that the researchers dropped from the sample those 911 calls that didn't receive the assigned intervention (i.e., co-response vs police-as-usual). This led to the loss of 30% of calls in the T group & 14% of calls in the C group.
This problem (an "intent-to-treat" violation) can easily undermine T & C group equivalence, & lead to inaccurate results. As an illustrative example, suppose mental health clinicians chose (due to staffing limitations) not to respond to T group calls they deemed less serious.
Dropping these no-response (less serious) cases from the T group would distill the T group down to more serious cases, potentially leading to an underestimate of program impacts. Other study limitations included a sample that was likely too small to detect meaningful impacts.
Comment:
The study is valuable in illustrating how RCT designs might be used to evaluate co-response programs. But unfortunately I think it fell short in execution. Disclosure: My former employer, Arnold Ventures, funded this study.