Does involving family in mental health care stop relapse?

No. Adults who received family-inclusive Open Dialogue care relapsed at almost exactly the same rate as adults who received standard NHS care. In this trial of 494 adults across England, the Open Dialogue group had about a 5% lower risk of relapsing after they recovered (hazard ratio 0.95), a gap far too small to mean anything.

The uncertainty around that number says more than the number itself. The true effect was very likely somewhere between about 33% lower risk and about 32% higher risk (95% CI 0.67 to 1.32). When the range runs that far in both directions, the honest reading is that the two approaches performed the same on the thing the trial was designed to measure. What separated them was everything else.

What is Open Dialogue?

Open Dialogue is a way of organizing mental health care rather than a single therapy technique. The trial describes it as a person-centred, transdiagnostic model that emphasises continuity, therapeutic relationships, and collaboration with the person’s social network. In practice, care runs through network meetings that include the service user, members of their social network such as family or close friends, and usually two practitioners who support that network for the whole duration of care.

That is a real departure from how most public mental health systems run. The comparison arm in this trial was treatment as usual, which the researchers define as the functional team model currently used throughout English mental health services. In that model, different teams handle different phases of care, so the faces around a person change as they move through the system.

What the data show

Researchers randomised at the level of primary care practices, not individual people, which is how you test a change to an entire service rather than a change to one patient’s treatment. Eighty practices formed 32 clusters, split evenly between the two arms. One mental health trust pulled out with its two clusters, leaving five NHS trusts and 30 clusters in the final trial. Between June 25, 2019, and December 9, 2021, 494 adults consented, 223 in the usual-care group and 271 in the Open Dialogue group. Their average age was 38.1 years, 266 (54%) were female, and 341 (69%) were White British.

Everyone was followed for two years. Of those enrolled, 174 (78%) in the usual-care group and 225 (83%) in the Open Dialogue group recovered from the crisis that brought them in and had enough data to determine whether they later relapsed. On that primary outcome, time to first relapse, there was no significant difference between the groups.

The secondary outcomes went the other way. Open Dialogue was associated with significantly lower probabilities of psychiatric inpatient admission and of re-referral to crisis care or secondary mental health services. People in that arm also reported better self-rated recovery, better health-related quality of life, and higher satisfaction with services. One result runs against the model’s own logic: there were no significant differences in the quality or size of people’s social networks, even though building around those networks is the whole premise.

Safety looked reassuring. There were 386 serious adverse events across the trial, 281 in the usual-care group and 105 in the Open Dialogue group, and 376 of them (97%) were judged unrelated to the intervention. The Open Dialogue arm recorded fewer of these events despite being the larger group.

Dr. Kumar’s Take

A trial that misses its primary endpoint usually gets filed away as a failure, and that would be the wrong lesson here. Time to relapse is a clean, countable outcome, which is exactly why trials pick it. It is not obviously the outcome that matters most to someone in crisis. Staying out of a psychiatric hospital, not being bounced back to crisis services, and feeling like your care was worth having are not consolation prizes.

The caution is that these are secondary outcomes in a trial that did not hit its primary one, so they are a strong reason to keep investigating rather than proof of superiority. Participants and clinicians also knew which model they were getting, and satisfaction and self-rated recovery are self-reported. A model built on being listened to has a natural advantage on questionnaires about whether you felt listened to. The inpatient admission and re-referral results come from electronic health records instead, which is harder to argue with.

The finding I find hardest to explain is the social network result. Open Dialogue is built on the idea that mental health improves when the people around you are pulled into care. Two years in, the networks were no bigger and no better rated. Something in this model is working, and it may not be the mechanism its founders proposed.

What this does not settle

The trial recruited from June 2019 through December 2021, so a large part of enrolment and follow-up overlapped with the pandemic and the disruption it caused to mental health services in England. The trial was also registered retrospectively, meaning after it began. Cluster randomisation left uneven groups, 223 versus 271, which is common when you randomise practices instead of people but still means the arms were not identical in size. And this was five NHS trusts in London and the South of England, so a system with different staffing, different funding, and different community services might not reproduce the result.

Practical Takeaways

If you or a family member is in mental health crisis care, ask whether relatives or a trusted friend can join appointments, since network involvement was the defining feature of the model that reduced hospital admissions here.
Ask whether you will see the same clinicians throughout your care, because continuity of practitioners is one of the specific things this model changed relative to standard team-based care.
Treat the hospital-admission and satisfaction findings as promising rather than settled, since they were secondary outcomes in a trial that found no difference on its main measure.
Do not read this trial as a reason to change or stop any medication, since it tested how services are organized, not which treatments people received.

FAQs

If relapse rates were the same, why does the hospital finding matter?

Relapse and hospitalization are not the same event. A person can have symptoms return and still be managed at home, in clinic, or through a crisis team without ever being admitted to a psychiatric ward. An inpatient admission is disruptive in a way a symptom flare is not, since it removes someone from their job, their home, and their family for days or weeks. It is also the single most expensive part of mental health care, which is why the authors framed the result as reduced acute inpatient bed use rather than as a symptom benefit.

Was Open Dialogue safe?

The trial found no significant safety concerns. There were 386 serious adverse events in total, and 376 of them, 97%, were judged unrelated to the intervention people received. That last figure matters more than the raw count, because trials in people presenting in mental health crisis will record serious events regardless of what treatment arm someone is in. The events were split 281 in usual care and 105 in Open Dialogue, and the Open Dialogue arm was the larger of the two groups.

No, and this is the most surprising result in the trial. Researchers measured social support using the Social Provisions Scale and network size using the Lubben Social Network Scale, and found no significant differences between the two arms after two years. So the meetings did not visibly expand or strengthen the networks the model is built around. Whatever produced the lower admission rates and higher satisfaction, it was not a measurable change in who was in people’s lives.

How strong is this evidence compared with a typical drug trial?

It is a different kind of evidence rather than a weaker one. Cluster-randomised trials are the right design for testing a whole service model, since you cannot deliver one philosophy of care to half the people in a single team without contaminating the other half. The trade-off is that nobody can be blinded to which model they are receiving. The researchers handled this by masking the chief investigator, the senior statistician, and the assessors of the primary outcome, and by pulling clinical outcomes from electronic health records rather than relying only on what participants reported.

Bottom Line

Open Dialogue did not delay relapse in adults presenting to English mental health services in crisis, and on the trial’s own primary measure it performed the same as usual care. On almost every other measure it did better: fewer psychiatric admissions, fewer re-referrals to crisis services, better self-rated recovery, better quality of life, higher satisfaction, and no safety signal. That combination is unusual and awkward. It suggests that how mental health services are organized changes what happens to people, even when it does not change the symptom timeline that trials are built to track.

Read the full study