Art Therapy for Psychosis: The MATISSE Trial and Why Services Still Offer It

Group art therapy for people with established schizophrenia has been tested in a large UK randomised controlled trial, and the trial found no difference between group art therapy, an activity group and standard care alone on either of its two primary outcomes at 24 months. That is a genuine null result rather than a gap in the literature, it is narrower than the way it is usually quoted, and it did not stop services from offering group art therapy, for reasons that are worth understanding rather than sneering at.

I have never been in an art therapy group, on either side of the room, and I am not going to write as though I have. I am a community arts practitioner who runs singing sessions in care homes and on a stroke unit, and what I know about this trial I know from reading it. What I do have is the reason I read it, which is the reason this page exists. At a regional creative health event a couple of years ago, a speaker told a room of about forty people that the arts therapies had been disproved by a large trial, and moved on to his next slide within about eight seconds. I caught him at the tea break and asked which trial. He said the big Cochrane review. It is not a Cochrane review. It is a single trial, reported in the BMJ in 2012 with the full report published in Health Technology Assessment in the same year, and he had not read either. I went home and read both, and found that almost every element of the sentence he had used it for was wrong: the trial tested group art therapy and not the arts therapies plural, so it says nothing about music therapy, dramatherapy, or individual art therapy; it tested one diagnosis; it measured two primary outcomes at one timepoint; and disproved is the wrong verb for a finding of no difference. I also found the thing he had not mentioned, the attendance figure below, which cuts against his case rather than for it, and I noticed in myself an immediate urge to lead with it as a rescue. That urge is the mirror image of his mistake and this page tries to resist both.

Every claim about effect on this site carries a grade. Tier 1 means supported, Tier 2 means promising but limited, and Tier 3 means one of two very different things: either nobody has run an adequately powered controlled study, or somebody has and the result was null. The page always says which. That distinction is the whole of this article, and the method behind it is set out on reading arts in health research.

What is offered, and why it was there in the first place

Group art therapy has been part of what UK psychosis services offer for decades, delivered by registered art therapists inside secondary care mental health services rather than as an add on.

Art therapist and art psychotherapist are two of the four arts therapy titles protected in UK law, alongside dramatherapist and music therapist, and in July 2026 the Health and Care Professions Council register listed 6,103 arts therapists across all four¹. A meaningful proportion of that small workforce has always worked in adult mental health. The professional body for the visual side publishes practice standards for the work², and the trial described below was delivered to nationally agreed standards rather than to whatever each site felt like doing, which is one of the reasons it is a useful trial.

The clinical rationale offered for it is not mysterious. People with psychosis may find sustained talking therapy difficult, particularly where negative symptoms, disorganised thought or a long history of being interviewed by professionals are in play. An image gives something else to attend to, allows a person to be in a room with others without having to speak continuously, and can hold material that is not yet sayable. That is a plausible mechanism and plausibility is not evidence, which is precisely why a trial was funded. What a session actually involves is described on what happens in an art therapy session, and the ward context on arts in mental health inpatient units.

The trial, in full

MATISSE was a three arm, rater blinded, pragmatic randomised controlled trial, and it is set out here in enough detail that a reader could go and check it.

Recruitment was from secondary care mental health and social services in four UK centres across 15 sites. 417 people were randomised: 140 to group art therapy, 140 to activity groups, and 137 to standard care alone. 355 (85%) were followed up at two years, which is a high retention rate for this population and this length of follow up.

The middle arm is the design feature that makes the trial unusual and worth this much space. The activity groups were an active comparator: an organised group activity without an art therapist, included so that the trial could distinguish the specific contribution of art therapy from the general effect of being invited somewhere weekly with other people. A great many trials in this field compare an intervention against nothing much, and consequently cannot separate the two.

The intervention was weekly group art therapy in addition to standard care for 12 months, in groups of up to eight people, 90 minutes at a time, delivered by registered art therapists to nationally agreed standards. The trial specified two outcomes as primary, both measured at 24 months: global functioning, measured on the GAF, and mental health symptoms, measured on the PANSS. Measuring at 24 months means the trial was asking about the position a year after the groups had finished, not only at the end of them.

What it found

No difference between the three arms on either primary outcome. That is the headline, and the numbers behind it are these.

Comparison at 24 months	Adjusted mean difference	95% CI
Art therapy against standard care, GAF	-0.9	-3.8 to 2.1
Art therapy against standard care, PANSS	0.7	-3.1 to 4.6

Both intervals sit across zero, and both are compatible with a small benefit, a small harm and nothing at all³. On the GAF the point estimate is very slightly worse than standard care and on the PANSS very slightly worse again, and neither point estimate should be read as a direction of travel: the honest reading of both rows is no detectable difference.

Three further findings belong beside them. No secondary outcome favoured art therapy. The one secondary difference the trial did find ran the other way: those referred to an activity group had fewer positive symptoms at 24 months than those randomised to art therapy⁴. And a secondary analysis found that attendance at art therapy groups was not associated with improvement in global functioning or mental health, which matters because the most natural defence of a null result in a poorly attended trial is that the people who came must have benefited. In this trial, they did not detectably do so.

On cost, the economic analysis found that total costs in the art therapy arm were lower than in the two comparison arms, and that referral to group art therapy was nonetheless judged not to be a cost effective use of resources⁴. Cheaper without being better is not value, and that pairing is misreported often enough to be worth stating explicitly. The way cost claims go wrong in this sector generally is worked through on does arts on prescription save money.

One caveat applies to every figure above and to every group result anywhere. These are differences between group averages. Inside a trial reporting no average difference, some individuals will have improved considerably and some will have deteriorated, and nothing in an adjusted mean difference tells you which one a particular person would have been.

The attendance figure, and what it does and does not license

Only 86 of the 140 people randomised to art therapy (61%) attended even one group, and 73 of 140 (52%) in the activity arm. This is the most important limitation of the trial and the most abused fact about it.

Start with what it genuinely establishes. Because the trial was pragmatic and analysed by intention to treat, what it tested was the effect of referring somebody to 12 months of weekly group art therapy, not the effect of a course of group art therapy actually received by somebody who turned up. Those are different questions. A service deciding whether to commission a group art therapy programme is asking the first question, so for that decision the trial answers exactly the right thing. A person sitting in a group every week for a year is asking the second, and the trial does not answer that as cleanly.

Now what it does not license. It is not a reason to set the result aside, and I want to be blunt about that because it is the standard move. Three things stand in the way. First, low uptake is itself a finding about the intervention as it can be delivered in ordinary services, not an accident that a better trial would have avoided; an intervention most people do not attend is not the same intervention as one they do. Second, the secondary analysis specifically looked at whether attendance was associated with improvement and found that it was not. Third, the activity arm had lower attendance still, at 52%, and the trial still detected a secondary difference in that arm’s favour.

The reason to be this careful is that both sides misuse the same number. Advocates use it to argue that the trial did not really test art therapy, so the null can be discounted. Sceptics ignore it entirely and quote the result as though it were a clean efficacy trial of therapy received. Neither is honest. The accurate sentence is longer and less useful in an argument: referral to group art therapy, in ordinary UK secondary care, with the uptake such services actually achieve, produced no benefit on global functioning or symptoms at two years.

How to grade it

This is Tier 3, and specifically the null controlled evidence kind rather than the nobody has looked kind. Those words are chosen deliberately and the distinction carries almost all of the meaning.

Tier 3 on this site covers two situations that are frequently collapsed into one. The first is a topic where no adequately powered controlled study exists, which describes most of what happens in care homes, community halls and day services, including my own work. The second is a topic where a controlled study does exist and came out null. The label is the same and the epistemic position is entirely different. An empty file means the question is open. A well conducted trial reporting no difference is a positive finding of absence on the outcomes it measured, and it does not become an open question again just because somebody would prefer it to be one.

So when this page says Tier 3, it means the second thing. It does not mean nobody has bothered to look, and it does not mean the finding is soft. Equally, and this is where a Tier 3 label gets misread in the other direction, it is not a verdict of does not work in general. It is a verdict of no detectable difference, on these outcomes, in this population, with this format, at this timepoint. The way a null result should be read alongside its interval and its certainty is set out on reading arts in health research, and the way policy documents tend to handle findings like this one is on creative health policy and reports.

What the trial does not cover

Everything outside the four corners of its design, and the list is longer than the trial.

It does not cover individual art therapy, which is a different intervention with a different relationship at its centre, and which was not tested here at all. It does not cover other diagnoses: this was people with established schizophrenia, so it is not evidence about art therapy in depression, in trauma, in eating disorders, in dementia, in cancer care or with children. It does not cover first episode psychosis, where the population, the prognosis and the service model are all different. It does not cover outcomes the trial did not measure, and there are many that people reasonably care about, including engagement with services, self reported quality of the experience, or anything a person might value that is not captured by the GAF or the PANSS. And it does not cover any timeframe beyond 24 months.

The related pages set out separately what is and is not known elsewhere: art therapy for trauma and PTSD, and for a different regulated modality working in the same services, dramatherapy in mental health services. The pillar, art therapy, deliberately quotes no effect size at all, for reasons it explains.

Why services still offer it

Because a null result on two outcomes is not the only input into a service decision, and the reasons that remain are more respectable than the sceptical version allows.

The first is patient preference. Some people ask for it, prefer it to the alternatives offered, and keep attending. A trial that finds no difference on the GAF and the PANSS at 24 months has not found that people were harmed, that they disliked it, or that the time was unpleasant. Preference is a legitimate consideration in a health service and it does not require an effect size to be respectable.

The second is what else is actually available. On many wards and in many community teams the realistic alternative to a group art therapy session is not a better evidenced psychological intervention. It is an empty afternoon. That comparison is uncomfortable and it is the true one in a great many places, and it is a reason to fund something for people to do rather than a reason to call that something a treatment.

The third is institutional. Provision that predates a trial does not evaporate when the trial reports. Posts exist, rooms are booked, referral pathways are printed, and people are mid course. Inertia is not a justification, and it is an honest description of how services change, which is slowly.

The fourth is a distinction clinicians draw and researchers are wary of: the outcomes a trial selects are not the whole of why something is offered. That argument can be used to make any intervention unfalsifiable, and it should be treated with suspicion for exactly that reason. It is also not always wrong. The reasonable middle is that a service offering group art therapy after MATISSE should be able to say which of these four reasons is operating, and should not describe the offer to a patient as a treatment expected to improve functioning or symptoms, because the best available evidence says it will not.

If you are the person being offered it

Take the offer on its own terms, ask the questions that a null trial makes sharper, and do not treat any of this as a reason to change anything else.

Worth asking: what is this expected to help with, in plain words. How many sessions, over what period, and in a group of what size. Who runs it, and are they registered, which you can check yourself in a minute on the statutory register of 6,103 arts therapists¹. What else is available if you decline. What happens if you go twice and hate it. And, if you want the honest version of the evidence question, whether the person offering it knows about MATISSE, because a service that does is a service thinking about what it provides.

None of this argues for refusing. If you want it, the trial gives you no reason not to have it, and the difference between an unregulated painting group and a clinical intervention delivered by a registered therapist is still real, and is set out on art therapy against an art class. If you are considering the profession from the other side, training as an art therapist covers the route.

The two cautions I would leave here are the ones this page has been built around. A group average is not a prediction for a person: no adjusted mean difference of -0.9 tells you what would happen to you. And the error runs in both directions, with the sceptical version rarer in this field but no less wrong, so a null on two outcomes in one population is not evidence that the arts therapies have been disproved, whatever a speaker with a slide deck says at the tea break. Nothing here is a reason to alter, delay or decline any other part of anybody’s treatment, and no arts intervention of any kind stands in for the care around it.