Creative Health Policy and Reports: What They Are For, and Why They Are Not Evidence

A creative health policy report is an argument assembled from evidence, and a systematic review is an attempt to measure something. Both are legitimate documents, both are useful, and only one of them can support a claim that an activity works. Almost every misquoted figure in this field starts with the two being read as the same kind of thing.

I have been a case study in one of these reports. A scheme I worked on was written up over two pages, with a photograph and a quotation from me, and the quotation was accurate. What was not accurate, though nobody wrote it down as a claim, was the impression the spread created. The scheme it described had run for eighteen months and closed nine months before publication, when its grant ended. The report was not lying; it was describing something that had happened. But a reader would have finished those two pages believing the thing was still running, and would have had no way of knowing that the most interesting fact about it was how it ended. I have thought about that ever since when reading anybody’s case studies, including the ones I now find persuasive.

Where a claim about effect appears below it carries a tier (Tier 1 supported, Tier 2 promising but limited, Tier 3 practice and anecdote), explained on reading arts in health research.

The two kinds of document

They have different purposes, different authors and different failure modes, and telling them apart takes about thirty seconds.

A policy or sector report exists to make a case: that something should be funded, commissioned, embedded or scaled. It draws on studies, evaluations, case studies, interviews and costings, and it is written by people who already believe the conclusion, which is not a criticism, because that is what advocacy documents are. Its failure mode is selection: the strongest available figure, the scheme that worked, the caveat left in a footnote.

A systematic review exists to answer a question: does this intervention change this outcome, and how confident can anybody be about the answer. It has a protocol written before the searching starts, states its inclusion criteria, reports what it found including the parts that do not help, and attaches a certainty rating. Its failure mode is different and more technical: pooling studies that should not be pooled, or a search that missed things.

The two get confused because reports cite reviews and look, at a glance, like the same genre. The tell is the conclusion. A report concludes that something should happen. A review concludes that an effect is of a certain size, with an interval around it, at a stated level of certainty.

The citation that outruns every correction

The most repeated false claim in this field is that the National Institute for Health and Care Excellence recommends music therapy in dementia. It does not.

A full text search of the dementia guideline NG97, Dementia: assessment, management and support for people living with dementia and their carers, returns zero occurrences of music, art, dance, drama or creative activity¹. Not a weak recommendation, not a mention in passing. Nothing.

The claim nonetheless appears in funding bids, service descriptions, training materials and press coverage, and almost everybody repeating it is sincere. It has been passed along enough times that checking it stopped seeming necessary, which is the exact mechanism this whole page is about.

Two things have to be said next to that correction, because a bare absence is easy to misuse in the other direction. Absence from a guideline is not a finding that something is ineffective. Guidelines cover what they cover, and NG97 has its own scope and its own reasons for it. And separately, there is real evidence about music in dementia, just not from NICE: the 2025 Cochrane review covers 30 studies and 1,720 participants randomised, and reports depressive symptoms at SMD -0.23 (95% CI -0.42 to -0.04) at moderate certainty, which is Tier 1 and small, alongside agitation and aggression at SMD -0.05 (95% CI -0.27 to 0.17) at moderate certainty, which is Tier 1, null result. It found no evidence of anything persisting four weeks after the sessions end and contains no separate quality of life estimate². Anything citing 22 studies and 1,097 participants is citing the superseded 2018 version. See music therapy for dementia and creative activity in dementia care.

How a nested count becomes a bigger field

The WHO Europe scoping review is the document most often used to establish that the evidence base is enormous, and reading its counts correctly shrinks it considerably.

It covers over 900 publications, which comprise 200 plus reviews and 700 plus individual studies. Those reviews between them cover over 3,000 studies³. The 3,000 sit inside the 900 rather than beside them: they are the studies covered by the reviews that are among the publications.

Written as “3,000 studies and 900 publications”, the same document appears several times larger than it is. That formulation is common and this site does not use it.

More important than the arithmetic is the genre. A scoping review maps a literature rather than pooling it. It does not compute an effect, it does not attach a certainty rating, and it cannot tell you whether anything works. It carries no tier here, and it is cited only for what it is: evidence that a field has been studied. That is a real and useful thing to know, and it is a different statement from evidence that anything in the field is effective.

The savings figures, as a worked example of a report chain

The clearest demonstration of what happens between a source document and a policy claim.

The source is a cost benefit summary of one Gloucestershire scheme, written by a GP in December 2011, which analysed 90 patients out of roughly 500 referred, had no control group, called itself “a simple observational study”, and stated in its own conclusion that it “does not imply causality”⁴. Three figures escaped it and none of them survived the journey intact.

• The 37% fall in GP consultations is the months 7 to 12 figure. The report’s own full year figure, printed in the same section, is 24%.
• The 27% is a reduction in overall NHS spend, £157,473 before against £115,050 after. It is quoted almost universally as a reduction in admissions, which appear separately as 54 before and 33 after with no percentage and no significance test attached.
• The per patient saving subtracts £360, which is £180,000 divided by the 500 patients referred, from £576, which contains £471 derived by dividing £42,423 by the 90 patients analysed. Numerator and denominator describe different populations.

None of that required anybody to lie. It required a report to quote a figure, and a second report to quote the first, and a bid to quote the second. The full arithmetic is on does arts on prescription save money.

The wider context that keeps the whole cluster at Tier 3 is that there is nothing better: the 2024 systematic review screened 7,805 records, included 25 studies, and found no randomised controlled trials at all⁵.

Six checks for any report

They take about ten minutes and settle most arguments faster than debating the conclusion does.

1. Who wrote it and what are they asking for? Not a disqualification, just context. Advocacy documents argue, and knowing what is being argued for tells you where to look for selection.
2. Follow the footnote to a study, not to another report. If the trail runs report to report to report, the figure has been forwarded rather than checked.
3. What design produced the number? Randomised, controlled but not randomised, or before and after with nothing to compare against.
4. Are the counts nested or parallel? Studies inside reviews inside a publication total is the classic case.
5. Is this the current version? Reviews get updated and conclusions change. Check the identifier and the date on the record, not the date on the thing quoting it.
6. Does the conclusion match the genre? A document that concludes something should be funded has argued; a document that concludes an effect is of a certain size has measured.

A seventh, for figures about scale rather than effect: check that a number belongs to the programme it is attached to. The 300 plus organisations quoted for an American pilot belong to a later statewide programme, and the pilot was 8 and then 12: see arts on prescription around the world.

What reports are genuinely good for

Reading them as descriptions rather than as evidence turns them from a hazard into a resource.

They tell you what exists and where, which no trial will. They tell you how something was organised: who referred, how groups were formed, what the venue arrangements were, who paid. They tell you what practitioners and participants say, in their own words, which is worth having even though it cannot establish causation. They surface problems the sector is willing to admit to, which is a slower but real signal. And read several in sequence, they tell you something none of them states, which is how many named schemes have quietly closed since publication.

The main UK collections are the sector’s own body for creative health⁶ and the national body for social prescribing⁷. Both are worth reading and both are descriptions of practice. Neither is an affiliation of this site and neither endorses anything here.

Prevalence figures, and the same discipline applied

One last example, because prevalence numbers drift in exactly the same way as effect sizes.

The current WHO dementia fact sheet, updated 3 July 2026, gives 57 million people worldwide living with dementia in 2021, and carries no 2050 projection⁸. Projections do exist and they belong to specific documents: WHO’s own 2021 figure of 139 million by 2050, and the Global Burden of Disease study’s 2019 figure of 152.8 million. Attributing either to the current fact sheet is wrong, and quoting a projection without naming which one it is and when it was made is the same failure as quoting an effect size without its interval.

The habit that prevents all of this is the same one. Trace the number to the document that produced it, quote it in the units that document used, and print its design and its date next to it. That takes ten minutes per claim, which is longer than forwarding it and considerably shorter than defending it later. Nothing on this page is a reason to change, delay or decline any part of anybody’s treatment, and no arts activity substitutes for one.