Music Before Surgery: An Unusually Clean Result on an Unusually Soft Outcome

Music offered to people in the period before an operation reduces how anxious they report feeling, and the review behind that claim is unusually clean for an outcome this soft: 26 trials, 2,051 participants, and anxiety 5.72 STAI-S units lower. The complications are not in the result. They are in what the intervention actually was, in a certainty rating that does not exist, and in how narrow the outcome is compared with what gets claimed around it.

I have no first hand experience of this one and I am not going to manufacture any. I have never worked in a theatre suite, a preoperative area or an anaesthetic room, and the closest I get to surgery is meeting people weeks later on a rehabilitation ward, by which point the operation is a thing that happened to them rather than a thing that is about to. Everything below comes from the review itself, from the reviewer who checks the figures on this site, and from what the trials describe themselves as having done.

The nearest thing I have to an observation is second hand and I will label it as exactly that. A nurse on the unit I visit told me that what patients ask her for most often, when they are anxious, is their own headphones. That is one experienced person’s impression of one ward, gathered over a conversation rather than a study. It is not evidence, it does not appear anywhere in the figures below, and I mention it only because it points at something the trials do not measure: the difference between music being given to you and music being yours.

Everything quantified here carries a tier, per outcome rather than per activity: Tier 1 supported, Tier 2 promising but limited, Tier 3 practice and anecdote. What a claim must show to earn each label is on reading arts in health research.

The finding

The Cochrane review of music interventions for preoperative anxiety, CD006908.pub2, published in 2013, pooled 26 trials and 2,051 participants and reports anxiety 5.72 STAI-S units lower in the music groups, with a 95% confidence interval of -7.27 to -4.17¹. Tier 1.

Look at the interval before the estimate. It does not cross zero, and even at its least favourable end there is still a reduction of a size that would be visible on the questionnaire. That is a firmer picture than most of this field can offer, and it is firmer than I expected to find when I first went looking, because anxiety self reports are usually where evidence goes to become inconclusive.

Two features of the design help explain why the result is as tidy as it is. Anxiety before an operation is intense, time limited, and measured at a moment everybody can agree on, which removes a great deal of the noise that ruins longer term outcome measurement. And because nearly all of the trials used the same instrument, the review could pool the results in the instrument’s own units rather than converting everything into a standardised effect size. That is why this figure is quoted in STAI-S units and not as an SMD, and it is a small piece of good luck for anybody trying to read it.

What the STAI-S is, and what 5.72 units tells you

The STAI-S is the state subscale of the State Trait Anxiety Inventory: a questionnaire people fill in themselves, asking how anxious they feel right now rather than how anxious a person they are.

That distinction between state and trait is doing more work here than anything else on this page. State anxiety is a response to a situation, and a hospital corridor on the morning of an operation is about as strong a situation as anybody meets. Trait anxiety is a stable characteristic, and clinical anxiety disorder is something different again, assessed and diagnosed by other means entirely. This review measured the first of those three. It is the easiest of the three to move, and moving it is worth doing, and it is not the same achievement as treating an anxiety disorder. That misreading is common enough to have its own page: see music therapy for depression and anxiety in adults.

The other thing to hold on to is that 5.72 units is a difference between group averages. Inside those 26 trials, some people were much calmer with music and some were not calmed at all, and a few will have found headphones in an anaesthetic room an irritation. Nothing in a pooled mean difference tells you which of those a particular person will be. A group effect is not an individual prediction, and that caveat applies just as firmly to a clean Tier 1 result as it does to a shaky one.

The missing certainty rating, and why nobody can supply it

This review carries no certainty rating at all, and the correct interpretation of that is a date rather than a downgrade.

Certainty assessment, which asks how much confidence a pooled estimate deserves given how its component trials were run, became a routine part of Cochrane reviews after this one was published. So CD006908.pub2 does not carry one, and none can be added to it now. Describing it as high certainty, or moderate certainty, or anything else, would attribute to the review’s authors a judgement they never made, and once a rating has been invented for one review nobody can tell which of the ratings on a page are real. This site therefore says that the review predates routine certainty assessment and stops there.

What can be offered in place of a rating is the material a reader can weigh directly: the trial count, the sample size and the interval. Twenty six trials, 2,051 participants, and an interval of -7.27 to -4.17. That is not nothing. It is a reasonable number of studies, a decent total sample, and an interval that stays on one side of zero across its whole length. Those three facts are why the claim sits at Tier 1 here rather than Tier 2, and they are the honest substitute for the rating that does not exist.

Notice how differently the absence would read if it were treated as a downgrade. A missing rating is not a low rating, and a low rating is not a missing one. The 2021 cancer care review, discussed further down, carries a rating and the rating is very low, which is a considerably weaker position than this review is in. Two absences that look similar in a summary can mean opposite things, which is the whole reason for reading past the abstract.

What the intervention usually was

This is the section most coverage of the review leaves out, and it changes who is entitled to claim the result.

In the trials pooled here the intervention was, in most cases, recorded music offered around a procedure. Headphones or a bedside player, a period of listening before or during surgery, sometimes a choice of genre offered to the patient and sometimes a selection made by the researchers. That is a music intervention. It is not a course of therapy delivered by a registered practitioner working to assessed goals with an assessment, notes, clinical supervision and a planned ending, which is what music therapy means and what the profession is trained and regulated to provide.

The distinction is not defensive pedantry, and it cuts against the profession’s interests rather than for them. If the effect is produced by recorded music through headphones, then the finding belongs to an intervention that costs almost nothing, requires no registered staff, and could be offered in every preoperative area in the country tomorrow. Attributing it to music therapy instead makes the intervention look scarcer, more expensive and more specialised than the evidence says it is. It also spends a real finding on the wrong argument. In July 2026 the Health and Care Professions Council register listed 6,103 arts therapists across all four protected titles for the entire United Kingdom², which is nowhere near enough people to staff a preoperative service and is not what this evidence asks for anyway. What a registered practitioner does, and why it is a different thing, is set out on music therapy.

What you can practically do before an operation

The practical version of this finding is unusually simple: your own device, your own headphones, and a question asked in advance rather than on the day.

Ask at the preoperative assessment appointment rather than in the anaesthetic room. The people to ask are the anaesthetic team and the nurse running the list, because they are the ones who know what the room needs. Choose the music beforehand, because the morning of an operation is not the moment to be scrolling, and pick something you already know rather than something you think ought to be relaxing. The nurse’s remark I quoted at the top, unevidenced as it is, at least fits the shape of the finding: the trials that offered people a choice of their own music were testing something closer to what a person would do for themselves than a piped playlist is.

There are circumstances where it will not be possible, and they are legitimate rather than obstructive. Some procedures need continuous verbal contact with the patient. Regional anaesthesia may require you to hear and respond to instructions. Some departments have infection control policies about personal equipment, and in an emergency admission nobody is going to be looking for your earphones. A hospital arts team, where one exists, may also have something already arranged: those programmes are described on arts in hospitals, and musicians playing in clinical areas raise a set of practical questions of their own, covered on live music on hospital wards. Anything offered to somebody in a hospital bed also involves consent in a setting where refusing is socially harder than it looks, which is set out on consent and safeguarding in arts in health.

What the finding does not cover, and the claim next door

It covers state anxiety before a procedure. It does not cover pain, recovery, complications, length of stay, satisfaction, or anything that happens after the person leaves theatre.

Those are separate outcomes and each would need its own evidence. This matters because the preoperative anxiety result rarely travels alone. It tends to arrive inside a broader argument about the healing environment, and it is usually accompanied by the single most overworked study in the whole of arts and health.

Ulrich’s 1984 paper compared patients recovering in rooms with a view of trees against patients whose windows faced a brick wall. It is 46 patients in 23 matched pairs, in one 200 bed Pennsylvania hospital, after cholecystectomies only, in one season, from retrospective records covering 1972 to 1981³. That is Tier 3, and it is Tier 3 of the kind that means no adequately powered controlled evidence, not the kind that means a null result: the study found what it found, and one small retrospective matched pair comparison from a single hospital cannot carry the weight put on it. Four decades on it is still underwriting claims in hospital design documents, business cases and funding bids. Set the two things side by side and the contrast is instructive. A pooled result from 26 trials and 2,051 participants is quoted less often, and less confidently, than a retrospective comparison of 46 patients, because the second one has a better story attached. The full trail, including the parts of Ulrich’s own table that cut against the headline, is on does hospital art speed recovery.

The contrast that teaches the most

Set the preoperative review beside the cancer care review and you have the clearest available demonstration that volume of evidence and strength of evidence are different things.

The 2021 Cochrane review of music interventions for people with cancer pooled 81 trials and 5,576 participants and reports anxiety 7.73 STAI-S units lower and pain at SMD -0.67, and rates both findings very low certainty⁴. Tier 2 for both.

Three times the trials. Nearly three times the participants. A larger anxiety estimate on the same instrument. And a weaker position on this site than the smaller 2013 review, because the certainty rating exists and it is very low, chiefly reflecting small trials at high risk of bias that disagreed with one another. A reader scanning abstracts would rank these two the other way round every time.

That comparison is also the best available answer to anybody who thinks tier labels are a way of being negative about the arts. The 2013 review is smaller and older and gets the higher label; the 2021 review is bigger and newer and gets the lower one. The labels are tracking how much the evidence can bear, not how much of it there is, and they move in whichever direction the sources push them. The cancer picture in the visual arts, where the certainty problem takes a similar shape, is on art therapy during cancer treatment.

One last observation about the shape of all this. The clean results in arts and health tend to share three features: a narrow outcome, a short measurement window, and an intervention specific enough that you could write it down and have somebody else reproduce it. Anxiety before an operation has all three, which is why the finding on this page holds together, and so does the gait work on rhythmic auditory stimulation. The claims that fall apart under checking are almost always the broad ones. None of which should be read as a reason to treat headphones as part of a treatment plan: music before an operation is a reasonable thing to ask for alongside everything the surgical and anaesthetic teams are doing, and never a substitute for any of it.