Music for insomnia: a Cochrane review of 13 trials where sleep felt better but did not measure better

Listening to something calming at bedtime is probably the most common sound-and-sleep practice there is, and this is the best evidence on it. Jespersen and colleagues searched eleven databases and two trial registers to December 2021 and found thirteen randomised trials, 1,007 adults with sleep difficulties, listening to prerecorded music for 25 to 60 minutes a day for anywhere from three days to three months. The trials ran in eight countries, mostly in participants’ own homes, and all of them used music described as sedative or relaxing.

On the headline outcome the result is clear. Pooling ten trials and 708 participants, music improved Pittsburgh Sleep Quality Index scores by 2.79 points (95% CI 1.72 to 3.86), which is roughly one standard deviation, and the review authors rate that moderate-certainty evidence. On a scale where anything above 5 marks a sleep problem, that is a real shift in how people rate their own nights.

Then the picture splits, and the split is the reason this summary exists. Three trials measured sleep with polysomnography rather than a questionnaire, and none of them found an effect on anything: not sleep-onset latency, not total sleep time, not sleep efficiency, not time awake after falling asleep. The same four outcomes measured by questionnaire instead moved in music’s favour, with low-certainty evidence each time: experienced sleep latency improved by 0.60 points on the PSQI subscale, sleep duration by 0.69, sleep efficiency by 0.96. Insomnia severity, measured with the Insomnia Severity Index in two trials, showed nothing at all, and the review grades that very low certainty. The two trials that enrolled people with a formal insomnia diagnosis both found better sleep quality and no change in insomnia severity.

The authors’ own summary of this, in the implications section, is the most careful sentence in the paper: the review “provides no evidence of an effect of the music intervention on objective measures of sleep-onset latency, total sleep time, sleep interruption and sleep efficiency”, while subjective measures of the same things improved. No trial reported any adverse event.

We grade it mixed rather than moderate, which is where the rubric score of 4.0 would otherwise put it. The partial-findings override applies: the subjective and objective halves of this review disagree, and reporting only the half that favours music would misdescribe the paper.

Limitations. Every trial compared music against no treatment or treatment as usual, so nothing here separates the music from the ritual of lying down with headphones on and expecting to sleep better. No participant could be blinded, which is why the review downgraded almost every outcome for risk of bias, and the people scoring the questionnaires knew who had listened. Heterogeneity in the main analysis was 81 percent, meaning the trials disagree substantially about the size of the effect. The review covers music, not binaural beats or other engineered audio; for those, see our summaries of Lee 2024 and Lin 2024.