Published on 2 October 2026
Which exercise, at what dose, for better sleep? What a network meta-analysis of 58 trials shows
The essentials
This network meta-analysis pools 58 randomised trials and 5,008 participants aged 13 to 85, drawn from very varied populations: its inclusion criteria required neither a sleep disorder nor poor sleep quality at entry. It compares four types of exercise (aerobic exercise, yoga included; traditional Chinese exercises such as tai chi or qigong; resistance training; combined exercise) and several ways of dosing them, with sleep quality measured by the PSQI as the outcome. On the probabilistic SUCRA ranking, the best-placed parameters are combined exercise (at least two types of exercise together, for example aerobic and resistance), at high intensity, four times a week, in sessions of 30 minutes or less, for nine to ten weeks. Each parameter was ranked in a separate analysis: the combination itself was never tested as such. Two caveats carry real weight. SUCRA places an option among the possible ranks, it does not say by how much that option outperforms the others, and in the published league table combined exercise does not differ significantly from any other type of exercise. And the publication reports no assessment of the certainty of the evidence (GRADE or equivalent), so the strength of these conclusions cannot be qualified.
Context
Patients who sleep poorly often ask whether they should exercise, and if so, what kind. The usual answer, “move more”, says nothing about what, how much or for how long. Conventional meta-analyses compare exercise with no exercise. They rarely say which form or which dose to choose.
A network meta-analysis is built for this question: it compares options that have not always been tested against each other in the same trial, by relying on a common comparator.
The study at a glance
| Item | |
|---|---|
| Population | |
| Participants with no restriction on age, sex or origin (13 to 85 years depending on the trial), from varied clinical and non-clinical populations, with no inclusion criterion relating to sleep: 5,008 participants in 58 randomised trials | |
| Interventions | |
| Aerobic exercise (yoga included), traditional Chinese exercises (tai chi, qigong, baduanjin), resistance training, combined exercise, at different frequencies, intensities and durations | |
| Comparator | |
| No exercise: usual activities, health education or standard care (passive controls in all trials) | |
| Outcome | |
| Sleep quality measured by the Pittsburgh Sleep Quality Index (PSQI), a self-report questionnaire, in all included trials | |
| Design | |
| Systematic review and network meta-analysis, protocol registered in PROSPERO (CRD42024555428), reported as compliant with PRISMA-NMA; five separate networks (type, frequency, session length, programme length, intensity), options ranked by SUCRA | |
Quality check
| Item | Judgement |
|---|---|
| Protocol registration | PROSPERO |
| FindingProtocol registered in PROSPERO under number CRD42024555428. The registration date is not reported in the publication. | |
| Volume of data | Large overall, thin per option |
| Finding58 randomised trials, 5,008 participants, but unevenly distributed across types and doses of exercise, which the authors acknowledge as a limitation. | |
| Certainty of evidence | Not assessed |
| FindingThe publication reports no GRADE assessment. The risk of bias of the trials was assessed (Cochrane tool), but how robust the conclusions are remains unknown. | |
| Homogeneity of populations | Low |
| FindingParticipants aged 13 to 85, very diverse populations (judging by the trial titles: cancer, sleep apnoea, rheumatoid arthritis, heart failure, pregnancy, students, psychiatric disorders). | |
| Funding and conflicts of interest | Public, no conflict declared |
| FindingFunded by the National Social Science Fund of China (Grant No. 22BTY011). The authors declare the absence of any commercial or financial relationship that could be construed as a conflict of interest. | |
Results
| Item | Ranked best by SUCRA |
|---|---|
| Type of exercise | Combined exercise (at least two types together) |
| FindingAerobic exercise alone also improves sleep compared with control, but ranks behind combined exercise, with no significant difference between the two (SMD -0.42, 95% CI -1.15 to 0.30). | |
| Intensity | High intensity |
| FindingRanked ahead of moderate intensity, with no significant difference between the two (SMD -0.41, 95% CI -1.52 to 0.69). Only two trial arms are rated as high intensity, both of resistance training in older people. | |
| Frequency | Four times a week |
| FindingBetter than control (SMD -1.09, 95% CI -1.92 to -0.26), but not significantly different from any other frequency in the league table. | |
| Session length | 30 minutes or less |
| FindingThe only session length significantly better than another (sessions of 40 to 55 minutes); no significant difference from sessions of 60 minutes or more. | |
| Programme length | Nine to ten weeks |
| FindingBest-ranked duration among those in the included trials, significantly better than six to eight weeks, with no significant difference from the other durations. | |
Critical appraisal
| Item | Judgement |
|---|---|
| Meaning of the SUCRA ranking | Often overinterpreted |
| FindingSUCRA summarises an option’s position across all possible ranks, not the size of its advantage. An option ranked first may do only marginally better than the second. | |
| Certainty | Cannot be qualified |
| FindingWithout GRADE, there is no way of telling whether confidence in each comparison is high or very low. | |
| Clinical heterogeneity | Substantial |
| FindingVery different populations are pooled, and the authors report high statistical heterogeneity. Extrapolating to patients under psychiatric care calls for caution. | |
| Origin of trials | Multinational |
| FindingTrials conducted in the United States (13), China (16, of which 7 are designated “Taiwan, China” and 4 “Hong Kong, China” in the publication), Iran (6), Spain and Brazil (4 each), Turkey (3), Canada, Ireland and India (2 each), and South Korea, France, Vietnam, Australia, Japan and Germany (1 each). | |
| Applicability | Direct |
| FindingThe parameters translate readily into concrete advice, provided they are presented as the best ranked, each in its own analysis, and not as proven superior. | |
Level of evidence
The level of evidence is that of a meta-analysis of randomised trials (Oxford CEBM level 1a), subject to the high heterogeneity reported by the authors. The overall effect of exercise on the PSQI score, against passive controls, is clear (SMD -0.68, 95% CI -0.89 to -0.47) and stable in sensitivity analysis. But its certainty was not qualified, and passive controls do not rule out a placebo or Hawthorne effect, as the authors acknowledge.
Confidence is low for the detail of the ranking. Without a GRADE assessment, with each parameter ranked in a separate analysis and with differences between options mostly non-significant in the league table, the “ideal” combination remains a reasonable hypothesis, not a proven prescription.
The colleague test
What an experienced colleague would say if you presented this study in two minutes, between two consultations.
“Combined, intense, four times a week, half an hour at most, a little over two months. I can tell a patient that, it’s concrete. But I’ll tell them it’s the best option available according to the ranking, not a certainty, since the certainty of the evidence was never assessed. And the 30 minutes is what will get them started.”
Translation for practice: precise exercise advice is better than vague advice, provided no result is promised.
What you can do with this
- What you can suggest: a combined exercise programme (for example aerobic and resistance), four times a week, in short sessions, over nine to ten weeks, adapted to the patient’s physical condition.
- What you can tell the patient who has no time: the best-ranked sessions last 30 minutes or less.
- What you can say honestly: each parameter is the best ranked in its own statistical analysis, the combination was not tested as such, and this is not a treatment with a precisely quantified effect.
- What you can teach: the difference between a position in a ranking and the size of an effect, using this example.
Frequently asked questions
What kind of exercise should I suggest to a patient who sleeps poorly?
In this network meta-analysis, combined exercise (for example aerobic and resistance training), high intensity, four sessions a week, sessions of 30 minutes or less and a programme of nine to ten weeks are the best-ranked parameters, each in a separate analysis. Put together, they have not been tested as a package. This is a ranking, not a demonstration of superiority.
What does SUCRA mean?
It is an index, expressed as a percentage, that summarises an option’s position across all possible ranks in the network of trials: the higher it is, the better the option ranks. It orders the options, it does not say by how much the first outperforms the next.
Do these findings apply to patients under psychiatric care?
With caution. The trials include participants aged 13 to 85 from very diverse populations, and the inclusion criteria did not require disturbed sleep at entry. Judging by their titles, a few trials concern people with depression, outpatients with various psychiatric disorders or drug users, but the publication reports no separate analysis of these patients.
Why is the absence of GRADE a limitation?
Because GRADE is the tool that indicates how far each estimate can be trusted, taking into account bias, imprecision and heterogeneity. Without it, there is no knowing whether confidence is high or very low.
Annotated bibliography
Source study. Li L, Wang C, Wang D, Li H, Zhang S, He Y, Wang P. Optimal exercise dose and type for improving sleep quality: a systematic review and network meta-analysis of RCTs. Frontiers in Psychology. 2024;15:1466277. DOI: 10.3389/fpsyg.2024.1466277. PMID: 39421847. PROSPERO protocol CRD42024555428. Funding: National Social Science Fund of China (Grant No. 22BTY011). Declared conflicts of interest: the authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Editorial collections
Tags
Verified on 2 October 2026 against the full text of the publication and its supplementary material where available. This analysis underwent an independent double reading. The English version was checked for conformity on 2 October 2026, against the figures of the French version and against the source. How we verify what we publish
This analysis is intended for healthcare professionals. It does not constitute a prescribing recommendation and does not replace individual clinical judgment.
Analysis from Psychiatry Evidence Base, evidence-based psychiatry, explained with rigor.