Published on 17 September 2026
Exercise in Bipolar Disorder: Which Symptoms Respond, and How Confident Are We?
Seven randomized trials, 576 patients with bipolar I disorder, bipolar II disorder, or an unspecified subtype, and one simple question: does structured exercise added to usual care change anything? For depressive symptoms, the standardized mean difference is minus 0.63 (95% confidence interval minus 1.11 to minus 0.14, p = 0.01), with heterogeneity across trials of 85%. For manic symptoms, the difference is minus 0.23 and does not reach significance. For anxiety, measured in only three trials, it is minus 0.70. The publication’s grading table rates the certainty of evidence as low for all three outcomes, and the authors calibrate their conclusions to that low level. That honesty is what makes the review usable, provided the effect size is never read apart from the certainty attached to it.
Background
Recommending exercise to a patient with bipolar disorder costs nothing, requires no prescription, and carries no drug interaction. That is precisely why the advice is often given without knowing what to expect from it. Two questions remain open in consultation: which symptom dimension the effect targets, and under what modalities.
The distinction between the depressive and manic poles is the one that matters. An intervention that improved both indiscriminately would be suspect; one that acted on only one of the two would be consistent with what is known about the dissociation of their underlying mechanisms. The review examined here separates the three dimensions, depression, mania and anxiety, and that separation is its real contribution.
Study at a glance
| Element | |
|---|---|
| Population | |
| Detail576 adults with bipolar I disorder, bipolar II disorder, or an unspecified subtype, diagnosed per DSM-5 or ICD-11, in a depressive, manic or remission phase, receiving usual care. Seven trials published between 2012 and 2024. | |
| Intervention | |
| DetailStructured exercise programme: walking, jogging, stair climbing, calisthenics, qigong, yoga, or a programme combining exercise with a psychological intervention. No resistance-training trial was included, although this format was eligible. Sessions of 30 to 120 minutes, 2 to 12 sessions per week, programmes of 3 to 24 weeks, per the text of the publication; the characteristics table reports fourteen weekly sessions for one trial, beyond the stated range, and reports neither session duration nor frequency for another. 288 patients. | |
| Comparator | |
| DetailUsual care without a dedicated exercise programme, in all seven trials. 288 patients. Trials whose control arm itself included exercise were excluded by the selection criteria. | |
| Outcomes | |
| DetailDepressive, manic and anxiety symptoms measured with clinical scales at the end of the intervention. Instruments differ from one trial to another, which requires the use of standardized mean differences. | |
| Design | |
| DetailSystematic review and meta-analysis of randomized trials, protocol registered on PROSPERO, compliant with PRISMA guidance and the Cochrane Handbook, methodological quality assessed with the PEDro scale, random-effects model, GRADE assessment of evidence quality. |
Quality control
| Point checked | Judgment |
|---|---|
| Protocol registration | Present |
| FindingThe review is registered on PROSPERO under identifier CRD420251041926, mentioned in the abstract and in the methods section. This limits after-the-fact selection of the outcomes analyzed. The publication does not reproduce the content of the protocol, so it is not possible to verify what was planned against what was not. | |
| Number of trials included | Seven, three for anxiety |
| FindingA pool of seven trials allows an estimate, not a fine-grained exploration. The anxiety result rests on three trials, making it the most fragile of the three, despite carrying the largest effect size. | |
| Heterogeneity for depression | 85% |
| FindingAt this level, the pooled estimate poorly describes the trials it summarizes. The question is no longer whether the average effect is correct, but whether there is an average effect that means anything at all. | |
| Grading of the evidence | Applied |
| FindingThe publication’s GRADE table rates the evidence as low for all three outcomes, depression, mania and anxiety, and the abstract says the same. The body of the text, however, writes very low for depression and mania: the publication contradicts itself on this point, and it is the table’s and abstract’s version that we retain. Either way, the authors calibrate their conclusions to a low level of certainty, which is uncommon and worth flagging. | |
| Geographic origin of the trials | Not reported by the publication |
| FindingThe characteristics table has no country column, and the origin of the trials is given nowhere in the text. The search drew on three Chinese databases in addition to four international ones, and the authors explicitly restrict their search to literature in Chinese and English. Three of the seven trials are cited in the bibliography only as entries in Chinese databases, without an author, a title or a journal. This is a gain in search sensitivity and a limit on generalizability. | |
| Safety and tolerability | No events reported |
| FindingThe review devotes a section to adverse events and states that none were reported across the seven trials. It does not say whether tolerability was systematically collected in these trials, and the characteristics table has no safety column. An absence of reporting should not be read as an absence of risk. | |
| Funding and conflicts of interest | None declared |
| FindingThe publication explicitly states that no financial support was received for the research or its publication, and that the authors have no commercial or financial relationship that could constitute a conflict of interest. This is an explicit statement of absence, not a missing section. | |
Results
| Result | Value |
|---|---|
| Depressive symptoms, seven trials, 576 patients | Standardized mean difference minus 0.63 (minus 1.11 to minus 0.14), p = 0.01, heterogeneity 85%, low certainty |
| ReadingThe interval excludes zero, so the effect is statistically present. It ranges from very modest to substantial, and the certainty the authors assign indicates that the estimate could change substantially with new trials. The trial-by-trial detail is harsher than the average: four of the seven trials find no between-group difference on depression, and the three that do are precisely the three Chinese database entries, which together account for 432 of the 576 participants. | |
| Manic symptoms, five trials | Standardized mean difference minus 0.23 (minus 0.67 to 0.21), not significant, heterogeneity 60.6%, low certainty |
| ReadingThe interval contains zero and extends on both sides. This result does not demonstrate that exercise has no effect on mania: it shows that five trials are not enough to decide. The distinction is not rhetorical, it forbids writing that exercise is ineffective on this pole. | |
| Anxiety symptoms, three trials | Standardized mean difference minus 0.70 (minus 1.26 to minus 0.15), heterogeneity 71.4%, low certainty |
| ReadingThe largest effect size in the whole body of work rests on the smallest number of trials. This is a classic configuration, and it calls for caution rather than enthusiasm. | |
| Modalities associated with the largest effect on depression | Sessions of one hour or less (minus 0.86), more than five sessions per week (minus 0.76), programmes of twelve weeks or less (minus 0.79) |
| ReadingThese values come from subgroup analyses, that is, comparisons between trials rather than within them. They describe associations, they do not prescribe a dosage. The publication does not state whether these subgroups were pre-specified in the protocol, and the high-frequency subgroup rests on only two trials totalling 123 patients. | |
Critical appraisal
| Domain | Judgment |
|---|---|
| Question and inclusion criteria | Clear |
| FindingPopulation, intervention and comparator are defined without ambiguity, and the three symptom dimensions are analyzed separately rather than merged into a single global score. | |
| Search and selection | Dual extraction |
| FindingSeven databases searched, four international and three Chinese, from inception to May 2025, supplemented by a manual reference search. Selection and extraction were performed in duplicate, with disagreements arbitrated by a third researcher. The flow diagram, however, is inconsistent: of the 158 articles retained after title and abstract screening, the exclusions detailed at full-text review remove 55, which should leave 103, not the seven finally included. | |
| Risk of bias in the included trials | PEDro 6 to 8, mean 6.29 |
| FindingPEDro scores range from 6 to 8, mean 6.29, which the authors themselves classify as good methodological quality, with no low-quality trial included. The detail is less flattering: no trial blinded the therapists, only one blinded the patients, four blinded the outcome assessors, and only two report concealed allocation. An exercise trial cannot blind the patient, which weighs on outcome measures that are themselves self-reported. | |
| Pooling of the data | Questionable for depression |
| FindingWith heterogeneity of 85%, the weighted average becomes a statistical object more than a clinical quantity. The authors’ sensitivity analysis identifies one trial as the main source of this dispersion: removing it drops heterogeneity from 85% to 33.6% and the pooled effect becomes minus 0.46 (minus 0.74 to minus 0.18). This one trial alone accounts for 242 of the 576 participants. The effect therefore survives its removal, smaller and more homogeneous. | |
| Publication bias | Not assessable |
| FindingThe authors explicitly state that they did not assess publication bias, the number of trials being too small, and seven trials indeed do not allow a meaningful test of funnel-plot asymmetry. The grading table in the same publication nonetheless lists undetected for all three outcomes: an absence of analysis is presented there as an absence of signal. It should not be read as an absence of bias. | |
| Adequacy of the conclusions | Calibrated |
| FindingThe authors write that current data suggest a benefit, not that they establish one. The non-significant result on mania is reported as such. | |
Level of evidence
Confidence is high in how the review was conducted: registered protocol, dual extraction, explicit grading of evidence, conclusions adjusted to that level. It is also high on the dissociation of the symptom dimensions, which is the real contribution of this work.
It is low on the size of the effect. Heterogeneity of 85% on the primary outcome means the trials are probably not measuring quite the same thing, and a certainty the authors themselves describe as low means the estimate is liable to be revised. What is suggested is a benefit on depressive symptoms. What is not established is its magnitude. What is not settled is the effect on manic symptoms, for lack of statistical power. The modalities associated with the largest effect remain, as things stand, a hypothesis to test rather than a recommendation.
The colleague test
What an experienced colleague would say if shown this study in two minutes, between two consultations.
« Seven trials, 85% heterogeneity, and authors who write in black and white that their evidence is low quality. I will keep recommending physical activity, as before, because the benefit-to-risk ratio justifies it. But I am not going to quote a patient an effect size on this basis, and I will not say it does not work on mania: nobody has shown that. »
Translation for practice: the review reinforces a habit, it does not ground a numbered prescription. Its real contribution is separating the poles, and reminding us that the documented effect concerns the depressive dimension.
What you can do with this
- Advising physical activity remains reasonable for a bipolar patient in a depressive phase, with one added argument: pooling seven randomized trials tips the same way. It is worth knowing that four of these seven trials, taken alone, show no between-group difference.
- Do not promise an effect on manic symptomatology. The available data do not settle the question, and presenting that silence as an absence of effect would be a misreading.
- The most studied modalities are relatively short, frequent sessions, over programmes of a few weeks. That is a reasonable starting point for setting a goal with the patient, not a validated dosage.
- Where a structured exercise-referral pathway is available locally, using it turns the advice into part of a formal care plan rather than leaving it to the patient’s own initiative.
- What still needs watching in practice: sleep and rhythm regularity in a patient who increases activity, independently of what this review shows.
Frequently asked questions
Is a standardized difference of minus 0.63 a large effect?
Taken alone, it would correspond to a moderate effect. But the confidence interval runs from minus 1.11 to minus 0.14, which covers everything from a clear effect to a marginal one, and the certainty assigned to this estimate is low. What can be retained is the probable existence of a benefit, not its size.
What does 85% heterogeneity actually mean?
That the spread of results across trials far exceeds what chance alone would explain. The interventions, populations and scales differ too much for a single average to represent them faithfully. It is an invitation to look at the trials one by one rather than to retain the pooled figure.
Could exercise trigger a manic switch?
The review did not test this. It reports that no adverse event was recorded across the seven trials, all conducted with low- to moderate-intensity activities, and its discussion recalls earlier observations of manic-like symptoms in bipolar patients undertaking marathon training or high-intensity interval training. Nothing is demonstrated in either direction. The usual clinical vigilance over sleep and rhythms remains warranted in a patient who changes their activity level.
Why is the effect size largest for anxiety?
Because only three trials measured it. Estimates from small pools are unstable and tend to overstate the effect. This is the most appealing result of the review and the least solid.
Can the identified modalities be used to prescribe a regimen?
No. These modalities emerge from comparisons between trials, not from randomization between formats. They can guide the design of a future trial, they do not fix a dose.
Annotated bibliography
Source study. Li X, Liu F, Ding F, Ma X, Zhu Y. Exercise interventions for depressive, manic, and anxiety symptoms in bipolar disorder: a systematic review and meta-analysis. Frontiers in Psychiatry. 2025 ; 16 : 1648008. DOI 10.3389/fpsyt.2025.1648008. PMID 41058644. Received 16 June 2025, accepted 25 August 2025, published 22 September 2025, open access under a Creative Commons licence. The authors declare no financial support and no competing interests.
Protocol registration. PROSPERO, identifier CRD420251041926.
Editorial collections
Tags
This analysis underwent an independent double reading in French, verified on 2 September 2026 against the full text of the publication and its supplementary material where available. The English version was checked for compliance on 17 September 2026 against the freely accessible full text on Frontiers in Psychiatry (CC BY licence). How we verify what we publish
This analysis is written for healthcare professionals. It does not replace individual clinical assessment or current guidelines, and it does not constitute therapeutic advice.
Psychiatry Evidence Base, evidence-based psychiatry, explained with rigour.