Published on 19 September 2026
Aerobic or mind-body exercise for substance use disorders: what does the ranking actually show?
The essentials
Fifty-seven randomized trials drawn from 33 articles, 2,922 participants treated for a substance use disorder, five outcome dimensions and two forms of exercise, aerobic and mind-body practice. No included trial directly compared the two modalities, so the authors built two separate networks whose nodes are not the interventions but the five outcome dimensions. The resulting rankings look decisive, with a surface under the cumulative ranking curve of 0.874 for physiological health in the aerobic network and 0.884 for sleep quality in the mind-body network. Of the twenty comparisons these two networks produce, only one reaches significance, and it opposes two outcome dimensions, not two interventions. No assessment of the certainty of evidence accompanies these results. What remains: nothing here allows one modality to be ranked against the other, and this publication reports no effect estimate of exercise compared with usual care.
Context
Physical exercise has long been part of the background of addiction care, without its place ever being firmly established. It is associated with better treatment adherence and with improvements in mood, sleep and fitness. The question that keeps coming up in practice is one of choice: should a patient be pointed toward running or toward yoga, and for what benefit.
This network meta-analysis sets out to answer exactly that, by ranking modalities dimension by dimension. The method is appealing, and the result, as it reads in an abstract, looks clean. It is worth looking at what the ranking actually covers, and above all at what was put into the network.
Mechanism, and what the study did not measure
The authors invoke molecular mechanisms to account for their rankings: activation of AMP-activated protein kinase, regulation of the hypothalamic-pituitary-adrenal axis, upregulation of GABA receptors. None of these mechanisms is measured in the included trials. They are hypotheses borrowed from experimental literature, often conducted in animals or in older adults, used to explain differences that, in the data actually presented, are for the most part not established. This gap between the level of explanation and the level of evidence is itself worth flagging to the reader.
The study at a glance
| Question (PICO) | |
|---|---|
| Population | |
| 2,922 participants treated for a substance use disorder, drawn from 57 randomized trials across 33 articles. Inclusion criteria targeted adults aged 18 and over meeting DSM-5 or ICD-10 criteria; the included-trials table lists several smoking-cessation trials conducted in smokers. | |
| Intervention | |
| Aerobic exercise, for example slow running, brisk walking or cycling, or so-called mind-body practices, tai chi, yoga, qigong. | |
| Comparator | |
| Usual care, including pharmacological and psychological treatment with no exercise component, or a waiting list. No effect estimate of exercise compared with this control group is reported in the publication. | |
| Outcomes | |
| Five dimensions: substance use, mental health and emotional symptoms, cognitive function, physiological health indicators, sleep quality. Ranking by surface under the cumulative ranking curve within each network. | |
| Design | |
| Systematic review and frequentist network meta-analysis in Stata 16.0. Six databases searched, PubMed, Web of Science, Cochrane Library, EMBASE, SCOPUS and ScienceDirect, from inception to July 1, 2025; 11,689 records identified, 2,890 read in full text, 33 articles retained. PROSPERO registration CRD420251144089, stated compliance with the PRISMA extension for network meta-analyses, trial quality assessed with the Cochrane risk-of-bias framework in Review Manager 5.4. With no included trial directly comparing aerobic and mind-body exercise, two independent networks were built, with the five outcome dimensions, not the interventions, as nodes. Oxford level of evidence 1a by design. |
Quality control
| Point checked | Judgment |
|---|---|
| Protocol registration | Present and identified |
| FindingPROSPERO registration, identifier CRD420251144089. Selection by two independent reviewers across six databases, with arbitration by a third. | |
| Risk-of-bias assessment | Done, reported as a figure |
| FindingSeven-domain Cochrane framework, two independent assessors, results reported as a summary figure. No trial-by-trial table appears in the published text. A funnel plot and Egger’s test for publication bias are announced in the methods; no corresponding result is reported. | |
| Certainty of evidence | Absent |
| FindingThe text mentions neither GRADE nor its network adaptation. The reader has no way of judging how much confidence to place in the rankings presented. | |
| Nature of the network nodes | Dimensions, not interventions |
| FindingThis is the decisive point. The authors state that no included trial compared aerobic exercise with mind-body practice, so they built two separate five-node networks, in which the nodes are the five outcome dimensions. The comparisons produced therefore oppose outcomes to each other, not interventions. A network meta-analysis built this way cannot, by construction, decide between two exercise modalities. | |
| Internal consistency of the network | Uninformative test |
| FindingThe node-splitting test found no inconsistency, with every p-value above 0.05, ranging from 0.082 to 0.989. But the associated standard errors reach 313, 630 or 835 in the aerobic network and up to 835 in the mind-body network. The authors themselves acknowledge that such wide intervals may reflect clinical heterogeneity or a lack of direct comparisons rather than genuine consistency. | |
| Geographic origin of the trials | Not reported |
| FindingThe authors state that they extracted each trial’s country, but the included-trials characteristics table has no country column and no geographic summary is presented anywhere. The geographic spread of the evidence base cannot be verified from the publication. | |
| Funding and conflicts of interest | Declared absent |
| FindingThe authors declare having received no financial support for this work or its publication, and no commercial or financial relationship that could constitute a conflict of interest. They also declare not having used generative artificial intelligence in drafting the manuscript. | |
Results
| Result | Value |
|---|---|
| Aerobic network, ranking of dimensions | 0.874 for physiological health |
| ReadingThen 0.468 for substance use, 0.446 for sleep, 0.427 for mental health and 0.286 for cognition. A surface under the cumulative ranking curve expresses a probability of occupying the best ranks, not an effect size, and not evidence of efficacy. The authors gloss this value as indicating “87.4% probability of ranking in the top tier,” a reading the statistic does not by itself support. | |
| Mind-body network, ranking of dimensions | 0.884 for sleep |
| ReadingThen 0.608 for cognition, 0.588 for mental health, 0.243 for physiological health and 0.177 for substance use. Same caveat applies. These values come from a network distinct from the one above: they are not directly comparable, term by term, with the aerobic network’s figures. | |
| Comparisons produced by the two networks | 19 of 20 not significant |
| ReadingA recount across the two comparison tables shows that all ten estimates in the aerobic network have a confidence interval that includes zero, and so do nine of the ten estimates in the mind-body network. This is the article’s central finding, and it does not appear in its title. Sharp rankings resting on comparisons that are not sharp point to an uninformative network, not to a hierarchy. | |
| The one significant comparison | Effect size 44.58 (3.30 to 85.85) |
| ReadingIt opposes substance use to sleep within the mind-body network, that is, two outcome dimensions rather than two interventions. A comparison of this kind carries no clinical content. There is also a scale problem: the methods announce standardized mean differences, yet the reported values in this network range from 0.12 to 318.96 with bounds exceeding 900, a range compatible with no standardized mean difference. | |
Critical appraisal
| AMSTAR 2 domain | Judgment |
|---|---|
| Question and study selection | Relevant, selection questionable |
| FindingA clinically relevant question, a broad search across six databases, a substantial base of 57 trials. But two entries in the included-trials table point to references that are not randomized trials conducted in substance use disorder: a meta-analysis on cognition in older adults and a meta-analysis on high-intensity interval training in overweight adults. Two further entries point to a single reference, listed twice under different sources, with mismatched sample sizes. | |
| Network structure | Not fit for the question |
| FindingThe nodes are the outcome dimensions, not the interventions, and the two modalities are analyzed in two separate networks. The method therefore cannot produce the comparison between aerobic and mind-body exercise that the conclusion claims, nor the personalized prescription it calls for. | |
| Interpretation of the rankings | Overreaching |
| FindingA high rank computed from non-significant comparisons is still a high rank: that is a mathematical property of the statistic used, not a clinical signal. The discussion goes a step further, calling mind-body practice “significantly superior to aerobic exercise” on cognition, 0.608 against 0.286, when the two values come from different networks and no test compares them. | |
| Certainty of evidence | Not assessed |
| FindingThe absence of grading deprives the reader of the one tool that would have put the rankings presented into perspective. | |
| Internal consistency of the publication | Faulty |
| FindingThree passages in the results section describe the interventions’ effect “on depressive symptoms,” two of them “in students,” a population unrelated to the study’s object. The methods also mention Chinese databases, although none of the six databases listed is Chinese. These inconsistencies do not invalidate the tables, but they weigh on how much credit the rest of the writing deserves. | |
| Match between claim and evidence | Broken |
| FindingThe tables themselves are given in plain sight, which lets an attentive reader reconstruct the actual situation. It is the abstract, the mechanistic discussion and the conclusion that go further, presenting as “significantly different advantages” between two modalities something the data never compare. | |
Level of evidence
The design is that of a meta-analysis of randomized trials, hence Oxford level 1a. The real level of confidence is considerably lower, given the lack of a certainty assessment and the network structure chosen. Confidence is nil on any hierarchy between the two modalities: the publication contains no comparison between them. It is also low on the size of exercise’s own benefit, since no effect estimate compared with usual care or a waiting list is reported, even though every included trial rests on that comparator. Two points need spelling out. First, the absence of an established difference between aerobic and mind-body practice does not mean the two are equivalent: it means the data presented cannot tell them apart. Second, the publication reports no adverse event, no tolerability data and no analysis of dropouts: that silence is not a safety signal, it is an absence of analysis.
The colleague test
What an experienced addiction specialist would say if shown this network meta-analysis in two minutes, between two consultations.
Fifty-seven trials, rankings around 0.87, and when I open the actual comparison tables there is exactly one significant result out of twenty, and it compares sleep to substance use, not tai chi to running. So I keep telling patients to stay active, for reasons that have nothing to do with this article, and I do not tell them tai chi beats running for sleep. That is not what these data show.
Translation for practice: the general recommendation to stay physically active does not change, a modality-specific prescription has no basis here, and the article makes an excellent teaching case on how to read a network meta-analysis.
What you can do with this
- Keep recommending regular physical activity as part of substance use disorder care, on the strength of the prior literature rather than this publication, which reports neither an effect versus a control group nor tolerability data.
- Leave the choice of modality to the patient, based on what they enjoy and can sustain. Nothing here justifies favoring aerobic exercise or a mind-body practice for a given goal, and nothing establishes that the two are interchangeable either.
- Faced with a network meta-analysis, start by checking what the network’s nodes actually are. If they are not the interventions you want to compare, the ranking does not answer that question.
- Then look for the pairwise comparison table before reading the ranking. A high rank built on non-significant comparisons means little.
- When a discussion invokes molecular mechanisms the included trials never measured, that is often a sign the results cannot carry the conclusion on their own.
Frequently asked questions
What is a surface under the cumulative ranking curve?
It is a statistic that summarizes, for each node in a network, the probability of occupying the best ranks. It is calculated even when no difference reaches significance, and that is its weakness: it always produces a ranking, even from data that distinguish nothing. It measures neither the size nor the clinical relevance of an effect.
Should yoga be recommended over running for sleep?
Not on the basis of this article. The two modalities are analyzed here in two separate networks, with no comparison between them, and no certainty assessment informs the rankings presented.
Does exercise actually help in addiction?
This publication cannot answer that question: it reports no effect estimate of exercise compared with usual care or a waiting list, and no tolerability data. The question remains open in the prior literature, of which several trials gathered here are part, but it is not settled by the analyses presented.
Why is the stated level of evidence high when the conclusion is cautious?
Because the Oxford level describes the type of study, here a meta-analysis of randomized trials, not the robustness of its results. A high-level design can still produce low-certainty conclusions, and that is precisely the case here.
Annotated bibliography
Source study. Liu T, Chen Z, Gao K. Effects of physical exercise interventions on multidimensional health outcomes in patients with substance use disorders: a network meta-analysis. Frontiers in Psychiatry. 2026;16:1732663. DOI 10.3389/fpsyt.2025.1732663 · PMID 41614099. Received October 26, 2025, revised December 7, 2025, accepted December 11, 2025, published January 14, 2026. Open-access article under a Creative Commons Attribution licence. The authors declare having received no financial support for this work or its publication, and no commercial or financial relationship that could constitute a conflict of interest.
Protocol registration. PROSPERO, identifier CRD420251144089.
Scope of verification. This analysis draws on no reference outside the source publication: every figure, count and quotation reported here comes from the full text and the five tables it contains. The figures, notably the bias-risk summary graphic and the cumulative ranking curves, could not be read as data.
Editorial collections
Tags
Verified on September 2, 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 September 19, 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.