Exercise for low mood

This is a network meta-analysis published in the BMJ in 2024, pre-registered on PROSPERO before it was run: 218 randomised trials, 495 trial arms, 14,170 participants, all of whom met clinical cut-offs for major depression. Its design lets it place every exercise type, psychotherapy, medication and the control conditions in a single statistical model, which is why it can say something about which exercise rather than only whether exercise. Walking or jogging came out at Hedges' g -0.62 against active controls, from 51 arms containing 1210 participants - the largest effect among the well-populated exercise types, with yoga, strength training, mixed aerobic exercise and tai chi all somewhat behind it. One thinly populated type did come out higher: dance, at g -0.96, but on five trial arms and 107 people, which is too little to build anything on. Now the honest weight of it. Of the 218 trials, one met the Cochrane criteria for low risk of bias. One. Participants and staff were hardly ever blinded, which is much harder to do with exercise than with a tablet, so expectation is inside these numbers and cannot be taken out. The reviewers graded their own confidence using the CINeMA tool and rated it low for walking or jogging and very low for everything else. They are fair to themselves in both directions: their sensitivity analyses suggested the effect sizes were not actually being driven by any particular bias, and they note that older trials may be judged harshly because they predate modern reporting checklists. Many of the trials were also small - the median walking or jogging arm held 17 people. The right way to hold this is that the direction is well supported and the exact size is not.

Noetel M et al. 2024

Depression improves on its own more than almost any condition studied this way, which makes the choice of comparison group the difference between a real finding and an illusion. This review handled it well and the detail is worth knowing. Its main comparison was not people left alone. "Active control" here means usual care, a placebo tablet, stretching, an educational session or social support, grouped together because they all performed similarly. Waitlist groups - people told they would get the treatment later - were deliberately kept separate, because waitlists reliably do worse and lumping them in would have flattered every treatment in the network. The scale of what that protects against is visible in the reviewers' own numbers: the active control groups improved by roughly a full standard deviation from start to finish, all by themselves. The exercise effect reported on this page is the ground gained on top of that. It is also why the reviewers set a deliberately modest threshold, Hedges' g -0.20, for what counts as a difference worth having against an active control - and why clearing it at -0.62 means something.

Noetel M et al. 2024

These get blurred constantly and they are not the same question. The randomised evidence, the strong stuff, is entirely about programmes: weeks of prescribed sessions, measured against a comparison group. It says nothing at all about how you will feel this afternoon. There is a separate literature on the single session, and it is genuinely weaker in a specific and important way. A meta-analysis of 103 studies measured mood before a bout of exercise and again within half an hour of finishing, and found people rated themselves better afterwards - Hedges' g 0.336. A second meta-analysis, of 18 studies restricted to people with depression, found a larger before-and-after change of 0.585. Neither had a control group. Nobody sat quietly for the same half hour. So moods drifting on their own, the tendency to answer a questionnaire differently the second time, and the simple expectation of feeling better after exercise are all still inside those numbers, and cannot be separated from the exercise. Both sets of reviewers say so; the second states it as their published limitation. What this means practically: the single walk is a reasonable thing to do and people consistently report feeling better after one, but the claim that carries real evidential weight is the repeated one. Whether the lift after one session predicts the benefit after ten weeks is an open question the researchers have explicitly asked somebody to go and answer.

Weinstein AA et al. 2024 · Bourke M et al. 2022

Three results shape what can honestly be recommended. Intensity mattered: light activity, for which the reviewers give walking as their own example, produced g -0.58 against active controls, while vigorous activity such as running or interval training produced g -0.74. The intervals around those two overlap a great deal, so read it as a tendency, not a rule - and note that light activity still cleared the bar for a difference that matters. Weekly amount was unresolved: the review's intervals were too wide to separate weekly doses, which is why no minutes-per-week target appears anywhere on this page. And then the awkward one. Trials that gave participants a choice over frequency, intensity, type or timing tended to show WEAKER effects, g -0.28, than trials that prescribed the regimen, g -0.75. The reviewers call this contrary to their own expectations. Handle it carefully: this is a comparison between whole trials, not between individuals, the interval around the choice group includes no effect at all, and a trial that prescribes a regimen also tends to supervise it, chase attendance and provide structure - any of which could be doing the work. It cannot support the claim that choosing your own walk makes it work less well for you. What it does suggest, and what the fix on this page is built on, is that the thing that got tested was a decided, scheduled session - so decide it in advance rather than on the day. What that means for how you actually walk is less prescriptive than most advice on this subject. The evidence that has been through randomised comparison is about programmes lasting weeks rather than a single heroic outing, and its benefit was still measurable when people were followed up to six months later. No trial removed anybody's therapy or medication, so nothing here can tell you what happens if you stop one. The reviewers could not resolve a difference between weekly amounts, so there is no target to fail at; harder tended to help a little more, and light activity - walking is the reviewers' own example - still cleared the bar they had set for a difference that matters. Start where you can start.

Noetel M et al. 2024

Because the review models everything in one network, it produces estimates for the other treatments too, and those numbers travelled a long way in the press. Inside this network, and against the same active controls: cognitive behaviour therapy alone came out at Hedges' g -0.55, exercise combined with antidepressants at -0.55, and antidepressants alone at -0.26. Set beside walking or jogging at -0.62, that reads like a ranking. It is not a fair one, and the reviewers say so themselves in the same paragraph. Their search was built to find exercise trials, so drug and therapy arms appear only where an exercise trial happened to include one; the estimates "should not usurp" reviews designed to study those treatments directly. The authors point to exactly such a review for medication, which reports a larger effect than the figure produced here, with a much tighter range. There are two further reasons the comparison does not hold. Drug trials can blind participants to what they are taking and exercise trials essentially cannot, so expectation loads onto one side of the ledger. And the people enrolled in an exercise trial are not the same people enrolled in a drug trial. The finding on this line that does hold up is the combination: adding exercise to medication, or to psychotherapy, did better than the comparison conditions. Nothing in this review tested taking a treatment away, and nothing on this page is a reason to stop one.

Noetel M et al. 2024

No, and the evidence cannot be read that way even by someone trying to. Every arm in these trials ADDED exercise; not one removed a treatment, so there is no data at all on what happens when something is taken away. The arms that combined exercise with psychotherapy or with medication did better than the comparison conditions, which is a finding about addition. Stopping an antidepressant without medical supervision can be genuinely dangerous, and relapse after stopping treatment is common. If you want to change what you are taking or doing, that is a conversation with your prescriber, with this evidence as one input among several - not a decision to take off a page.

Partly, and it is worth being precise about which part. The strong randomised evidence comes entirely from people who met clinical cut-offs for major depression, so strictly it tells you about them and not about an ordinary flat fortnight. The reviewers did find the benefit similar across different starting severities within that group, which is mildly reassuring but is not the same as evidence in people below the threshold. The single-session research, which does include general populations, is the part that speaks to a low day - and that research has no control groups, so it can tell you people report feeling better after a session, not that the session is what did it. The reasonable position: a brisk walk is a sensible thing to do on a low day, the case for it is much stronger if you are actually depressed, and it is free either way.

It is, and this evidence does not solve it. Loss of drive is a symptom of depression rather than a failure of willpower, and every trial here had a researcher organising sessions, following up and keeping people going, which is help most readers do not have. That is worth saying rather than hiding, because a fix that assumes motivation you do not have is not a fix. Two honest observations from the trials themselves. Dropout was low enough in several exercise types for the reviewers to describe them as well tolerated, so people who were genuinely depressed did keep going once something was organised for them. And the sessions that worked were decided in advance rather than left to the day, which is the one piece of that structure you can copy for free. If you cannot start, that is a reason to ask for help, not evidence that you have failed at a walk.