The Number Is Not the Prescription
The subject here is not exercise in the abstract. It is what two 2026 evidence syntheses of exercise for postpartum mental health can—and cannot—tell us when a population-level result is carried into an individual life. One synthesis models how much structured aerobic exercise is associated with changes in depressive symptoms, anxiety, and health-related quality of life. The other compares different exercise-based approaches for postpartum depressive symptoms. The practical question is how to use those findings without turning a modeled “best dose” into a rule that ignores recovery, caregiving, baseline symptoms, or uncertainty.
That distinction matters because postpartum mental health is both common and unusually context-sensitive. The first year after childbirth can combine hormonal change, interrupted sleep, physical recovery, and a new burden of care. Depressive symptoms may affect motivation, functioning, and the mother–infant relationship; anxiety can involve persistent worry and physiological tension; quality of life also includes physical capacity, social participation, and the ability to manage daily tasks. An intervention can help one of these domains more clearly than another. A useful reading of the evidence therefore begins by asking what was measured, in whom, and against what—not by asking which number sounds most authoritative.
Postpartum mental health is not one outcome
Exercise is appealing in this setting for reasons that are practical as well as biological. A structured walking, cycling, water-exercise, or similar program may be delivered at home, in a community, or with supervision. It can potentially fit alongside ordinary routines while offering physical and psychological benefits. The supplied studies discuss pathways such as behavioral activation, improved sleep, self-efficacy, stress regulation, and social contact. Those possibilities help explain why exercise has become an important non-pharmacological option to study, especially when access to psychotherapy is uneven or when people have concerns about medication during pregnancy or lactation.
But “non-pharmacological option” is not the same as “replacement for care.” The evidence syntheses compare exercise programs with specified control conditions; they do not establish that exercise alone is adequate for every person with clinically significant depression, nor that it replaces assessment or other treatments. That is especially important because the outcomes are not interchangeable.
Depressive symptoms are usually measured with a validated scale. A lower score is meaningful, but it is still a change in a measured symptom burden unless the trial explicitly assessed diagnosis. It does not automatically mean remission, prevention of a depressive disorder, or restoration of functioning in every part of life. Anxiety is related to depression but not identical to it. Worry about an infant’s safety, intrusive thoughts, sleep loss, and a sense of diminished control may require forms of support that are not captured by the same pathways through which exercise affects mood.
Health-related quality of life is broader again. It can reflect energy, physical functioning, emotional well-being, social participation, and the capacity to carry out ordinary responsibilities. Someone may experience a tangible improvement in daily life without showing the same change on a depression scale, or may have lower depressive symptoms while anxiety remains a problem. Once these outcomes are separated, the phrase “exercise improves postpartum mental health” becomes a starting point for inquiry rather than a sufficient conclusion.
This is why the word best is unstable. Best for reducing depressive symptom scores is not necessarily best for anxiety. Best for quality of life may involve a different balance of activity and recovery. And best for a group average is not automatically the best starting point for a person whose sleep, pain, support network, or physical recovery looks different from the trial participants. The number cannot be interpreted before the outcome has been named.
Two studies, two meanings of “best”
The aerobic-exercise review asks a dose question. It pools randomized controlled trials of structured aerobic exercise and uses a model to examine how weekly exercise volume relates to several outcomes. Its headline result is a modeled relationship between a composite exposure—intensity, time, and frequency—and average changes in outcome scores. The question is not simply whether participants moved. It is whether different amounts of a defined exercise exposure appear to line up with different average results across the included trials.
The network meta-analysis asks a comparative question. Instead of treating exercise as one category, it separates structured aerobic exercise, mind–body exercise, resistance training, pelvic-floor training, lifestyle physical-activity promotion, and exercise combined with other interventions. It also distinguishes passive controls from active controls that provide education, attention, social support, or other structured contact. This allows the analysis to compare interventions that were not always tested directly against one another, but it also means the conclusion depends on how comparable those interventions and populations are.
The studies do not even have identical boundaries around “postpartum.” The dose review’s eligibility criteria allowed pregnant or postpartum women, and its included set contained interventions delivered during pregnancy as well as after childbirth, even though the article presents its implications in postpartum terms. The network review enrolled women within the first 12 months after childbirth and included both women with depressive symptoms and women considered at high risk or recruited for prevention. These are relevant populations, but they are not interchangeable denominators.
The difference between the questions is easy to lose in a headline. A dose analysis can suggest that an intermediate weekly volume is associated with better depressive-symptom or quality-of-life results than very low or very high volumes in the observed data. A network analysis can suggest that one modality has clearer comparative support than another, or that a different modality ranks favorably in a subgroup. Neither result answers every other question. A volume is not a modality, a ranking is not a dose, and a favorable point estimate is not the same as a reliable superiority claim.
Imagine two programs that produce a similar weekly MET-minute total. One is structured brisk walking; another combines movement with breathing and mindfulness. The total volume alone does not tell us that the programs work through the same route, impose the same burden, or offer the same experience. Conversely, two programs with the same label may differ in supervision, social support, adherence, or the amount of activity participants actually completed. The apparent optimum belongs to the evidence definition used by the review, not to an exercise category detached from its delivery conditions.
The translation problem behind a precise number
A pooled number can look more personal than it is. The dose review expresses exercise as MET-minutes per week, a useful common language for trials with different schedules. Yet some intensities were converted from heart-rate, oxygen-consumption, or perceived-exertion targets, while others were estimated from the activity description and a physical-activity compendium. The investigators then rounded doses into categories to make the dose-response network workable. That process improves comparability, but it also means the number is partly reconstructed rather than directly observed in the same way for every participant.
The distinction becomes sharper at the individual level. A prescribed weekly load is not the same as completed exposure. Two people assigned the same program may attend different numbers of sessions, experience different effort at the same nominal intensity, or face different consequences for sleep and recovery. The review itself notes uncertainty from nonspecific exercise descriptions and variation in energy expenditure. Its study set also leaves important personal characteristics, including parity and baseline body mass index, inconsistently reported. A model can estimate the center of a pattern without telling us how to progress through that pattern safely or sustainably.
The authors translate their approximate range into ordinary minutes of moderate-intensity activity at about four METs. That arithmetic is helpful: it lets readers connect an abstract research unit with familiar activity. But it remains an illustration built on an assumed intensity. It does not prove that one exact schedule, number of days, or bout length is required, nor that every shorter or longer bout is equivalent for every outcome. The translation makes the finding understandable; it does not convert an average into an order.
A disciplined interpretation therefore has several filters. First ask which outcome matters: depressive symptoms, anxiety, quality of life, or some combination. Then ask which population the evidence most resembles, whether the goal is treatment or prevention, and what the comparator contributed. Next consider whether the exercise form is feasible alongside sleep disruption, caregiving, physical recovery, and available support. Finally, keep the uncertainty visible: the trials have limited long-term follow-up and incomplete reporting of adherence and adverse events, and the two syntheses do not provide a complete individualized starting or progression protocol.
The result is not an excuse to say only “it depends.” The syntheses do provide useful anchors: they make structured exercise a serious candidate for postpartum mental-health support, identify outcomes with more consistent evidence than others, and offer a way to think about volume and modality. The responsible conclusion is narrower and more useful: a modeled dose can orient a decision, but the decision still has to pass through outcome, population, comparator, feasibility, recovery, adherence, and evidence-certainty checks.
Before trusting any average effect, then, the necessary next question is not “What number should everyone follow?” It is “What, exactly, did the studies count?”