What the Meta-Analysis Actually Counted
This chapter examines the trial base behind the aerobic-exercise headline: who was enrolled, what researchers called exercise, which outcomes were measured, and what the comparison groups received. That reconstruction matters because an average effect is only as meaningful as its denominator. Once the boundaries are visible, a reader can interpret the review as evidence about multi-week randomized programs—not as a verdict on every form of movement or every postpartum circumstance.
The trial set behind the postpartum headline
The dose–response review included 17 randomized controlled trials involving 2,865 women. A randomized trial assigns participants to different conditions by a process intended to balance characteristics between groups, allowing a stronger comparison than an uncontrolled observation. A meta-analysis then combines results from multiple trials. It does not create a new population; it gives greater statistical weight to a set of populations that were already selected, measured, and compared in particular ways.
The included studies were published from 2003 to 2024. Their sizes varied substantially, from 19 to 784 participants, and the programs were not identical. Nine studies used aerobic exercise as the intervention, while eight combined aerobic exercise with resistance training. The exercise intensity reported or reconstructed by the review ranged from 4.0 to 5.5 METs; sessions lasted 30 to 60 minutes, occurred between once and five times a week, and continued for four to 26 weeks. Those ranges describe the material available to the analysis. They are not a menu of equally tested options.
The eligibility criteria were deliberately narrower than the word exercise might suggest. Programs had to be structured aerobic exercise: rhythmic, sustained activity involving large muscle groups, with aerobic exercise as the primary active component. They could be supervised or unsupervised, delivered in a facility, at home, or remotely. A multi-week walking program and a remotely guided activity program could therefore qualify if the aerobic component was identifiable. A single exercise session could not. Neither could a package in which diet, sleep manipulation, or other lifestyle components dominated and the aerobic dose could not be separated.
The review restricted inclusion to randomized controlled trials in peer-reviewed journals and required at least four weeks of training. That choice makes the evidence more relevant to an intervention than to an acute mood response, but it also excludes forms of activity that are difficult to standardize or that have not been tested in a qualifying trial. The review searched several major databases through October 31, 2025, applied no language restriction at the search stage, and used validated measures of depressive symptoms, anxiety, or health-related quality of life. Methodological care improves confidence in the search process; it does not remove the limits of the studies that survive it.
The most important scope qualification is easy to miss. The methods allowed women who were pregnant or postpartum, and the included set contained five interventions delivered during pregnancy and 12 conducted postpartum. The article’s framing and conclusion emphasize postpartum care, and parts of the results describe the full set as postpartum women. But a pool containing both stages is not identical to a pool limited to the first year after childbirth. Pregnancy and postpartum involve different physiological states, time demands, and sources of distress. If a result appears in a headline about postpartum exercise, the reader should check whether the underlying denominator is purely postpartum or includes related perinatal evidence.
That is not a reason to discard the review. It is a reason to use the result at the level it can support. The synthesis is relevant to the broader question of exercise around the perinatal period and contains a substantial postpartum component. It does not, by itself, prove that the same average applies unchanged to someone at a particular point after delivery, with a particular delivery history, symptom profile, or recovery course. The review reports that parity and baseline body mass index were inconsistently available across studies, which further limits the ability to identify who was represented well enough for a tailored conclusion.
“Included postpartum women” is not a complete description of the people behind the estimate. The studies may differ in when participants entered a program, whether they already had depressive symptoms, how much support accompanied the exercise, and how faithfully participants completed the assigned sessions. The synthesis can preserve an average across those differences, but it cannot fill in characteristics that were not consistently reported. The network review’s narrower first-year-postpartum eligibility does not solve this problem for the aerobic dose estimate, because it is a separate synthesis with a different study set and question.
This matters most when a reader tries to move from population relevance to personal suitability. The evidence can make structured exercise a reasonable subject for discussion in postpartum care. It cannot, from these counts alone, tell us that a person with a particular delivery history, pain pattern, medical condition, sleep situation, or level of support was adequately represented. Nor does a study’s failure to report a characteristic prove that the characteristic had no influence. It simply leaves applicability less certain. A careful interpretation treats missing information as a boundary around the claim, not as permission to assume that all postpartum circumstances are equivalent.
The trial set can be represented as a chain of decisions rather than as a single black-box number:
Each step changes what the final estimate means. The first limits the design. The second defines the population. The third separates a structured training intervention from a single bout or an inseparable lifestyle package. The fourth determines what kind of change can be seen. The fifth determines what the exercise is being compared against. Only then does the pooled effect appear.
What the average effect does—and does not—measure
The review’s primary outcomes were depressive symptoms, anxiety symptoms, and health-related quality of life. These were obtained from validated instruments, but a validated instrument is not automatically a diagnostic interview. Most results represent symptom severity at the end of an intervention or at follow-up. A lower score can indicate less reported distress; it does not necessarily establish that a participant no longer meets criteria for a disorder, that a disorder was prevented, or that the person’s functioning recovered in every relevant domain.
This distinction is particularly important for depressive symptoms. A screening scale can be sensitive to change across a group, which makes it useful for research. But the scale’s points are not the same thing as clinical states. If an exercise group’s average score falls more than a control group’s average score, the evidence supports a difference in average symptom change under the conditions studied. It does not tell us that every participant improved, that the size of the improvement was the same for everyone, or that exercise is sufficient for a person with severe or persistent symptoms.
Health-related quality of life is also not a second name for depression. It can include physical functioning, vitality, social participation, emotional well-being, and the practical capacity to manage daily life. A program may improve how capable or connected someone feels without producing an identical movement on a depression scale. Conversely, a reduction in depressive symptoms may not resolve anxiety, pain, exhaustion, or the logistical strain of caring for an infant. The outcome determines the claim.
To combine results measured with different instruments, the review used a standardized mean difference, or SMD. In plain language, an SMD expresses how far apart the average changes in two groups are relative to the amount of variation within those groups. A negative SMD in the depression analysis favors exercise because lower symptom scores are better; a positive SMD in the quality-of-life analysis favors exercise because higher scores are better. Standardization allows unlike scales to be placed on a common statistical frame, but it also makes the result less immediately interpretable than a change of, say, three points on one named questionnaire.
The confidence interval around an SMD matters as much as its central estimate. It describes a range of values compatible with the analysis and the data, not a promise that the true effect lies at the midpoint. A narrow interval suggests more precision; a wide interval leaves more possibilities open. Whether an interval crosses zero helps identify whether a difference is statistically distinguishable from no difference in that analysis, but statistical distinguishability is not the same as clinical importance. Nor does it tell us whether the result applies to a person outside the studied population.
An SMD is also not a conversion table for a single participant. If one trial uses a depression questionnaire with a wide numerical range and another uses a shorter scale, standardization helps compare their relative group differences, but it does not say that a person should expect a particular number of points to change. The same standardized difference can arise from different combinations of average change and variation. A large spread of responses can make a group contrast look different from a similarly sized average change in a more uniform group. That is one reason the central estimate should be read with its interval and with the design that produced it.
The review’s analysis also used the difference between groups, not a simple before-and-after story. Both exercise and control participants may change over time because of natural recovery, changing routines, regression toward the average, or attention from the research team. The relevant evidence is the contrast in change between conditions. This protects against the tempting but weaker statement, “participants exercised and then felt better.” It still does not identify which parts of a structured program produced the difference or whether the same contrast would appear against another form of care.
The control condition supplies another part of the meaning. In the dose review, comparators included no treatment, wait-list conditions, education or attention controls, and usual care. These are often grouped as inactive or minimal controls, but they are not conceptually identical. A wait-list participant may receive little structured contact, while an education or attention condition may provide information, expectation, or social interaction. The observed contrast can therefore include not only the physiological effects of movement but also the effects of support, routine, attention, and belief that accompany the program.
A simple example shows why the denominator and comparator belong in the same sentence. Suppose a trial reports that the exercise group’s average symptom score fell more than the usual-care group’s score. The result may reflect the exercise itself, the regular contact that helped participants sustain a routine, the social experience of the program, or several of these together. If another trial compares exercise with a support-rich active program, its estimate answers a harder question: what additional difference remains when both groups receive attention or education? The two results should not be read as interchangeable evidence of the same magnitude.
The label “usual care” deserves the same scrutiny as the label “aerobic exercise.” Routine postpartum management may differ across settings, and an education or attention condition may be more active than its short description suggests. When a review combines several minimal-control arrangements, it is summarizing a family of comparisons rather than one perfectly uniform counterfactual. That does not invalidate the estimate, but it limits how confidently a reader can translate it into the question, “How much better is this than the care available to me?”
The randomized design still matters. By assigning groups rather than merely observing people who choose to exercise, it reduces some forms of selection bias. Yet exercise trials cannot usually blind participants or instructors to whether movement is being delivered. Expectations, adherence, missing follow-up data, and differences in program contact can remain important. The dose review judged ten of its 17 studies to have low overall risk of bias and seven to have some concerns; all were judged low risk for outcome measurement and selective reporting, while some concerns most often involved deviations from the intended intervention and incomplete outcome data. Those judgments do not erase the pooled result, but they prevent a clean statistical estimate from being mistaken for perfect evidence.
The right reading is therefore layered. The review counted randomized, multi-week programs with a defined aerobic component and validated outcomes. It found average differences under specified comparator conditions. It did not count every kind of postpartum movement, did not use one universal diagnostic outcome, and did not establish a personal forecast. A headline such as “aerobic exercise helps postpartum depression” is a compressed summary of that chain. To understand the dose attached to it, we next have to examine how intensity, time, and frequency were converted into one apparently precise quantity.