How Do You Synthesize Evidence Across Sex or Gender Groups?

Should evidence from different sex or gender groups be combined?

You may encounter studies with very different participant compositions. Some include mostly women, others mostly men, some report several gender categories, and many provide only an overall effect without disaggregated results.

That creates a practical synthesis problem. Should all of these studies contribute to one estimate? Should findings be separated by sex or gender? And if one group appears to benefit more than another, when is that difference credible?

The answer depends on more than demographic composition. Sex or gender should influence the synthesis when there is a defensible reason that the characteristic could alter the effect, outcome, experience, exposure, implementation, or applicability being investigated.

Combine evidence when appropriate, but preserve meaningful differences

Sex and gender can matter, but not necessarily in the same way

Sex and gender are often entered into a dataset as if they were interchangeable demographic variables. For evidence synthesis, that shortcut can become consequential.

Depending on the research question, sex-related characteristics may be relevant through biological pathways, while gender may operate through social roles, identities, behaviors, expectations, access to resources, exposure patterns, or institutional treatment. These pathways can overlap, and individual studies do not always define or measure them clearly.

Your synthesis should therefore preserve the terminology and constructs actually measured by the primary studies rather than silently converting one into the other.

Begin with a plausible pathway, not a routine subgroup analysis

A subgroup analysis is most informative when there is a substantive reason to expect an effect to differ. Cochrane guidance recommends that potential effect modifiers be justified and, where possible, specified before results are examined. Conducting many post hoc subgroup analyses increases the opportunity to find apparently interesting differences by chance.

The relevant pathway depends on the topic. Pharmacological effects might plausibly vary through biological characteristics. A workplace intervention could interact with gendered occupational roles. Access to health services may reflect gender-related social or institutional barriers. In another research question, neither sex nor gender may plausibly modify the effect at all.

Demographic importance does not automatically imply effect modification.

Separate prognostic differences from effect modification

One group may have a different baseline risk of an outcome without responding differently to an intervention. Cochrane distinguishes prognostic factors, which predict an outcome, from effect modifiers, which influence how an intervention affects that outcome.

Suppose an outcome occurs more frequently in one group before treatment. Even if an intervention produces the same relative effect in both groups, their absolute benefits or harms can differ because they started at different baseline risks.

This distinction affects interpretation. “The groups have different outcomes” and “the intervention works differently between the groups” are not equivalent claims.

Compare the groups directly

Suppose an intervention has a statistically significant effect among women but not among men. That does not establish that its effect differs by sex.

The two effect estimates might be almost identical while one subgroup simply has fewer participants and a wider confidence interval. Cochrane explicitly cautions against comparing statistical significance within separate subgroups. The appropriate analysis evaluates the difference between subgroup effects, such as through a formal interaction test when suitable.

Participant composition is not the same as subgroup evidence

Suppose Study A contains 80% women and Study B contains 80% men. If Study A reports a larger effect, you cannot safely conclude that women benefit more.

The studies may also differ in intervention delivery, participant age, setting, risk of bias, follow-up, baseline risk, or numerous other characteristics. Comparisons based on the proportion of participants in each study are study-level observational comparisons and may be confounded.

This is one reason within-study comparisons are generally more persuasive than comparisons between sets of studies. Cochrane notes that replicated within-study relationships can add confidence to subgroup findings.

Individual participant data may answer questions that published aggregates cannot

Published articles frequently provide only an overall intervention effect even when multiple groups participated. That can make subgroup synthesis impossible from aggregate reports.

Individual participant data meta-analysis can address some of this limitation because interactions between participant characteristics and intervention effects can be estimated within individual studies and then synthesized. Cochrane identifies this as the most reliable, and sometimes the only practical, approach for investigating whether intervention effects vary by participant characteristics such as sex.

That does not mean individual participant data are required for every review. It means you should recognize when published aggregate evidence cannot answer the subgroup question you would like to ask.

Absence of subgroup evidence is not evidence of equal effects

If studies do not report disaggregated results, you cannot conclude that groups respond identically. You may simply lack the data needed to assess whether they differ.

The reverse is equally important. Sparse subgroup evidence should not be stretched into claims of meaningful differences.

Cochrane's equity guidance recommends considering sex and gender among potentially relevant population characteristics while emphasizing prespecification and justification of subgroup analyses. It also recognizes that analyses may sometimes be impossible because the necessary data are insufficient.

Think about representation as well as interaction

Even if there is no demonstrated subgroup effect, the evidence base may disproportionately represent one group. That raises a different question: how directly does the synthesis apply to people who were scarcely represented?

This is an applicability issue rather than proof of effect modification. The distinction matters because “we have no evidence that effects differ” is not equivalent to “we have strong evidence that effects are the same.”

When representation is poor and there is a credible reason the finding might differ, the broader question becomes whether evidence from another population is directly relevant to the group of interest.

Context can interact with sex or gender

Sex- or gender-related patterns may not be constant across countries, institutions, cultures, or resource environments. A gender-related barrier operating in one setting may be weaker, stronger, or qualitatively different elsewhere.

Consequently, apparent subgroup differences sometimes need to be interpreted together with meaningful contextual differences between countries rather than treated as universal characteristics of a group.

When different subgroup results do not necessarily mean different effects

This distinction prevents one of the most common subgroup-analysis errors. Different P values can reflect different amounts of information rather than different underlying effects.

Common mistakes when interpreting sex- or gender-specific evidence

Preserve the distinction between representation and effect modification

When planning the synthesis, decide what role sex or gender could plausibly play. Then extract enough information to determine who was studied, what construct was measured, and whether group-specific effects are actually available.

The final synthesis may legitimately conclude that the overall evidence supports an effect while evidence about differences between sex or gender groups remains uncertain. Those conclusions answer different questions, and keeping them separate makes the review more informative.

Before synthesizing evidence across sex or gender groups, check:

Questions about synthesizing evidence across sex or gender groups

Do not force either sameness or difference onto the evidence

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.