Evidence Synthesis

Evidence synthesis is the process of systematically identifying, evaluating, and combining findings from multiple individual studies to draw broader, more reliable conclusions than any single study can provide. It encompasses a family of methods including systematic reviews, meta analyses, scoping reviews, rapid reviews, umbrella reviews, realist reviews, and qualitative evidence syntheses. Each method addresses different types of questions and uses different techniques for identifying, appraising, and integrating evidence. Systematic reviews and meta analyses of well conducted studies are often placed at the top of traditional evidence hierarchies, and evidence syntheses provide the foundation for clinical guidelines, health policy decisions, and research priority setting.

Individual studies, however well conducted, provide limited evidence. They are conducted in specific populations, settings, and time periods, and their results are subject to random error and systematic bias. Evidence synthesis addresses these limitations by bringing together evidence across studies, exploring sources of variation in results, and providing a comprehensive picture of what is known and unknown about a research question; when results can be pooled statistically in a meta analysis, it also increases statistical power and improves the precision of effect estimates. Clinical practice guidelines from organizations such as WHO, NICE, and professional societies are based on evidence syntheses. In the US, the FDA issued draft guidance in 2018 on how it evaluates meta analyses of randomized trials used to assess the safety of drugs and biological products. Research funders use evidence syntheses to identify gaps and set priorities. Without evidence synthesis, decision makers would face the impossible task of individually evaluating thousands of primary studies.

The evidence synthesis process varies by method but shares common elements. A clearly formulated question guides the entire process. A comprehensive search strategy, designed to minimize the risk of missing relevant studies, is conducted across multiple databases and supplementary sources. Studies are screened against predefined eligibility criteria, typically by two independent reviewers. In systematic reviews, included studies are critically appraised for risk of bias using validated tools (such as the Cochrane Risk of Bias tool, RoB 2, for randomized trials, or QUADAS-2 and its successor QUADAS-3, published in February 2026, for diagnostic accuracy studies); scoping reviews, which map the evidence, often omit this step.

Data are extracted from each study and organized for synthesis. In meta analysis, quantitative data are statistically combined to produce a pooled effect estimate, typically displayed in a forest plot. The I squared statistic and Q test assess statistical heterogeneity. Subgroup analyses and meta regression explore sources of variation. Sensitivity analyses test the robustness of results to different assumptions and decisions.

When statistical pooling is not appropriate due to excessive heterogeneity or incompatible data, narrative synthesis organizes and interprets findings without pooling. Qualitative evidence synthesis methods, such as meta ethnography and thematic synthesis, integrate findings from qualitative studies using interpretive techniques.

The certainty of the overall body of evidence is assessed using frameworks such as GRADE, which considers risk of bias, inconsistency, indirectness, imprecision, and publication bias. The final report follows a reporting guideline (PRISMA 2020 for systematic reviews, PRISMA-ScR for scoping reviews, MOOSE for meta analyses of observational studies) that supports transparency and reproducibility.

Evidence synthesis should be undertaken before conducting a new primary study, to determine what is already known and to identify genuine gaps in the evidence. It is essential for developing clinical guidelines, informing policy decisions, and prioritizing research funding. The choice of synthesis method depends on the question: meta analysis for quantitative pooling of intervention effects, diagnostic test accuracy review for diagnostic questions, qualitative evidence synthesis for understanding experiences and barriers, scoping reviews for mapping the evidence landscape, and rapid reviews when time constraints require an accelerated process. Evidence synthesis is not a substitute for primary research when no primary studies exist; it can only synthesize what has been studied.

A common error is conducting a meta analysis on studies that are too heterogeneous to be meaningfully combined, producing a pooled estimate that does not represent any real population. Another mistake is equating evidence synthesis with meta analysis; many evidence synthesis questions are best addressed through narrative synthesis, qualitative synthesis, or framework based approaches. Failing to search comprehensively (searching only one or two databases) risks missing relevant studies and biasing the synthesis. Using inappropriate quality assessment tools (such as applying a tool designed for RCTs to observational studies) produces invalid risk of bias assessments. Publishing an evidence synthesis without registering the protocol in advance (on PROSPERO or a similar registry) reduces transparency and invites suspicion of selective reporting. Finally, some researchers conduct evidence syntheses without adequate methodological expertise, producing reviews that look comprehensive but contain systematic errors in search strategy, data extraction, or statistical analysis.

In a hypothetical example, a review team synthesizes evidence on the effectiveness of school based physical activity programs for reducing childhood obesity. Their search across eight databases and two trial registries identifies 8,400 records, from which 62 cluster randomized and individually randomized trials meet inclusion criteria. Meta analysis of 48 trials reporting BMI z score shows a pooled mean difference of negative 0.10 (95% CI: negative 0.14 to negative 0.06) favoring the intervention. Subgroup analysis reveals that programs lasting more than 12 months produce larger effects (MD negative 0.15) than shorter programs (MD negative 0.06). GRADE assessment rates the certainty of evidence as moderate, downgraded for inconsistency. Findings like these can inform guidelines on physical activity promotion in schools.