Systematic Review and Meta-Analysis

A systematic review with meta-analysis combines two methods: a structured, reproducible review that gathers and appraises all eligible studies on a question, and a statistical procedure that pools their results into a single weighted estimate of effect. The review controls the bias of study selection; the meta-analysis quantifies the combined signal and the variability around it. Together they form the prototypical method of evidence-based intervention research.

Definition

A systematic review with meta-analysis is a review that uses explicit, reproducible methods to identify and appraise all eligible studies on a question and then statistically combines their effect estimates into a pooled estimate, characterising both the central effect and the heterogeneity among studies.

Scope

This topic covers the conduct of a systematic review with quantitative pooling: protocol and eligibility, searching and screening, risk-of-bias appraisal, the fixed-effect versus random-effects choice, weighting, heterogeneity, and the reporting and certainty standards that govern the result. It is a methodological reference, not clinical guidance.

Core questions

• Are the included studies similar enough to justify pooling their results?
• Should a fixed-effect or random-effects model be used?
• How much do the study results vary beyond chance (heterogeneity)?
• How is the risk of bias within studies reflected in the pooled estimate?
• How certain is the combined evidence overall?

Key concepts

• Protocol and pre-specified eligibility
• Effect measure (e.g. risk ratio, odds ratio, mean difference)
• Inverse-variance weighting
• Fixed-effect versus random-effects model
• Heterogeneity and the I-squared statistic
• Forest plot
• Risk-of-bias assessment
• Certainty rating (GRADE)

Mechanisms

After eligible studies are identified and appraised, each study contributes an effect estimate with a measure of precision. Meta-analysis combines these by weighting each study, typically by the inverse of its variance, so that larger and more precise studies count more. A fixed-effect model assumes a single common true effect; a random-effects model assumes the true effect varies across studies and incorporates that between-study variance. The spread of true effects beyond sampling error is heterogeneity, often summarised by the I-squared statistic, and the pooled result is conventionally displayed in a forest plot. Reporting follows PRISMA, within-study bias is appraised with tools such as the Cochrane risk-of-bias tool, and the certainty of the pooled evidence is rated with GRADE (higgins-handbook-2019; page-2021-prisma; higgins-2011-rob; guyatt-2008-grade).

Clinical relevance

Meta-analyses of randomised trials provide much of the quantitative evidence cited in guidelines and health technology assessments. Critically reading a meta-analysis — checking what was pooled, how heterogeneity was handled, and how certain the evidence is rated — is part of evidence appraisal. The method describes how pooled estimates are produced; it does not prescribe treatment for an individual.

Evidence & guidelines

Conduct and reporting are standardised: PRISMA 2020 (with its 2009 explanation-and-elaboration lineage) governs reporting, the Cochrane Handbook describes accepted methods, the Cochrane risk-of-bias tool structures within-study appraisal, and GRADE rates certainty across the body of evidence (page-2021-prisma; liberati-2009; higgins-handbook-2019; higgins-2011-rob; guyatt-2008-grade).

History

Statistical combination of studies dates to early twentieth-century agricultural and medical statistics, and the term meta-analysis was coined in 1976. The systematic review consolidated the surrounding process during the 1990s, especially through the Cochrane Collaboration. Reporting standards evolved from QUOROM to PRISMA (2009, updated 2021), heterogeneity statistics such as I-squared were popularised, and GRADE supplied a structured certainty framework, together defining the modern method (page-2021-prisma; higgins-handbook-2019).

Debates

When is heterogeneity too great to pool?

Combining clinically or statistically dissimilar studies can produce a misleading average; reviewers debate thresholds and whether to favour random-effects models, subgroup analysis, or a narrative synthesis instead of pooling.

Key figures

• Julian Higgins
• David Moher
• Matthew Page
• Gordon Guyatt
• Cynthia Mulrow

Seminal works

• page-2021-prisma
• higgins-handbook-2019
• guyatt-2008-grade

Frequently asked questions

Does every systematic review include a meta-analysis?

No. When studies are too dissimilar in population, intervention, or outcome, pooling can mislead, and the review reports a structured narrative synthesis instead of a single combined estimate.

What is a forest plot?

A forest plot displays each study's effect estimate and confidence interval alongside the pooled estimate, making the contribution of each study and the overall result visible at a glance.