Systematic Review and Meta-Analysis

A systematic review is a structured synthesis of all the studies that address a defined question, and a meta-analysis is the statistical pooling of their results into a single estimate. Together they sit at the top of the evidence hierarchy for questions of intervention effect, and within drug information they are the principal tools for summarising what the literature collectively says about a medicine. This entry treats the topic within clinical pharmacy and literature evaluation; a parallel epidemiology entry exists.

Definition

A systematic review uses explicit, reproducible methods to identify, appraise, and synthesise all studies relevant to a defined question; a meta-analysis is the statistical combination of comparable study results within such a review to produce a pooled effect estimate.

Scope

This topic covers the systematic-review process — protocol, comprehensive search, screening, risk-of-bias assessment, and synthesis — and the meta-analytic methods used to pool results, including fixed-effect and random-effects models and the assessment of heterogeneity. It is a methodological and reference topic about evidence synthesis, not a source of treatment instructions.

Core questions

• How is a review question and protocol defined to make the search reproducible?
• How are studies identified, screened, and appraised for risk of bias?
• When can results be pooled, and what model should be used?
• How is heterogeneity between studies measured and interpreted?
• How is the resulting synthesis reported and its quality appraised?

Key concepts

• Pre-registered protocol and reproducible search
• Study screening and selection
• Pooled effect estimate
• Fixed-effect versus random-effects models
• Heterogeneity and the I-squared statistic
• Publication bias
• Reporting and appraisal standards (PRISMA, AMSTAR 2)

Mechanisms

A systematic review follows a pre-specified protocol: a comprehensive search of multiple databases, duplicate screening against explicit criteria, risk-of-bias assessment of included studies, and synthesis. Where studies are comparable, meta-analysis pools their effect estimates, weighting each by its precision. A fixed-effect model assumes one common true effect, whereas a random-effects model — formalised by DerSimonian and Laird — assumes the true effect varies across studies and incorporates that between-study variance. Heterogeneity is quantified by statistics such as I-squared, introduced by Higgins and colleagues, to express the proportion of variation due to genuine differences rather than chance. Publication bias, in which studies with positive results are more likely to appear, is examined because it can distort the pooled estimate. PRISMA standardises how the whole process is reported, and AMSTAR 2 appraises the methodological quality of the completed review.

Clinical relevance

Systematic reviews and meta-analyses provide the summarised evidence behind formulary decisions, guideline recommendations, and many drug information answers. This topic describes how that evidence is synthesised and supports its critical reading; it is a reference resource and not a basis for individualised diagnostic or treatment decisions.

Evidence & guidelines

Evidence synthesis is governed by established standards: the PRISMA statement (2009, updated 2020) for reporting systematic reviews and meta-analyses, and the AMSTAR 2 instrument for appraising their methodological quality. Heterogeneity quantification via I-squared and random-effects pooling via the DerSimonian-Laird method are standard analytic components.

History

Quantitative pooling of study results has roots in early twentieth-century statistics, and the term meta-analysis was coined in the 1970s. DerSimonian and Laird's 1986 random-effects method became a workhorse of medical meta-analysis, and Higgins and colleagues' 2003 I-squared statistic standardised the description of heterogeneity. The PRISMA statement, first issued in 2009 and updated in 2020, then formalised transparent reporting of systematic reviews.

Debates

Fixed-effect versus random-effects pooling

The choice of model embodies an assumption about whether studies estimate a single common effect or a distribution of effects; random-effects models give more weight to smaller studies and wider intervals, and the appropriate choice depends on the heterogeneity and the inferential goal.

Key figures

• David Moher
• Matthew Page
• Julian Higgins
• Rebecca DerSimonian
• Nan Laird

Seminal works

• moher-2009-prisma
• dersimonian-1986
• higgins-2003
• page-2021-prisma

Frequently asked questions

What is the difference between a systematic review and a meta-analysis?

A systematic review is the structured process of finding, appraising, and synthesising all relevant studies; a meta-analysis is the optional statistical step within it that pools comparable results into a single estimate.

What does the I-squared statistic tell me?

It estimates the proportion of the total variation across studies that is due to real heterogeneity rather than chance, helping judge whether pooling the results into a single estimate is appropriate.