TextbookPublic Health & EpidemiologySystematic Reviews and Meta-Analysis

Systematic Reviews and Meta-Analysis

Systematic reviews use explicit, reproducible methods to identify, appraise, and synthesise all relevant research on a specific question, with meta-analysis providing a quantitative pooled estimate of effect that sits at the apex of the evidence hierarchy.

PLAB 1UKMLA0 questions

Key Facts

Systematic reviews are the highest level of evidence in the evidence hierarchy They use explicit, reproducible methodology to identify, select, appraise, and synthesise all relevant studies Meta-analysis is the statistical pooling of results from multiple studies to generate a single summary effect estimate Forest plot: graphical display showing individual study effects and pooled estimate; diamond represents overall effect Heterogeneity assessed by I² statistic: 0-25% low, 25-50% moderate, 50-75% substantial, >75% considerable Funnel plot: assesses publication bias; asymmetry suggests bias (small negative studies less likely published) PRISMA statement is the reporting standard for systematic reviews and meta-analyses Cochrane Library is the largest repository of systematic reviews of healthcare interventions

Overview

Key Facts

Systematic reviews and meta-analyses provide the most reliable evidence for clinical decision-making when conducted rigorously. They synthesise evidence from multiple studies, increasing statistical power and precision.

Steps in Conducting a Systematic Review

  1. Define research question (PICO format)
  2. Develop protocol (register on PROSPERO)
  3. Systematic literature search (multiple databases: MEDLINE, Embase, CENTRAL)
  4. Study selection (screening by ≥2 reviewers, inclusion/exclusion criteria)
  5. Data extraction (standardised forms)
  6. Quality assessment (risk of bias tool)
  7. Data synthesis (narrative and/or meta-analysis)
  8. Assessment of certainty of evidence (GRADE)
  9. Reporting (PRISMA)

Meta-Analysis Methods

  • Fixed-effect model: assumes all studies estimate the same underlying effect; appropriate when heterogeneity is low
  • Random-effects model: assumes true effects vary between studies; accounts for between-study heterogeneity; more conservative
  • Mantel-Haenszel method: common fixed-effect method for dichotomous outcomes
  • DerSimonian and Laird: common random-effects method

Assessing Quality

  • Risk of bias: assessed for each included study (Cochrane RoB 2 for RCTs, ROBINS-I for non-randomised)
  • Heterogeneity: clinical (population/intervention differences), methodological (design differences), statistical (I² and Chi² test)
  • Publication bias: funnel plot asymmetry, Egger test, trim and fill method

Clinical Presentation

Interpreting Forest Plots

  • Each study represented by a square (size proportional to weight) and horizontal line (95% CI)
  • Diamond at bottom represents pooled estimate (width = CI)
  • Vertical line at RR/OR = 1 (line of no effect)
  • If diamond does not cross line of no effect: statistically significant
  • I² value and p-value for heterogeneity reported

Interpreting Funnel Plots

  • Plots study effect size (x-axis) against study precision (y-axis, usually SE)
  • Symmetric funnel shape expected in absence of publication bias
  • Asymmetry (missing small negative studies) suggests publication bias

Subgroup and Sensitivity Analyses

  • Subgroup analysis: explores whether effect varies by patient/study characteristics
  • Sensitivity analysis: tests robustness (e.g. excluding high risk-of-bias studies)
  • Meta-regression: explores sources of heterogeneity using study-level covariates

Differential Diagnosis

FeatureSystematic ReviewNarrative ReviewMeta-Analysis
MethodsExplicit, reproducibleAuthor-selected, subjectiveStatistical pooling
SearchSystematic, comprehensiveNon-systematicPart of systematic review
Quality assessmentFormal (risk of bias tools)Informal/noneIncluded studies assessed
SynthesisNarrative ± quantitativeNarrative onlyQuantitative pooled estimate
Bias riskMinimisedHigh (selection bias)Depends on included studies
ReproducibilityHighLowHigh

Diagnosis / Investigation

Critical Appraisal (CASP Systematic Review Checklist)

  1. Did the review address a clearly focused question?
  2. Did the authors look for the right type of papers?
  3. Were all important relevant studies included?
  4. Did the authors assess the quality of included studies?
  5. If results were combined, was it reasonable to do so?
  6. What is the overall result?
  7. How precise are the results?
  8. Can results be applied to the local population?
  9. Were all important outcomes considered?
  10. Are the benefits worth the harms and costs?

Key Statistical Concepts

  • : percentage of variability due to heterogeneity rather than chance; >50% = substantial
  • Chi² (Q) test: tests whether observed differences in results are compatible with chance alone
  • Prediction interval: range of effects expected in a new study (wider than CI of pooled estimate)
  • GRADE certainty: High, Moderate, Low, Very Low

Management

Using Systematic Reviews in Practice

  • Check Cochrane Library for existing reviews on clinical questions
  • Assess GRADE certainty of evidence before applying to patients
  • Consider applicability to your patient population
  • NNT from meta-analysis may be the most reliable estimate available
  • Living systematic reviews: continuously updated as new evidence emerges

Limitations

  • Only as good as included studies ('garbage in, garbage out')
  • Publication bias: negative studies less likely published
  • Heterogeneity may limit pooling
  • Ecological fallacy: pooled results may not apply to individual patients
  • Time lag: may not include most recent evidence

Prognosis

  • Systematic reviews and meta-analyses form the basis of clinical guidelines (NICE, WHO, Cochrane)
  • They provide the most precise estimates of treatment effects through pooled analysis
  • Updated/living reviews improve timeliness of evidence synthesis
  • Network meta-analyses (comparing multiple treatments simultaneously) increasingly used for guideline development
  • Individual patient data meta-analyses provide the most detailed evidence but require data sharing

Other Relevant Information

I² Heterogeneity Interpretation

I² ValueHeterogeneity LevelAction
0-25%LowFixed-effect model appropriate
25-50%ModerateConsider random-effects
50-75%SubstantialRandom-effects; explore sources
>75%ConsiderableMay not be appropriate to pool; explore/explain

GRADE Certainty of Evidence

LevelMeaning
HighVery confident effect estimate is close to true effect
ModerateModerately confident; true effect likely close to estimate
LowLimited confidence; true effect may be substantially different
Very LowVery little confidence; true effect likely substantially different