Meta-Analysis Guide

Conduct meta-analyses to synthesize quantitative research findings across studies. Covers the complete process from protocol development through statistical synthesis and reporting.

Usage

Describe the research question you want to synthesize and the studies you've collected. The guide walks through effect size extraction, statistical synthesis, heterogeneity analysis, and publication bias assessment.

Parameters

• Effect type: Mean difference, Odds ratio, Risk ratio, Correlation, or Standardized mean difference
• Model: Fixed-effect, Random-effects, or Mixed-effects
• Software: R (metafor/meta), Stata, RevMan, or Python
• Stage: Protocol, Data extraction, Analysis, or Reporting

Examples

1. Treatment Efficacy: Random-effects meta-analysis of 15 RCTs comparing a therapy to waitlist control — SMD extraction, forest plot, I² heterogeneity, subgroup analysis by treatment duration.

1. Diagnostic Accuracy: Bivariate meta-analysis of sensitivity and specificity for a diagnostic test across 20 studies, with SROC curve and heterogeneity investigation.

1. Dose-Response Analysis: Non-linear dose-response meta-analysis examining the relationship between exercise volume and depression reduction across observational studies.

1. Network Meta-Analysis: Compare 5 interventions simultaneously using network meta-analysis when direct head-to-head comparisons are limited — transitivity assumption checking and SUCRA rankings.

Guidelines

• Protocol is registered (PROSPERO) before data extraction begins
• Inclusion criteria clearly specify PICO/PECO elements for study selection
• Effect sizes are extracted or calculated consistently across all included studies
• Heterogeneity is assessed with I², Q-test, and prediction intervals
• Random-effects models are used when studies differ in population, setting, or implementation
• Subgroup analyses and meta-regression explore sources of heterogeneity
• Publication bias is assessed with funnel plots, Egger's test, and trim-and-fill
• Sensitivity analyses test robustness to study removal, quality thresholds, and model choice
• Forest plots display individual study effects with confidence intervals and pooled estimate
• Reporting follows PRISMA 2020 guidelines with all required checklist items