History and Growth of Meta-analysis

  • Coined in 1976 by statistician Gene Glass
  • First modern meta-analysis published in 1904 by statistician Karl Pearson
  • Early examples of meta-analyses in occupational aptitude testing and agriculture
  • First model meta-analysis published in 1978 on psychotherapy outcomes
  • Growth of meta-analysis from 334 published meta-analyses in 1991 to 9,135 in 2014

Steps in a Meta-analysis

  • Formulate research question using PICO model
  • Search literature and select studies based on quality criteria
  • Decide inclusion of unpublished studies to avoid publication bias
  • Determine allowed dependent variables or summary measures
  • Select meta-analysis model and examine sources of between-study heterogeneity

Literature Search

  • Identify appropriate keywords and search limits for efficient database search
  • Use Boolean operators and search limits to assist the literature search
  • Choose the most appropriate sources from available databases
  • Search reference lists of eligible studies for additional sources (snowballing)
  • Detail search results in a PRISMA flow diagram to track study selection process

Data Collection

  • Use data collection form to collect standardized data from eligible studies
  • Collect effect size information, such as Pearson's statistic, for correlational data
  • Exclude partial correlations from meta-analysis due to potential inflation of relationships
  • Use plot digitizers as a last resort to collect data points from scatterplots
  • Collect study characteristics and measures of study quality to assess evidence

Meta-analysis Models and Approaches

  • Two types of evidence in meta-analysis: individual participant data (IPD) and aggregate data (AD)
  • Different approaches for synthesizing aggregate data
  • Statistical models for aggregate data: fixed effect model and random effects model
  • Issues with random effects model and alternative models like IVhet and quality effects model
  • Network meta-analysis methods and Bayesian framework
  • Frequentist multivariate framework for network meta-analysis
  • Types of meta-analysis and the use of robust methods
  • Aggregating IPD and AD using the generalized integration model (GIM)
  • Validation of meta-analysis results and challenges in meta-analysis
  • Publication bias and the file drawer problem

Meta-analysis Data Sources

Reference URL
Knowledge Graph