Key Terms

Meta-analysis

Meta-analysis is the statistical method for combining both qualitative and quantitative outcomes from multiple studies to draw statistically stronger conclusions. Researchers may want to use meta-analysis because of the greater statistical power it creates by means of having a larger number of subjects and results than any individual study. Meta-analysis also provides estimates of treatment effects, which can identify any asymmetries in the studies, such as publication bias and heterogeneity.

​​Univariate vs. Multivariate meta-analysis

  • Univariate meta-analysis (UMA): Univariate meta-analysis is the process of aggregating multiple studies to better understand the effect of one outcome.
  • ​Multivariate meta-analysis (MMA): Multivariate meta-analysis aggregates multiple studies to better understand the effect of multiple outcomes.

​Fixed effects vs. random effects meta-analysis model

  • Fixed effects meta-analysis model: The fixed-effect meta-analysis model is one of the two main methods of meta-analysis.
    • This model assumes that the studies included in the meta-analysis come from a single homogeneous population and the observed heterogeneity is due to within-study variations.
    • Under these assumptions, an overall estimate of the effect(s) can be calculated by simply averaging the studies’ estimates.
  • Random-effects meta-analysis model: The random-effects meta-analysis model is the other main method of meta-analysis.
    • This model relies on the assumption that study effect estimates are often more variable than assumed in the fixed effects model, where observed heterogeneity is attributed to both within-study and between-study variation.
    • This variability is captured by a normally distributed random variable with mean 0 and variance as the between-study heterogeneity.

Challenges in meta-analysis

  • Heterogeneity in meta-analysis: There are two main types of heterogeneity that may occur in meta-analysis: clinical heterogeneity and statistical heterogeneity.
    • Clinical heterogeneity arises from the way the study was conducted or designed, such as differences in patients, type of treatment intervention(s), and outcome definition(s).
    • Statistical heterogeneity results from differences in the methods used to analyze the results of the studies.
    • Heterogeneity may contribute to biased estimates and it may be identified using the forest plot visualization or statistical tests for heterogeneity such as Cochran’s Q test or the Index I .
  • Small study effect:
    • The small study effect arises from the bias toward publishing studies with positive outcomes.
    • This increased likelihood that small studies are published if their results are significant may skew the conclusions drawn from meta-analysis as the studies being pooled largely represent studies selected for their favorable results and larger effects.
  • Publication bias:
    • Publication bias (PB) is the phenomenon that occurs when the outcome of a study influences the decision to publish it or not; typically, studies having favorable results are more likely to be published or submitted for approval than those with unfavorable results.
    • The implications of PB in meta-analysis include drawing biased inferences from pooling potentially biased studies together, which may lead to observing an effect that does not reflect the true effect.
    • Publication bias may be identified using the funnel plot visualization tool, the Trim and Fill method, and other techniques such as the Copas selection model and sensitivity analysis.