Meta-analysis Basics: Effect Size, Heterogeneity and Interpretation

A practical introduction to pooled estimates, confidence intervals, I², tau-squared and why a meta-analysis is more than producing a forest plot.

A meta-analysis combines comparable quantitative evidence

Meta-analysis statistically synthesises estimates from multiple studies. The exact effect measure depends on the outcome and design: examples include risk ratios, odds ratios, mean differences, standardised mean differences and proportions. Studies should not be pooled merely because they address a broadly similar topic; the effect measure and underlying question must be sufficiently comparable.

The pooled estimate needs uncertainty

A meta-analysis reports a pooled effect with a confidence interval. The interval communicates statistical uncertainty around the synthesis. A precise pooled estimate can still be misleading if the included studies are biased, measure different constructs or represent very different populations.

Heterogeneity describes between-study variation

Cochran’s Q tests whether observed variability is more than expected from sampling error, while I² expresses the proportion of observed variability associated with heterogeneity rather than chance. Tau-squared estimates the between-study variance in a random-effects model. These statistics should be interpreted together with the clinical or methodological differences between studies.

Random effects does not solve incompatibility

A random-effects model allows the underlying effect to vary between studies, but it does not make fundamentally different studies comparable. Researchers should inspect populations, thresholds, follow-up periods, instruments and study designs before pooling. Subgroup or sensitivity analyses may help investigate important sources of heterogeneity.

Prediction intervals can improve interpretation

Where appropriate, a prediction interval estimates the range in which the true effect of a future comparable study might lie. With substantial heterogeneity, this can be more informative than focusing only on the average pooled effect. It also makes clear that an average effect may not represent every setting.

Quality and sensitivity matter

Meta-analysis should be accompanied by risk-of-bias assessment and sensitivity analysis. Results may change when high-risk studies, reconstructed estimates or different statistical assumptions are used. Transparent reporting helps readers judge how robust the pooled conclusion really is.