May 21, 2002Statistics in Medicine
Quantifying heterogeneity in a meta‐analysis
Key Points
- To develop mathematical measures for evaluating the impact of heterogeneity on meta-analyses that remain independent of the number of studies and the chosen treatment effect metric.
- Derived mathematical formulations for three statistics: H (normalized chi-squared statistic), R (standard error ratio of random to fixed effects models), and I-squared (proportion of total variation due to heterogeneity).
- Evaluated the properties, confidence intervals, and interpretability of the proposed metrics across five published meta-analytic example datasets exhibiting varying degrees of heterogeneity.
- Established I-squared as an intuitive transformation of H that quantifies the percentage of between-study variability in meta-analytic estimates.
- Identified that H and I-squared overcome the limitations of standard significance tests for heterogeneity and can be directly calculated from standard published summary statistics.
Authors
Discussion
No takes yet. Share an insight, caveat, or question.
Implication
Methodological study derives scale-independent variance metrics in meta-analyses, indicating that the I-squared statistic reliably quantifies heterogeneity regardless of study count or effect scale.
Cite This Study
Higgins et al. (2002) studied this question.