Statistical Primer: heterogeneity, random- or fixed-effects model analyses?†
Key result
Random-effects models account for heterogeneity in meta-analyses by assuming different underlying true effects across trials, utilizing statistics like Q, T2, and I2 to quantify dispersion.
Design
Review
Authors
Discussion
Loading...
Member takes
Implication
Aids model selection in heterogeneous meta-analyses; leaves open standardization of I² thresholds in cardiovascular syntheses.
Key Points
- Explain the concept of statistical heterogeneity in meta-analyses and provide guidance on choosing between fixed-effects and random-effects models.
- Methodological review evaluating statistical principles for synthesizing trial outcomes with differing underlying treatment effects.
- Comparison of fixed-effects assumptions against random-effects models incorporating both within-study and between-study variance components.
- Five core metrics quantify heterogeneity: the Q statistic and associated p-value test for departure from homogeneity, T2 and T calculate absolute variance magnitude, and I2 reflects the proportion of total variation driven by heterogeneity.
- Random-effects models account for distributed true effects across trials, yielding wider confidence intervals that incorporate between-study uncertainty compared with fixed-effects models.
Cite This Study
Barili et al. (2018) reported a review. Random-effects models account for heterogeneity in meta-analyses by assuming different underlying true effects across trials, utilizing statistics like Q, T2, and I2 to quantify dispersion.