Accounting for heterogeneity in meta-analysis using a multiplicative model-an empirical study

Area of Science:

• Biostatistics
• Medical Statistics
• Systematic Reviews

Background:

• Meta-analysis commonly employs random-effects models to address statistical heterogeneity.
• These models typically assume heterogeneity has an additive effect on effect size variance.
• Multiplicative heterogeneity models, though less common in medicine, are standard in fields like particle physics.

Purpose of the Study:

• To compare the performance and implications of additive versus multiplicative heterogeneity models in meta-analysis.
• To evaluate differences in goodness of fit and statistical outcomes between the two models.
• To provide guidance on selecting the appropriate meta-analysis model based on underlying assumptions.

Main Methods:

• A random sample of 448 meta-analyses was selected from the Cochrane Database of Systematic Reviews.
• The study involved comparing goodness of fit between additive random-effects models and multiplicative heterogeneity models.
• Statistical outcomes, including proximity to the null and confidence interval width, were analyzed.

Main Results:

• Differences in goodness of fit between the two models were generally modest.
• The multiplicative heterogeneity model tended to produce results closer to the null hypothesis.
• Confidence intervals generated by the multiplicative model were typically narrower compared to the additive model.

Conclusions:

• Both additive and multiplicative models rely on distinct assumptions regarding meta-analysis outcomes.
• The choice between models may hinge on the plausibility of the multiplicative model's assumption of a single, underlying effect size.
• Understanding these model assumptions is crucial for accurate interpretation of meta-analysis results.