The interpretation of random-effects meta-analysis in decision models

Area of Science:

• Biostatistics
• Health Economics
• Clinical Trial Analysis

Background:

• Random-effects models are common in meta-analysis for summarizing heterogeneous treatment effects.
• Decision models rely on meta-analysis results to inform healthcare choices.
• The interpretation of meta-analysis models can influence decision model outcomes.

Purpose of the Study:

• To investigate how different interpretations of random-effects models in meta-analysis affect decision model results.
• To explore alternative approaches for representing treatment effect heterogeneity in future implementations.
• To demonstrate the impact of model choice on expected net benefits.

Main Methods:

• Analysis of sources of variation in meta-analysis (outcome definition, patient groups, protocols, implementation).
• Comparison of different models for relating observed heterogeneity to future effect sizes.
• Application of a probabilistic, Bayesian posterior framework for illustration.

Main Results:

• The mean treatment effect from random-effects meta-analysis is often an inadequate representation of future efficacy.
• Alternative models lead to different computations and expected net benefits, especially with non-linear efficacy.
• Parameter uncertainty and heterogeneity require careful consideration in modeling.

Conclusions:

• Modelers should consider the predictive distribution of future treatment effects or assume a distribution of effects.
• Standard interpretation of random-effects meta-analysis may not suffice for robust decision modeling.
• Accurate representation of heterogeneity is crucial for reliable health economic evaluations.