Network Meta-Analysis with Class Effects: A Practical Guide and Model Selection Algorithm.

Area of Science:

• Biostatistics
• Health Services Research
• Clinical Epidemiology

Background:

• Network meta-analysis (NMA) synthesizes evidence from multiple randomized controlled trials.
• Class effect NMA models group interventions into classes to improve evidence synthesis, especially with sparse data or disconnected networks.
• Existing literature lacks a comprehensive guide for class effect NMA models, including their assumptions, implementation, and model selection.

Purpose of the Study:

• To provide a comprehensive modeling framework for class effect NMA.
• To propose a systematic approach for model selection in class effect NMA.
• To offer practical guidance for implementing class effect NMA using the `multinma` R package.

Main Methods:

• Described hierarchical NMA models with random/fixed treatment effects and exchangeable/common class effects.
• Detailed methods for testing assumptions of heterogeneity, consistency, and class effects.
• Proposed a model selection strategy and assessed model fit.

Main Results:

• Developed a modeling framework and practical guidance for class effect NMA.
• Illustrated the approach with a large NMA of 41 interventions across 17 classes for social anxiety.
• Provided implementation details using the `multinma` R package.

Conclusions:

• Class effect NMA models offer a valuable approach for synthesizing evidence across intervention classes.
• The proposed framework and R package facilitate the application and selection of appropriate class effect NMA models.
• This work addresses a critical gap in the literature, enabling more robust evidence synthesis for treatment recommendations.