Random main effects of treatment: A case study with a network meta-analysis

Area of Science:

• Biostatistics
• Medical Research Methodology
• Evidence Synthesis

Background:

• Traditional network meta-analyses often assume fixed treatment effects.
• Large numbers of treatments in meta-analyses present analytical challenges.
• Existing random-effect models primarily address treatment-by-trial interactions or trial main effects.

Purpose of the Study:

• To propose and evaluate a hierarchical model for random main effects of treatments in network meta-analysis.
• To explore the utility of random-effect models for within-trial variances.
• To demonstrate the application of this approach using a real-world example.

Main Methods:

• Development of a hierarchical modeling framework to treat individual treatment effects as random realizations from a normal distribution.
• Application of the model to a network meta-analysis involving 44 treatments across 10 trials.
• Investigation of incorporating random-effect models for within-trial variances.

Main Results:

• The hierarchical model successfully produced shrunk (regressed toward the overall mean) estimates for individual treatment effects.
• This method offers improved precision for effect estimates when dealing with numerous treatments.
• Graphical representations of results were developed to aid interpretation.

Conclusions:

• Modeling the main effect of treatment as random is a viable and beneficial approach in network meta-analysis, especially with a large number of treatments.
• The proposed hierarchical method enhances the estimation of individual treatment effects.
• Further consideration of random-effect models for within-trial variances is warranted.