Simulation and data-generation for random-effects network meta-analysis of binary outcome.

Area of Science:

• Statistics
• Biostatistics
• Medical Informatics

Background:

• Statistical method performance is often assessed using simulation studies.
• Existing data-generating models (DGMs) for network meta-analysis of binary data are limited, often excluding multi-arm trials or only accommodating fixed-effect models.

Purpose of the Study:

• To propose a novel simulation framework for random-effect network meta-analyses (NMA) of binary data.
• To extend existing DGMs to include multi-arm trials, a common scenario in NMA.
• To address the restrictive assumptions of current pairwise DGMs when applied to random-effects NMA.

Main Methods:

• Developed a simulation framework based on pairwise meta-analysis data generation.
• Modified an existing DGM to be applicable to random-effects NMA settings.
• Derived a simulation procedure utilizing odds ratios as the effect measure.
• Evaluated the proposed procedure using both synthetic datasets and an empirical example.

Main Results:

• The proposed framework successfully simulates random-effect network meta-analyses including multi-arm trials with binary outcomes.
• The modified approach overcomes limitations associated with existing DGMs in NMA.
• The use of odds ratios as the effect measure provides a viable alternative for simulation.

Conclusions:

• The developed simulation framework enhances the evaluation of statistical methods in network meta-analysis.
• This approach provides a more flexible and realistic simulation of random-effect NMA with binary data.
• The findings support the use of this method for assessing the performance of statistical models in complex NMA scenarios.