Network meta-analysis of rare events using penalized likelihood regression

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Health Research Methods

Background:

• Network meta-analysis (NMA) for rare events is underexplored.
• Traditional inverse-variance NMA can yield biased estimates with rare events due to poor normal approximations.
• Existing methods like Mantel-Haenszel NMA or noncentral hypergeometric distribution have limitations.

Purpose of the Study:

• To propose a novel common-effect NMA approach for analyzing rare events, including zero events.
• To address limitations of existing methods in synthesizing data with low event frequencies.
• To provide a robust method for meta-analysis in challenging data scenarios.

Main Methods:

• Implementation of Firth's penalized likelihood function within the logistic NMA model.
• A two-stage approach to incorporate heterogeneity using a multiplicative overdispersion term.
• Simulation studies to evaluate performance across various scenarios.

Main Results:

• The penalized likelihood NMA method demonstrates consistent performance and reduced bias compared to other methods.
• The approach effectively handles networks with extremely low or zero events without data imputation or exclusion.
• Simulations confirmed its reliability across tested scenarios.

Conclusions:

• The penalized likelihood NMA is a promising method for rare event binary outcomes.
• It is particularly suitable for networks with few studies and low control group event risks.
• This approach offers improved accuracy and reduced bias in meta-analyses of rare events.