Using structural equation modeling for network meta-analysis.

Area of Science:

• Statistical modeling
• Biostatistics
• Evidence synthesis

Background:

• Network meta-analysis (NMA) integrates multiple treatment comparisons, overcoming limitations of pairwise meta-analysis.
• Current NMAs utilize Bayesian hierarchical linear models or frequentist generalized linear mixed models.
• Structural Equation Modeling (SEM) is a statistical method for causal relations, adaptable for complex random effect structures in NMA.

Purpose of the Study:

• To demonstrate the implementation of network meta-analysis within the Structural Equation Modeling (SEM) framework.
• To showcase SEM's flexibility for standard fixed and random effect NMA models.
• To explore a novel NMA approach using the unrestricted weighted least squares (UWLS) method within SEM.

Main Methods:

• Utilized an example dataset with 26 studies comparing three treatments (A, B, C) for preventing first bleeding in liver cirrhosis patients.
• Implemented standard fixed and random effect NMA models using SEM.
• Applied SEM to a new NMA approach based on the unrestricted weighted least squares (UWLS) method.

Main Results:

• SEM produced results consistent with traditional NMA for fixed and random effect models.
• UWLS models in SEM yielded identical point estimates to fixed effect models but wider confidence intervals.
• The UWLS model with unique variance adjustment better reflected pairwise comparison heterogeneity.

Conclusions:

• SEM offers a highly flexible framework for conducting univariate and multivariate meta-analyses.
• SEM's potential for advanced meta-analysis applications warrants further exploration.
• This study highlights SEM as a powerful tool for sophisticated evidence synthesis.