Hierarchical Bayesian approaches for detecting inconsistency in network meta-analysis.

Area of Science:

• Biostatistics
• Medical Informatics
• Health Services Research

Background:

• Network meta-analysis (NMA) integrates direct and indirect evidence for treatment comparisons.
• Bayesian NMA methods are increasingly complex and popular.
• Inconsistency arises when direct and indirect evidence in NMA diverge.

Purpose of the Study:

• To propose a novel arm-based random effects model for detecting inconsistency in NMA.
• To introduce discrepancy factors for quantifying evidence of inconsistency.
• To offer an alternative to contrast-based (CB) models for inconsistency detection.

Main Methods:

• Developed an arm-based random effects model for NMA.
• Utilized fixed effects to detect discrepancies between direct and indirect evidence.
• Employed random effects to flag potentially extreme trials.
• Defined novel discrepancy factors to characterize inconsistency.

Main Results:

• The proposed model effectively detects discrepancies in direct and indirect evidence.
• Discrepancy factors provide a novel measure of NMA inconsistency.
• Comparison with loop-based CB methods on real and simulated data shows comparable or superior performance.
• The new approach offers powerful inconsistency detection capabilities.

Conclusions:

• The novel arm-based random effects model provides a powerful and effective method for detecting inconsistency in network meta-analysis.
• This approach offers a valuable alternative to existing contrast-based methods.
• The introduction of discrepancy factors enhances the characterization of evidence divergence in NMA.