A comparison of two models for detecting inconsistency in network meta-analysis

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Health Research Methodology

Background:

• Network meta-analysis (NMA) is increasingly used for synthesizing evidence from multiple treatment comparisons.
• Ensuring consistency between direct and indirect evidence in NMA is vital for reliable clinical decision-making.
• Inconsistency detection is a critical step in NMA to validate the use of results for clinical guidance.

Purpose of the Study:

• To comprehensively review and explore the relationship between the design-by-treatment interaction model and side-splitting models for inconsistency detection in NMA.
• To compare the performance of these models under different data structures using a frequentist approach.
• To provide practical guidance on selecting appropriate models for inconsistency assessment in NMA.

Main Methods:

• The study reviews two primary models for NMA inconsistency: the design-by-treatment interaction model and side-splitting models.
• A frequentist approach is employed to analyze these models, treating NMA data structures as missing data.
• Analytical and numerical studies are conducted to explore the relationship and performance of the models.

Main Results:

• Side-splitting models are identified as specific instances of the design-by-treatment interaction model, applicable under certain data structures or additional assumptions.
• The design-by-treatment interaction model demonstrates superior and robust performance in detecting inconsistency across diverse data structures compared to side-splitting models.
• The study confirms the relationship between the models through data structure parameterization and analysis.

Conclusions:

• The design-by-treatment interaction model is recommended for general inconsistency detection in NMA, especially when the location of inconsistency is unknown.
• Side-splitting models can be valuable supplementary tools for detailed inconsistency assessment, particularly in smaller networks or when examining local inconsistencies.
• This research provides a framework for understanding and applying different NMA inconsistency detection methods for improved evidence synthesis.