Influence Analyses of "Designs" for Evaluating Inconsistency in Network Meta-Analysis.

Area of Science:

• Biostatistics
• Evidence Synthesis
• Health Research Methodology

Background:

• Network meta-analysis (NMA) synthesizes evidence from multiple treatments but relies on consistency.
• Existing statistical tests for NMA inconsistency have limitations in power and handling multi-arm studies.
• Inconsistency in NMA can stem from design-by-treatment interactions, necessitating methods to identify influential designs.

Purpose of the Study:

• To propose an alternative framework for evaluating inconsistency in NMA using influence diagnostics.
• To quantitatively assess the influence of individual study designs on overall NMA results.
• To provide methods for prioritizing investigation into sources of bias and heterogeneity in NMA.

Main Methods:

• Developed four novel influence diagnostic methods: averaged studentized residual, MDFFITS, Φd, and Ξd.
• Implemented a "leave-one-design-out" analysis framework to quantify individual design influence.
• Introduced a summary measure, the O-value, for straightforward interpretation and prioritization of influential designs.

Main Results:

• The proposed methods accurately identified potential sources of inconsistency in a network meta-analysis of antihypertensive drugs.
• Simulation studies confirmed the effectiveness of the new methods in locating inconsistency.
• The influence diagnostics provided quantitative insights into the impact of individual designs on NMA outcomes.

Conclusions:

• The novel influence diagnostics offer a valuable alternative to existing test-based methods for assessing NMA inconsistency.
• These methods enable quantitative evaluation of individual design influence, enhancing the interpretability of NMA.
• The proposed framework improves the identification and understanding of heterogeneity and potential bias in evidence synthesis.