A leave-one-out algorithm for contribution analysis in component network meta-analysis.

Area of Science:

• Biostatistics
• Health Research Methodology
• Evidence Synthesis

Background:

• Component network meta-analysis (CNMA) allows for the separation of individual component effects within multicomponent treatments.
• Current CNMA methods lack established approaches to quantify the contribution of evidence from specific comparisons to component effect estimates.
• This deficiency hinders the interpretability and transparency of results from CNMA.

Purpose of the Study:

• To develop and validate a novel method for quantifying the contribution of constituent comparisons to component effect estimates in CNMA.
• To enhance the interpretability and precision of evidence synthesis for complex interventions.

Main Methods:

• A leave-one-out algorithm was proposed, iteratively excluding each comparison (edge) in the network.
• The algorithm recomputes component effect variances and quantifies precision leverage based on induced variance inflation.
• Special rules were developed for unidentifiable components, and the method decomposes estimates into direct and additive evidence sources.

Main Results:

• The leave-one-out algorithm successfully identified pivotal evidence sources by detecting significant variance changes upon exclusion.
• Precision leverage effectively quantified the importance of comparisons isolating specific components.
• Application to real-world data demonstrated the method's precision in complex networks and alignment with parameter decomposition.

Conclusions:

• The leave-one-out algorithm provides a robust, variance-based framework to quantify evidence contributions in CNMA, addressing a critical methodological gap.
• This method reliably identifies crucial evidence sources for component identifiability and precision.
• It significantly enhances the transparency and interpretability of evidence synthesis for complex interventions.