Two new approaches for the visualisation of models for network meta-analysis

Area of Science:

• Biostatistics
• Medical Informatics
• Evidence-Based Medicine

Background:

• Meta-analysis combines evidence from multiple studies to estimate pooled treatment effects.
• Network meta-analysis extends this by simultaneously comparing multiple treatments.
• Interpreting complex network meta-analysis models, especially with many treatments, is challenging.

Purpose of the Study:

• To propose two novel visualization methods for easier interpretation of network meta-analysis models.
• To enhance the intuitive understanding of fitted models in network meta-analysis.

Main Methods:

• Developed distance measures based on treatment effects, standard errors, and p-values.
• Categorized treatments as "close" or "connected" based on distance thresholds.
• Utilized community detection algorithms from network analysis to group treatments.
• Introduced a second method using parametric bootstrapping and heat maps for visualization.

Main Results:

• Illustrated methods on a dataset with 22 treatments.
• Identified two distinct communities of treatments with similar efficacy.
• Successfully pinpointed well-identified and less-identified parts of the network.

Conclusions:

• The proposed visualization methods offer an intuitive way for network meta-analysts to understand model implications.
• These methods can be used informally to identify key findings or formally in published reports.
• Enhanced interpretability aids in synthesizing evidence from complex treatment comparisons.