Tipping point analysis in network meta-analysis

Area of Science:

• Biostatistics
• Health Economics
• Clinical Epidemiology

Background:

• Network meta-analysis (NMA) synthesizes evidence from multiple treatments, but data sparsity, especially in arm-based NMA (AB-NMA), complicates correlation estimation.
• Accurate correlation estimates are crucial for robust conclusions on relative treatment effects in healthcare decision-making.

Purpose of the Study:

• To introduce a novel tipping point sensitivity analysis for arm-based network meta-analysis (AB-NMA).
• To assess the influence of correlation parameters on the robustness of conclusions regarding relative treatment effects.

Main Methods:

• Developed a Bayesian tipping point analysis specifically for correlation parameters in AB-NMA.
• Evaluated changes in conclusions based on the 95% credible interval including the null value (interval conclusion) and point estimate magnitude.
• Applied the method to 112 treatment pairs across multiple NMA datasets.

Main Results:

• Identified tipping points for interval conclusion change in 13 pairs (11.6%) and for magnitude change (≥15% threshold) in 29 pairs (25.9%).
• Findings suggest potential commonality in tipping points across different NMA datasets.
• Demonstrated the utility of the analysis through a case study with visual interpretation.

Conclusions:

• The proposed tipping point analysis is crucial for AB-NMA, particularly in networks with sparse direct comparisons or wide correlation credible intervals.
• Integrating this sensitivity analysis into standard practice can enhance the reliability of NMA findings.
• The approach provides valuable insights into the stability of relative treatment effect conclusions.