A review of methods for comparing treatments evaluated in studies that form disconnected networks of evidence

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Health Technology Assessment

Background:

• Network meta-analysis (NMA) enables simultaneous comparison of multiple treatments in randomized controlled trials.
• Estimating treatment effects across disconnected networks of evidence presents unique statistical challenges.
• Imbalances in prognostic variables and treatment effect modifiers require specialized modeling assumptions.

Purpose of the Study:

• To review and discuss statistical methods for comparing treatments within disconnected evidence networks.
• To highlight the importance of clinical context and data availability in selecting appropriate methods.
• To identify gaps in current methodologies regarding uncertainty quantification.

Main Methods:

• Review of existing statistical approaches for disconnected network meta-analysis.
• Discussion of methods addressing prognostic variable and treatment effect modifier imbalances.
• Analysis of uncertainty sources, including sampling variation and external information incorporation.

Main Results:

• Several methods exist for disconnected network meta-analysis, with varying applicability based on clinical context and data.
• Most methods address sampling variation but may not fully account for other uncertainty sources.
• The need for further research into method properties and incorporation of external information is highlighted.

Conclusions:

• Comparing treatments in disconnected networks requires careful consideration of statistical methods and data.
• Current methods often require further development to comprehensively address all sources of uncertainty.
• Future research should focus on robust methods that incorporate external data to improve parameter and structural uncertainty estimation.