Modeling missing binary outcome data while preserving transitivity assumption yielded more credible network...

Area of Science:

• Biostatistics
• Evidence Synthesis
• Health Research Methodology

Background:

• Missing participant outcome data (MOD) poses challenges in network meta-analysis (NMA).
• The transitivity assumption is fundamental for valid NMA but can be compromised by MOD.
• Conventional approaches to MOD may lead to biased or unreliable NMA findings.

Purpose of the Study:

• To conceptually evaluate the transitivity assumption in binary MOD within NMA.
• To highlight the importance of statistical modeling for addressing MOD.
• To propose modifications to scenarios that compromise transitivity in complex networks.

Main Methods:

• Conceptual evaluation of the transitivity assumption in the context of binary MOD.
• Identification and modification of scenarios that compromise transitivity in complex networks.
• Application of a published NMA to demonstrate the implications of different MOD handling strategies.

Main Results:

• Arm-specific scenarios for MOD, common in conventional meta-analysis, violate transitivity in complex networks.
• Imputing MOD in these scenarios yields estimates in the opposite direction with inflated between-trial variance.
• Modeling MOD after scenario modification provides robust estimates but wider credible intervals, reducing between-trial variance.

Conclusions:

• Modifying arm-specific scenarios for binary MOD is essential in complex networks to ensure transitivity.
• Statistical modeling of MOD, rather than exclusion or imputation, is recommended for bias-adjusted NMA results.