Assessing the consistency assumptions underlying network meta-regression using aggregate data.

Area of Science:

• Biostatistics
• Health Economics
• Evidence Synthesis

Background:

• Network meta-regression (NMR) analyzes relative treatment effects across multiple therapies.
• NMR relies on two key consistency assumptions: effect at covariate zero and consistent regression coefficients.
• Violations of these assumptions can lead to unreliable NMR results and mask true treatment-covariate interactions.

Purpose of the Study:

• To introduce novel models for assessing the consistency assumptions of network meta-regression (NMR).
• To extend existing NMR models to incorporate treatment by covariate interactions and assess consistency.
• To provide methods for evaluating both consistency assumptions simultaneously or individually.

Main Methods:

• Outlined existing NMR models with treatment by covariate interactions.
• Introduced and extended node-splitting, unrelated mean effects, and design by treatment inconsistency models.
• Applied Bayesian framework to trial-level antimalarial data and fabricated datasets.

Main Results:

• Developed models to assess NMR consistency assumptions for aggregate data, including covariate interactions.
• Demonstrated application using antimalarial treatment data with average age as a covariate.
• Illustrated key scenarios with fabricated datasets to highlight model performance.

Conclusions:

• The proposed models enhance the reliability of network meta-regression (NMR) by assessing critical consistency assumptions.
• These methods improve the detection of treatment by covariate interactions, which may be masked by inconsistency.
• Provides a thorough understanding of consistency in NMR, crucial for evidence synthesis and clinical decision-making.