Regression augmented weighting adjustment for indirect comparisons in health decision modelling

Area of Science:

• Health economics
• Biostatistics
• Pharmaceutical research

Background:

• Accurate health resource allocation requires comparing all interventions, but novel drugs are often only tested against placebos.
• Indirect comparisons are vital for evaluating relative efficacy against alternative treatments.
• Population adjustments are necessary when treatment effect modifiers differ across studies, with Matching-Adjusted Indirect Comparisons (MAIC) being a common but potentially unstable method.

Purpose of the Study:

• To introduce G-MAIC, a novel method combining outcome regression and weighting-adjustment.
• To address the limitations of existing methods, particularly MAIC's instability under poor population overlap.
• To evaluate G-MAIC's performance against standard methods in diverse simulation scenarios.

Main Methods:

• Developed G-MAIC, integrating Bayesian survey inference and Bayesian bootstrap for uncertainty propagation.
• Compared G-MAIC against non-adjusted methods, MAIC, and Parametric G-computation.
• Conducted a simulation study with 18 scenarios varying sample sizes, population overlap, and covariate structures.

Main Results:

• MAIC showed instability, increased bias, or non-sensible variance under poor overlap and small sample sizes.
• G-MAIC demonstrated comparable performance to parametric G-computation.
• G-MAIC achieved this with a reduced reliance on parametric assumptions, offering improved robustness.

Conclusions:

• G-MAIC is a robust alternative to MAIC for population-adjusted indirect comparisons.
• The G-MAIC framework is flexible, accommodating advanced nonparametric models and weighting schemes.
• This method enhances the reliability of indirect treatment comparisons in pharmaceutical research.