Parametric G-computation for compatible indirect treatment comparisons with limited individual patient data

Area of Science:

• Biostatistics
• Epidemiology
• Health Economics

Background:

• Matching-adjusted indirect comparison (MAIC) is common for cross-trial treatment effect comparisons.
• MAIC relies on propensity score weighting, sensitive to poor covariate overlap and limited extrapolation.
• Existing outcome regression methods extrapolate but yield incompatible conditional effects for indirect comparisons.

Purpose of the Study:

• To propose a novel marginalization method for population adjustment in indirect treatment comparisons.
• To develop a method that overcomes limitations of MAIC and conventional outcome regression.
• To enable accurate estimation of marginal treatment effects with limited patient-level data.

Main Methods:

• Developed a marginalization method using parametric G-computation.
• Applied to generalized linear models and Cox models for outcome regression.
• Separated covariate adjustment from marginal treatment effect estimation, allowing Bayesian integration.

Main Results:

• Parametric G-computation provided more precise and accurate estimates than MAIC, especially with poor covariate overlap.
• The method yielded unbiased marginal treatment effect estimates.
• Marginalized regression-adjusted estimates were more precise and accurate than biased conditional estimates from conventional outcome regression.

Conclusions:

• Parametric G-computation offers a robust alternative for population adjustment in indirect treatment comparisons.
• The method effectively recovers marginal treatment effects, improving precision and accuracy.
• This approach facilitates reliable comparisons of treatment effects in real-world evidence synthesis.