Assessing the performance of population adjustment methods for anchored indirect comparisons: A simulation study.

Area of Science:

• Health economics and outcomes research
• Biostatistics
• Epidemiology

Background:

• Standard meta-analysis assumes balanced populations, which is often unmet.
• Population adjustment methods (ML-NMR, MAIC, STC) use individual patient data to address this.
• These methods are increasingly used in health technology appraisals.

Purpose of the Study:

• To assess the performance of ML-NMR, MAIC, and STC under various assumption failures.
• To investigate the impact of sample size, missing effect modifiers, and covariate distributions.
• To compare these methods against standard indirect comparisons.

Main Methods:

• Extensive simulation study.
• Evaluated ML-NMR, MAIC, and STC.
• Assessed performance under scenarios with missing effect modifiers, extrapolation issues, and varying covariate distributions.

Main Results:

• ML-NMR and STC demonstrated robust performance, eliminating bias when assumptions were met.
• MAIC performed poorly across most scenarios, potentially increasing bias.
• All methods showed bias when key effect modifiers were missing.

Conclusions:

• ML-NMR and STC are reliable for population adjustment when all relevant effect modifiers are included.
• ML-NMR offers advantages in handling larger networks and estimating effects in any target population.
• Careful selection of effect modifiers is crucial for all population adjustment methods.