OBJECTIVES: Matching adjusted indirect comparison (MAIC) has become an increasingly popular method to conduct population adjustment for indirect comparisons between single-arm studies. The assumptions required for performing MAIC in this setting are strict, so there is need to empirically assess the method’s operating characteristics. Previous simulation studies evaluating MAIC for single-arm trials ignore the non-collapsibility of the treatment effect measure and therefore incorrectly specify the target of estimation. We conduct a comprehensive simulation study to assess the performance of MAIC for single-arm trials in a variety of scenarios.

METHODS: We evaluate the performance of MAIC when prognostic characteristics and effect measure modifiers are excluded from the matching scheme. We focus on the setting where the treatment effect measure is non-collapsible and consider both binary and time-to-event outcomes. We vary the trial sample size, level of imbalance in patient characteristics between trials, and prognostic strength of the excluded factors. Performance is summarized in terms of bias, precision, effective sample size, and the empirical coverage of confidence intervals.

RESULTS: Across settings, exclusion of any prognostic factor or effect measure modifier from the matching scheme significantly impacts MAIC's performance for single arm trials. The detrimental effect on performance is proportional to the amount of imbalance between trials and prognostic strength of the excluded factors.

CONCLUSIONS: Users of MAIC for single-arm trials should review its assumptions and plan sensitivity analysis to evaluate the strength of conclusions made from its application.