OBJECTIVES: The objective of this research is to understand the influence of different effect modifier (EM) scenarios on indirect treatment comparison within the setting of anchored matching adjusted indirect comparison (MAIC).
METHODS: Patient-level example datasets for two oncology trials were simulated. Trial 1 (individual patient data-accessible trial) comprised Treatment A and Placebo patients; Trial 2 (aggregate level data-accessible trial) comprised Treatment B and Placebo patients. The primary endpoint was progression-free survival. It was assumed that Trial 2 reported subgroup analyses from which EMs were identifiable. The following dichotomous variables were identified as EMs: genetic marker (EM in Trial 1 but not Trial 2), brain metastases (EM in Trial 2 but not Trial 1), late stage (EM in both trials; same direction), and laboratory test positivity (EM in both trials; opposite directions). Per Decision Support Unit recommendations, MAIC was implemented on overall trial populations adjusting for all EMs simultaneously to obtain an adjusted Treatment A versus Treatment B hazard ratio (HR). Stratified analyses using overall MAIC weights were performed for each category of EM to obtain eight distinct EM-stratified adjusted HR estimates. The naive HR was computed as an unadjusted reference point.
RESULTS: The HRs for Trials 1 and 2 were 0.737 and 0.604, respectively, yielding a naive Treatment A versus Treatment B HR of 1.220 (95% CI: 0.911, 1.634). Overall MAIC yielded an adjusted HR of 0.991 (95% CI: 0.736, 1.333). EM-stratified MAICs yielded the following adjusted HRs (yes category; no category): genetic marker (0.652; 1.307), brain metastases (2.292; 0.881), late stage (0.802; 1.107), laboratory positivity (0.586, 1.893).
CONCLUSIONS: In this simulation, overall and EM-stratified MAIC analyses yielded notably different HR estimates in some scenarios. Consideration of reference population characteristics is imperative in the context of treatment effect.
