POSA308 A Comparison of STC and Maic Under Misspecification of the Treatment Effect and Effect-Modifier Relationship

Abstract

Recent simulations comparing the accuracy of simulated treatment comparisons (STCs) and matching adjusted indirect comparisons (MAICs) to adjust for imbalances in effect modifiers (EMs) have revealed deficiencies in MAICs when matching on continuous covariates, raising concerns about use of this approach. MAIC may offer an advantage, however, when the form of effect modification is misspecified (e.g., adjusting for age groups when EM is a linear function of age) since the EM-outcome relationship is not modeled explicitly. We examined this in a simulation study. Simulations considered two continuous EMs and a dichotomous outcome to compare treatments B and C, anchored to A. Data are generated assuming the relationship between EMs and the log-odds ratios (LORs) for B vs. A and C vs. A is either linear or a step-function. STC and MAIC are performed adjusting for EMs as continuous when the LOR changes as a step-function, and as dichotomous when LOR changes linearly. Performance under correct specification was also assessed, assuming B and C have equal efficacy, but varied distribution of EMs in the trials. Average LORs (aLORs) for C vs. B across replications are used to assess accuracy. When correctly specified, the absolute bias was small but relatively larger with MAIC when EM was continuous (aLOR: 0.06 vs. 0.02). MAIC was more accurate with categorical EMs (0.003 vs. 0.02). Under misspecification, STC and MAIC were similarly biased (0.09 vs. 0.08) when the EMs are incorrectly matched as dichotomous, while larger bias was observed with STC when EMs were incorrectly adjusted as continuous (0.22 vs. 0.03). As shown previously, MAICs appear deficient with continuous EMs even under correct specification but are relatively more robust to misspecification. Findings highlight the importance of understanding both the presence and shape of effect modification to avoid bias with STC from misspecification.