PNS308 METHODOLOGICAL APPROACHES FOR INCORPORATING INVESTIGATORS' CHOICE COMPARATOR ARMS INTO NETWORK META-ANALYSES

Abstract

In investigators’ choice (IC) arms of randomized controlled trials (RCT), patients receive one of several different treatments based on the treating physician’s choice. This can pose a challenge in network meta-analyses (NMA), as these treatments are effectively ‘lumped’ into a single trial arm. Our objective was to evaluate a range of approaches for incorporating IC RCTs into NMAs, and to demonstrate their use with simulated and real-world examples. The three main approaches involved: 1) using a ‘lumped’ NMA node consistent with the IC arm; 2) splitting the NMA node into individual treatments, and 3) applying exchangeable treatment effects on a split NMA node. For the approaches involving split nodes, data inputs from the IC RCT involved: a) treatment effect estimates for the lumped IC comparison, applied to each split IC node; b) treatment-specific subgroup estimates; and c) a novel approach involving a weighted average of the lumped IC estimates. Using a simulated data set, the outputs from each approach were compared against the ‘true’ treatment effects underlying the simulated data set. Additionally, the approaches were applied to a real-world NMA in first-line advanced lung cancer. Using the simulated data set, the weighted approach yielded outputs closest to the ‘true’ values. The illustrative analysis in first-line lung cancer showed differences across the implementations in progression-free survival, whereas overall survival results were similar. The major limitations included incorporating non-randomized evidence (split nodes approach using subgroup data), and the potential for violating the transitivity assumption of the NMA (lumping and splitting nodes approach using lumped IC estimates); the weighted approach avoided such drawbacks. Our findings demonstrate that when including IC arms into NMAs, design choices can artificially impact treatment effect estimates. Researchers should consider the range of available methods and evaluate their underlying assumptions prior to conducting an NMA involving IC arms.