Presentation (CTI)

OBJECTIVES: Population-adjusted methods such as matching-adjusted indirect comparison (MAIC) and simulated treatment comparison (STC) have sought to address the susceptibility of network meta-regression (NMR) to aggregation (ecological) bias, via analyses that incorporate both aggregate-level data (AgD) and individual patient data (IPD). However, MAICs and STCs are only able to perform pairwise comparisons between two trial populations. Furthermore, inferences made by such analyses are only valid within the AgD population, which is seldom the desired population of interest for the decision problem at hand. Recent methodological advances have yielded multi-level NMR (ML-NMR), which can mitigate aggregation bias, include any number of interventions from both AgD and IPD studies, and yield estimates that are valid in any given study population. However, the ability of ML-NMR to address aggregation bias has yet to be scrutinised. Here, we seek to assess the ability of ML meta-regression (ML-MR) to disentangle effect modification at the individual level from that at the study level in a simple pairwise analysis scenario via simulation.
METHODS: Four pools of 10 trials were simulated to model the association between an effect modifier (x) and two interventions (A & B) on a continuous outcome (y) in 200 participants reflecting four aggregation bias scenarios: 1) participant-level and study-level effects of equal magnitude and direction; 2) positive effect at the participant-level, no effect at the study-level; 3) positive effect at the study-level, no effect at the participant-level; and 4) positive effect at the participant-level, negative effect at the study level. Within each pool, the number of aggregated trials was incrementally increased, starting with no trials (i.e. IPD-MR), 1-9 trials (i.e. ML-MR), and ending with the aggregation of all trials (i.e. AgD-MR).
RESULTS: Simulation inputs will be presented, along with estimates of relative treatment effects and 95% credible intervals from each of the input scenarios.
CONCLUSIONS: To be presented