Multi-Level Network Meta-Regression (ML-NMR) vs Matching-Adjusted Indirect Comparison (MAIC): A Comparative Analysis via Simulations to Illustrate the Importance of Population Adjustment

OBJECTIVES: Matching-Adjusted Indirect Comparison (MAIC) is a commonly used population adjustment method to assess the relative treatment effect of two treatments in heterogeneous trials. However, MAIC has known limitations, including the limited validity of estimates to the comparator’s population, due to the reweighting process based on treatment effect modifiers (TEMs) distribution from the aggregated data (AgD). The Multi-level Network Meta-Regression (ML-NMR) approach, introduced by Phillippo in 2019, is the most methodologically robust population-adjusted indirect comparison as endorsed in the recent HTA Coordination Group Guidelines for Quantitative Evidence Synthesis. This work aims to illustrate through simulations the importance of the population used for the adjustment and how ML-NMR overcomes this known MAIC limitation.

METHODS: A time-to-event outcome following a Weibull model and three covariates from two randomized clinical trials (RCTs), comparing drugs A vs B and A vs C, were simulated. Both MAIC and ML-NMR approaches were performed using individual patient data (IPD) for the AB trial and AgD for AC.

RESULTS: Estimates obtained through the ML-NMR were closer to the true values compared to the MAIC. MAIC was found to be limited in adjusting for heterogeneity across different populations, providing one estimate per comparison, only valid for the AgD population. In contrast, the ML-NMR demonstrated greater ability to estimate true relative treatment effects for various populations. The ML-NMR also allowed to get an overview of the impact of the population characteristics on the relative estimates.

CONCLUSIONS: The relative treatment effect estimates obtained through the MAIC or ML-NMR are specific to the population characteristics in terms of TEMs. While the MAIC can compute estimates for the comparator’s population only, the ML-NMR can compute estimates applicable to different populations, making it a more flexible and potentially less biased method for indirect comparisons.