Assessing the Impact of a Matching-Adjusted Indirect Comparison in a Bayesian Network Meta-Analysis

If IPD is available for some or all trials in a network meta-analysis (NMA), then incorporating this IPD into an NMA is routinely considered to be preferable. However, the situation often arises where a researcher has IPD for trials concerning a particular treatment (eg, from a sponsor) but none for other trials. Therefore, one can reweight the IPD so that the covariate characteristics in the IPD trials match that of the aggregate data (AgD) trials, using a matching-adjusted indirect comparison (MAIC). We assess the impact of using the reweighted aggregated data, obtained by the MAIC, in a Bayesian NMA for a connected treatment network. We apply this method to a network of multiple myeloma treatments in newly diagnosed patients (ndMM), where the outcome is progression free survival. We investigate the reliability of the methods and results through a simulation study. The ndMM network consists of three IPD studies comparing lenalidomide to placebo (Len-Placebo), one AgD study comparing Len-Placebo, and one AgD study comparing thalidomide to placebo (Thal-Placebo). We therefore investigate two options of weighting the covariates: (a) All three studies are weighted separately to match the AgD Thal-Placebo trial. (b) Patients are weighted across all three IPD studies to match the AgD Thal-Placebo trial, but the NMA considers each trial separately. We observe limited benefit to MAIC in the full network population. While MAIC can be beneficial as a sensitivity analysis to confirm results across patient populations, we advise that MAIC is used and interpreted with caution.