OBJECTIVES: In an MAIC, the relative efficacy is predicted in the aggregated data population (AgD) instead of the intervention population. Ideally, in HE models, the prediction should occur in the intervention population. Therefore, in HE models, the inverse hazard ratio (HR) from an MAIC is often applied to the intervention curves to obtain extrapolated comparator curves, hereby assuming shared-effect modification (SEM). We explored a novel method for unanchored MAICs, using similar principles as used in parametric network meta-analyses, for constructing comparator curves for HE models when the PH-assumption is violated.

METHODS: MAIC was run based on the IPD from MagnetisMM-3 (NCT04649359) and published data from a real-world evidence (RWE) study in relapsed-refractory multiple myeloma. Six standard parametric models were fitted based on the MAIC-weighted data and RWE. To allow for PH violations, we applied treatment effects on all coefficients, e.g., shape and scale, of the parametric models to the MagnetisMM-3 curve to derive the MAIC-adjusted comparator curves in the appropriate population.

RESULTS: After weighting, the PH assumption was violated between MagnetisMM-3 and the RWE study. The constructed comparator curves shifted as expected with all tested parametric models, based on the observed relative efficacy shown in the weighted intervention curves and the original comparator curves in the MAICs. Mean survival for the RWE arm was smaller with the novel method and therefore incremental survival was also increased compared to the standard method.

CONCLUSIONS: The proposed approach is a promising method for extrapolating comparator curves using unanchored MAIC outputs when non-PH presents. Like the HR approach, by assuming SEM, it allows for the comparison of intervention and comparator consistently on the intervention population in HE models.