OBJECTIVES

To compare the long-term survival benefit of treatments that have not been investigated head-to-head, a parametric network meta-analysis (PNMA) can be conducted. PNMAs may predict clinically implausible hazards over time that are lower than general population mortality (GPM) hazards. This can be corrected retrospectively. Alternatively, one can consider GPM in the likelihood function of the PNMA to warrant more clinically plausible survival predictions. This study investigated the impact of accounting for trial-specific age-adjusted GPM in a PNMA and the impact of varying ages across trials in the network.

METHODS

Five treatment regimens (VMP/MP/MPT/Rdcont/Rd18) investigated in four different multiple myeloma trials (median age varying from 71-79 years) were included in the analysis. Four Weibull PNMAs methods were applied to investigate the impact on predicted survival: 1) standard PNMA; 2) standard PNMA using GPM hazards when PNMA predicted hazards are below GPM hazards; 3) PNMA which considers a single GPM for all trials in the likelihood function (age 71 based on reference trial comparing VMP and MP 1); and 4) PNMA which considers a trial-specific age-adjusted GPM in the likelihood function.

RESULTS

All PNMAs fitted the data well. Predicted mean survival (years) by PNMA method were for VMP) 13.9/10.1/8.7/8.8; MP) 4.7/4.6/4.5/4.5; MPT) 8.4/7.7/6.9/7.5; Rdcont) 10.2/8.9/7.8/8.5; and Rd18) 10.4/9.0/7.9/8.5. The impact of considering trial-based ages (method 3 vs 4) resulted in longer survival predictions for MPT (6.9 vs 7.5), Rdcont (7.8 vs 8.5) and Rd18 (7.9 vs 8.5). The impact of method 3 vs 4 is more pronounced for these three treatments as these were investigated in other trials with older patients than the reference trial.

CONCLUSIONS

Choice of PNMA method considering GPM has an impact on the predicted survival. Age differences between trials could be an important consideration when conducting a PNMA, especially when substantial heterogeneity in age is observed over trials in the network.