OBJECTIVES:

Network meta-analyses (NMAs) are often used to derive hazard ratios (HRs) of relative treatment effect when modelling survival estimates in cost-effectiveness analyses (CEAs). A ‘reference’ treatment, for which Kaplan-Meier (KM) data are available, is selected and survival estimates for remaining treatments are generated by applying NMA-derived HRs. However, limited guidance exists for choosing this ‘reference’ when using NMA-derived HRs in CEAs. This research investigated the impact of the choice of reference treatment on survival estimates.

METHODS:

A published NMA for non-small cell lung cancer was used as a case study. To generate ‘reference’ curves, three survival functions for which the proportional hazards assumption holds (exponential, Weibull, Gompertz) were fitted to published progression-free survival (PFS) KM data from afatinib, dacomitinib, gefitinib and osimertinib trials used in the NMA. The fitted extrapolations for each treatment were used as reference curves in turn, and HRs from the NMA were applied to each reference treatment to generate survival curves for the remaining treatments. Mean PFS was calculated for the extrapolated and HR-derived survival curves of each treatment.

RESULTS:

Using the best statistically fitting extrapolation for all treatments (Weibull), the mean PFS estimates derived by applying the NMA-derived HRs to different reference curves varied by 1.7 months for gefitinib (14% of the 12.4-month Weibull-extrapolated gefitinib PFS), to 5.3 months for osimertinib (26% of the 20.1-month Weibull-extrapolated osimertinib PFS). Despite this, the Weibull-extrapolated PFS estimates fell within the ranges of the HR-derived mean PFS estimates in all cases, except osimertinib (Weibull-extrapolated mean PFS: 20.1 months; HR-derived mean PFS: 21.0–26.3 months).

CONCLUSIONS:

This research demonstrates that survival estimates generated using HRs are sensitive to the choice of reference treatment, which could have unforeseen consequences in CEAs. The findings support a need for clearer guidance on selecting the reference treatment when using NMA-derived HRs to derive survival estimates in CEAs.