OBJECTIVES: Both comparative effectiveness and survival extrapolations are needed to assess survival benefits with novel therapies compared to standard-of-care. In HRNMA, PHA must hold and, for extrapolation purposes, should not be combined with non-PH models such as log-logistic and log-normal. If the PHA does not hold, then the use of PNMA is justified. Hence, in this application, we aimed to assess the impact of applying PNMA versus HRNMA on survival extrapolations in aNSCLC. Additional alternative methodologies such as mixture, mix-cure model (MCM), non-MCM, piecewise, spline, and fraction polynomial NMAs were also assessed.

METHODS: First, as an example, a published HRNMA in aNSCLC was retrieved. The HRNMA included four different studies, and three different comparators: pembrolizumab (PMB), chemotherapy (CH), and PMB+CH. Individual patient data were constructed and validated from Kaplan-Meier curves for overall survival. Reported HR for CH versus PMB was 1.58 (95%CI:1.11-2.27), and for PMB+CH versus PMB was 0.80 (95%CI:0.36-1.79). Second, the PHA was evaluated using the Schoenfeld test. If PHA was violated in at least one study, third, PNMA was applied. Last, extrapolations from both the PNMA and the HRNMA based on Weibull distribution were compared.

RESULTS: PHA was violated in one study comparing PMB versus CH; therefore, it invalidated the HRNMA and justified the use of PNMA. In PNMA, the estimated overall mean survival was 3.22, 3.62, 1.68 years for PMB, PMB+CH, and CH, respectively. On the other hand, based on the published HRNMA, the estimated overall mean survival was 2.77, 3.54, 1.64 years for PMB, PMB+CH, and CH, respectively.

CONCLUSIONS: In this example, there were differences in the overall mean survival, which may have an impact on subsequent reimbursement decisions. Furthermore, PNMA offered an additional advantage in addition to no need to follow proportionality assumptions: NMA results can directly be implemented into a health economic model.