Background: For HTA purposes, immature survival from clinical trials needs to be extrapolated. Some trials might include two groups of patients with different survival expectations (short- versus long-term survivors). Standard parametric models (SPM), generally used for extrapolation purposes, cannot capture these two survival trends. This study aims to assess the performance of parametric mixture models (PMM) that use two distributions to capture these conflicting survival expectations.

Methods: SPMs and PMMs were fitted on immature data of three trials in melanoma, breast cancer (BC) and multiple myeloma (MM). The best fitting models were selected based on LOOIC. The extrapolation on immature data was compared to mature trial data in terms of ΔMean-Absolute-Deviation (ΔMAD) (lower ΔMAD implies a better prediction) in months for all arms in the trial. Restricted mean survival time (RMST) over all arms of the mature data was determined to assess the relative size of ΔMAD. Mixtures of exponential, Gompertz, loglogistic, lognormal and Weibull were compared.

Results: For the melanoma dataset, where the RMST of the mature dataset was 67.88 months, ΔMAD was 19.59 and 23.73 months for the best fitting PMMs and SPMs, respectively. For the BC dataset (RMST: 123.42 months) ΔMAD was 9.87 and 44.07 months, and for the MM dataset (RMST: 92.55 months) ΔMAD was 11.24 and 7.21 months, respectively. The best fitting distribution on immature data did not always correspond to the lowest ΔMAD.

Conclusions: For two out of three trials, the PMM extrapolations over the immature data were more aligned with the mature data than the SPM. In case it is plausible that a trial population consists of a mixed patient population in terms of survival expectations, PMMs should be considered in future HTAs. Final selection of an extrapolation model should also consider external data and expectations on long-term comparative effectiveness.