Hancock H, Brockbank J, Theodorou E, Brodtkorb TH. Survival analysis methods used in immuno-oncology NICE appraisals. Poster presented at the ISPOR Europe 2022; November 6, 2022. Vienna, Austria.

OBJECTIVES: Survival analysis is often required in economic evaluations to capture survival functions seen in clinical trial data and extrapolate outcomes into the future. It is important to consider the hazard function during and after the trial when selecting appropriate survival models. Standard parametric models may not adequately capture complex time-dependent hazard shapes, such as those seen with immuno-oncology treatments where patients can have a delayed response to treatment and long-term survival. NICE Technical Support Document (TSD) 21 describes a range of more sophisticated survival modelling approaches that can be used for complex hazard functions. The objective of the study was to review the use of different survival models in immuno-oncology NICE technology appraisals (TAs) completed after the publication of TSD 21 in January 2020.

METHODS: The NICE website was searched to identify all completed TAs for immuno-oncology published from January 2020 to the end of May 2022. Information about the use of different survival models was extracted from TA documents available on the NICE website. 

RESULTS: 26 TAs were identified for 6 immunotherapies (atezolizumab, avelumab, durvalumab, ipilimumab, nivolumab, and pembrolizumab) in different oncology indications (7 lung, 3 bladder, 3 renal, 3 skin, 3 esophageal, 2 head and neck, 2 colorectal, 1 breast, 1 lymphatic, and 1 liver). Of the 26 TAs, 6 (23%) considered standard parametric survival models only and 20 (77%) explored more sophisticated models. Flexible spline-based models and piecewise models were most used and accepted by NICE when standard parametric models could not capture complex hazard functions.

CONCLUSIONS: Most immuno-oncology NICE TAs completed after the publication of TSD 21 explored more sophisticated survival analysis methods than standard parametric models to both fit the trial data and produce plausible extrapolations. The results of this study may be used to inform the survival modelling approach for future TAs.

Share on: