OBJECTIVES: Extrapolation of survival data in clinical trials is key to estimate lifetime benefits of novel oncology products. This review aims to understand the overall survival (OS) extrapolation methods used in National Institute for Health and Care Excellence (NICE) oncology technology appraisals.

METHODS: We reviewed 84 partition survival models (PSMs) informed by OS extrapolations in NICE oncology technology appraisals completed between December 2019 and December 2022. Information on the extrapolation methods used and corresponding suitability assessments as recommended by NICE in Technical Support Documents (TSD) 14 and 21 were extracted, along with ERGs’ critiques and Committees’ preferences.

RESULTS: Most cases considered all common standard parametric distributions (83/84), and 40% (33/84) further assessed flexible methods. The suitability of extrapolation methods was verified using relative statistical goodness-of-fit (83/84), visual inspection (81/84), tests of proportionality of the hazard (77/84), clinical expert validation (75/84), shape of the hazard (68/84), and comparison with external data (45/84). Weibull (18/67) and log-logistic (18/67) were the most frequently selected base-case standard parametric distributions (67/84) by companies, whereas flexible methods were preferred in 20% (17/84) of the cases, particularly piecewise models (7/17). ERGs preferred Weibull (15/59) and exponential (12/59) standard parametric distributions (59/84) for the base-case, and flexible methods in 23 out of 84 PSMs. In circumstances where ERGs and companies (45/84) differed on the choice of base-case extrapolation approach, Committees preferred ERGs’ base-case assumptions (25/45) more often than companies’ (12/45), or were indifferent between the two (8/45).

CONCLUSIONS: Most company submissions followed NICE TSDs 14 and 21 in their extrapolation approach. Standard parametric methods were widely used, and flexible methods were typically employed when standard approaches provided poor fitting to observed survival estimates. ERGs and companies differed on the base-case approach in over half of the cases. Committees' preferences for ERG’s base-case approach in such cases likely led to different cost-effectiveness estimates.