Identifying and Extrapolating Immature OS Data in Survival Analysis: A Review of the Variation in Methods Used in Oncology STAs for NICE

OBJECTIVES: Technology appraisals (TAs) submitted to the National Institute for Health and Care Excellence (NICE) describe cost-effectiveness analyses conducted on medical therapies, including survival analysis carried out on time-to-event (TTE) data. This research explores the variety in statistical methods used to model immature overall survival (OS) data and how these were viewed by the NICE external assessment group (EAG) or evidence research group (ERG) committee, with the aim of determining whether guidelines on extrapolation methods for immature OS data should be more firmly outlined by NICE documentation.

METHODS: All publicly available NICE single technology assessments (STAs) in oncology up to April 13, 2023 were identified. From these, information on the maturity of OS data, approach used to fit OS data, and EAG/ERG opinion on the related methods were extracted and analysed in a descriptive manner. Where OS data was mature, information on the approach used to fit OS data was not extracted.

RESULTS: The 92 most recent NICE STAs were extracted, reflecting 97 distinct treatment comparisons, of which 90 included survival analysis. 60% of the appraisals stated that they had immature data with a median follow-up duration of 20.6 months (IQR: 16.4 – 28.1 months). The most common method(s) used to fit immature OS data were standard parametric models alone (53.7%), standard parametric models with spline models (14.8%), and standard and semi-parametric models (9.3%). In 59.5% of the studies the EAG/ERG did not agree with the survival methods used and recommended using alternative distributions or modelling methods.

CONCLUSIONS: Standard parametric methods are most frequently used when dealing with immature OS data in oncology NICE TA submissions, however, the EAG/ERG did not agree with the methods used in over half of the STAs. This suggests that guidelines on extrapolation methods for immature OS data should be more firmly outlined by NICE documentation.