Survival Extrapolation for Immuno-Oncology Drugs: Pessimism or Optimism? a Targeted Literature Review and a Validation-Based Case Study
Author(s)
Phung TH
University of York, York, UK
Presentation Documents
Background: The objectives of this study are (1) examine the justification of selecting complex survival extrapolation models among ones suggested in the DSU Technical Support Document 21 and how Evidence Review Groups and appraisal committees responded to the use of complex models, (2) assess the accuracy of the used and unused methods in the NICE technical appraisal 581 to establish whether alternative methods could have provided better prediction. Methods: All NICE TAs for immune-oncology drugs (IOs) conducted from January 2018 to July 2021 were included. Information of justifications for extrapolation methods, critiques of ERGs, and final appraisal determination of committees over these methods were extracted. A range of extrapolation methods (fully-fitted parametric, semi-parametric, cubic spline-based, and cure fraction) were performed using the 2-year dataset of PFS of the CHECKMATE 214 trial. The 4-year dataset was used to validate the projections. Results: A total of 23 TAs were included. Once standard-parametric models were inappropriate (n = 16), the majority of TAs (n = 14/16) failed to justify selecting complex models among alternatives. ERGs and committees generally agreed on the methods modelling observed hazards (semiparametric, cubic spline-based models) whilst against modelling the unobserved hazards (response-based and cure fraction models). All TAs incorporating a “cured” proportion in their models (n = 3) was rejected. In the validation-based case study, only cure fraction models provided accurate 4-year survival predictions compared to the updated dataset. All other alternatives including models used in TA581 by the company and ERG substantially underestimated 4-year survival. Conclusion: Though the targeted literature review showed conservative standpoints from ERGs and committees over response-based and cure fraction models, the validation-based case showed that cure fraction could provide more accurate prediction for long-term PFS of dual IOs. This encourages the future NICE TAs to collect further long-term information to reassess the previous models.
Conference/Value in Health Info
2022-05, ISPOR 2022, Washington, DC, USA
Value in Health, Volume 25, Issue 6, S1 (June 2022)
Code
HTA72
Topic
Economic Evaluation, Health Technology Assessment
Topic Subcategory
Cost-comparison, Effectiveness, Utility, Benefit Analysis, Decision & Deliberative Processes
Disease
Oncology