Vickers AD, Hawe E. An evaluation of survival curve extrapolation techniques using long-term observational cancer data. Poster presented at the ISPOR 21st Annual European Congress; November 13, 2018. Barcelona, Spain.


OBJECTIVES: Uncertainty in survival prediction beyond trial follow-up is highly influential in cost-effectiveness analyses of oncology products. Several extrapolation techniques have been proposed in the literature. This research provides an empirical evaluation of the accuracy of alternative methods and recommendations for their implementation.

METHODS: Mature (15-year) survival data were reconstructed from a published database study for four treatments in early-stage non-small cell lung cancer. Censored data sets were created from these data to simulate immature trial data (for 1 to 10 years follow-up). Four methodological approaches were used to fit models to the simulated data and extrapolate beyond trial follow-up, which included conventional methods and refinements to recently published methods, which directly use mature external data. Model performance was evaluated by comparing mean survival estimates with those from the original mature survival dataset, using probability of equivalence and magnitude of error.

RESULTS: An ensemble of flexible spline (Weibull) models and parametric models fitted to the simulated trial datasets, which directly used mature external data, and general population data (to incorporate the increasing risk of death from other causes) performed best and produced survival estimates that were within 4% of the observed value when simulated trial follow-up was 3 years or more. Hybrid methods using Kaplan-Meier data prior to a cut-point (selected via Chow break tests) and parametric functions fitted to the remaining data thereafter gave the least accurate predictions. Single survival distributions and a Bayesian method to simultaneously model the data sets provided intermediate accuracy and underestimated uncertainty.

CONCLUSIONS: Models that directly used mature external data produced the most accurate survival predictions. Care is needed to capture the uncertainty. Reliance on single distributions resulted in underestimation of error and possible bias. Recommendations on the assumptions to use for the treatment effect after trial follow-up are made.

Share on: