An Evaluation of Survival Curve Extrapolation Techniques UsingLong-Term Observational Cancer Data

摘要

Uncertainty in survival prediction beyond trialfollow-up is highly influential in cost-effectiveness analyses of oncologyproducts. This research provides an empirical evaluation of the accuracy ofalternative methods and recommendations for their implementation. Mature (15-year) survival data were reconstructed froma published database study for “no treatment,” radiotherapy, surgery plusradiotherapy, and surgery in early stage non–small cell lung cancer in anelderly patient population. Censored data sets were created from these data tosimulate immature trial data (for 1- to 10-year follow-up). A second data setwith mature (9-year) survival data for no treatment was used to extrapolate thepredictions from models fitted to the first data set. Six methodologicalapproaches were used to fit models to the simulated data and extrapolate beyondtrial follow-up. Model performance was evaluated by comparing the relativedifference in mean survival estimates and the absolute error in the differencein mean survival v. the control with those from the original mature survivaldata set. Model performance depended on the treatmentcomparison scenario. All models performed reasonably well when there was a smallshort-term treatment effect, with the Bayesian model coping better with shorterfollow-up times. However, in other scenarios, the most flexible Bayesian modelthat could be estimated in practice appeared to fit the data less well than themodels that used the external data separately. Where there was a large treatmenteffect (hazard ratio = 0.4), models that used external data separately performedbest. Models that directly use mature external datacan improve the accuracy of survival predictions. Recommendations on modelingstrategies are made for different treatment benefit scenarios.