OBJECTIVES: Assessing the influence of different survival extrapolation techniques on cost-effectiveness analyses is crucial. This is particularly significant in situations where the need for projections extends beyond clinical trial durations, given the constraints of limited follow-up. This research aims to recognize and overcome obstacles related to parametric survival models by incorporating general population mortality rates directly into the modelling process through an excess hazard (EH) model.

METHODS: To incorporate the excess hazard (EH) directly into the modeling process, a clinical trial assessing a treatment with potential curative properties in non-small cell lung cancer (NSCLC) was selected. Seven standard parametric models were fitted to overall survival and progression-free survival data. The same seven models were subsequently applied within the EH framework, both with and without a cure parameter, integrating general population mortality rates to prevent unrealistic projections. The EH model separates mortality rates into background (expected) and excess rates, enabling more realistic long-term projections. Model fit evaluation was assessed by the Akaike information criterion (AIC) and visual inspection. The fitting of models and extrapolations was conducted using the "flexsurv" and "flexsurvcure" packages in R.

RESULTS: Based on the AIC and visual inspection, the Log-normal model exhibited the best fit in both frameworks-standard parametric model (AIC = 799.8) and EH framework (AIC = 765.4). The introduction of a cure assumption in EH framework led to an overall improvement in the goodness of fit results for most parametric models, except for the Exponential and Gompertz. Additionally, comparisons of root mean squared prediction error (RMSPE) further affirmed the improved fit of EH models for specific parametric distributions.

CONCLUSIONS: As excess hazard may decrease over time, extrapolating excess hazard beyond trial follow-up is easier than all-cause hazards, especially in oncology trials. EH cure models enhance stability and reliability, particularly when cure is plausible.