OBJECTIVES: Extrapolation for cost-effectiveness models is usually performed by fitting standard parametric distributions to Kaplan-Meier (KM) curves. However, in some therapeutic areas, patients can eventually be cured, which leads to a plateau in the KM curve. Our goal is to provide an overview of the methods that can be used to account for cured patients in cost-effectiveness models.

METHODS: A targeted literature review was performed on Embase and Pubmed to identify the main methods used to deal with plateau in KM curves. We digitalized KM curves from a RCT in advanced melanoma (NCT01844505) using Guyot algorithm. We applied the identified methods and standard parametric methods on the derived survival data. We then integrated these extrapolations in a cost-effectiveness model and assessed the sensitivity of the results to the different methods.

RESULTS: The main methods identified were restricted cubic splines (RCS), mixture cure models (MCM), non-mixture cure models and frailty models. For each method, we selected the distribution which was minimizing AIC and BIC. Extrapolations were sensitive to the method used. For instance, in arm A, the best standard parametric model (generalized gamma distribution) predicted a PFS of 1 % at 20 years [range of all standard distributions: 0.0%-7.2%], whereas RCS predicted 5.7%, MCM predicted 8.2% and MCM with background mortality predicted 6.2%. Restricted mean survival time at 20 years was 13 months [range: 8-24] for the best parametric model, 21 months for RCS, 25 months for MCM and 24 months for MCM with background mortality. Cost effectiveness results will be provided for the conference.

CONCLUSIONS: Our study showed that the survival method used had a significant impact on mean survival time, which is a key driver of the ICER in oncology. Non-standard parametric models should be taken into consideration when extrapolating survival in a cost-effectiveness model in the therapeutic areas where cure is possible.