Modelling Time-to-Death Utilities as a Continuous Function, Using a Rich Dataset of Patients With Non-Small Cell Lung Cancer

OBJECTIVES: Utility estimates are a key input for economic evaluations. The approach of modelling utility values as time-to-death (TTD) health states was developed in response to perceived shortcomings of the progression-based approach in oncology. We present an analysis of data from patients treated with nivolumab for advanced non-small cell lung cancer (aNSCLC), modelling utility values as a continuous function of TTD.

METHODS: Utility values showed a clear decline in the 6 months prior to death, with no clear groupings observed visually. The two preferred models out of six evaluated were in the form 1/square root of TTD and log(TTD). These models yielded a good visual fit as well as the smallest MAEs (0.177 and 0.178) and QICu (3628 and 3628) values. The continuous time models compare favourably with the use of previously published discrete TTD categories (Chaudhary ISPOR EU 2020; PCN 163) in having similar MAE (0.178) but lower QICu (3631) and a more parsimonious model.

RESULTS: Utility values showed a clear decline in the 6 months prior to death, with no clear groupings observed visually. The two preferred models out of six evaluated were in the form 1/square root of TTD and log(TTD). These models yielded a good visual fit as well as the smallest MAEs (0.177 and 0.178) and QICu (3628 and 3628) values. The continuous time models compare favourably with the use of previously published discrete TTD categories (Chaudhary ISPOR EU 2020; PCN 163) in having similar MAE (0.178) but lower QICu (3631) and a more parsimonious model.

CONCLUSIONS: The use of a continuous function avoids the need for arbitrary grouping of time to death and provides a more generalizable method to implement health state utilities. Further research is required to compare continuous TTD utilities to those derived from progression related states, as well as validation in other oncology datasets.