Presentation (CTI)

OBJECTIVES: Parametric survival models can predict lifetime outcomes for cost-effectiveness studies. Models typically are fitted to individual-patient data (IPD) using maximum likelihood estimation. Where IPD are unavailable, e.g. in early cost-effectiveness studies, models can be fitted to landmark survival estimates. An existing tool, SurvInt (Gallacher 2024), has focused on standard models with simple hazard functions, which may not adequately represent the complex hazards observed in oncology. Here, we expand on the existing methods by introducing an open-source function for fitting flexible Royston-Parmar (RP) cubic-spline models to landmark survival estimates.
METHODS: An R function was developed to fit RP cubic-spline models to a vector of user-specified landmark survival estimates. Using the optim function and the flexsurvspline RP spline functions, the function identifies the set of parameters that minimise the root mean squared error (RMSE) of the model predictions versus the landmark survival. To test the function, we fitted RP cubic-spline models to landmark survival estimates from publicly available datasets (cancertrials.io) and compared survival projections to the flexsurvspline RP cubic-spline models fitted to IPD from the same datasets.
RESULTS: The function successfully minimised the RMSE (<0.05) across datasets, predicting curves that aligned to the user-defined landmark survival estimates. The predictions performed well when compared to the flexsurvspline outputs. Deviation in the tails of the predicted curves and the IPD curves was present and was sensitive to the position of the final landmark timepoint. However, there was substantial overlap across the two methods.
CONCLUSIONS: Our function expands on an existing tool and increases options for using complex survival distributions where only landmark data are available. This functionality offers users the ability to capture more complex survival functions, reducing the need to rely on simpler models which may fail to capture the true survival trajectory of the disease.