Presentation (CTI)

OBJECTIVES: Flexible parametric non-mixture cure models (fp-nMCMs) are methods used to estimate complex hazards with a “cure” time-point, dictated by the placement of a boundary knot. To our knowledge, these have not yet been used in health technology assessments (HTAs) but may allow more accurate extrapolations of survival data beyond an observed trial period. Published guidelines for deriving outputs exist but practical guidelines for implementation into a HTA cost-effectiveness model (CEM) remain limited. With no published precedent for their direct implementation in Microsoft Excel-based CEMs, there could potentially be a barrier preventing potential HTA submissions including fp-nMCMs.
METHODS: A targeted literature review (TLR) of fp-nMCM methodology approaches was conducted. Key concepts relevant to cost-effectiveness modelling were summarised, including the estimation of cure fractions alongside extrapolated survival and hazard functions. Model coefficients from R were investigated to translate the statistical coefficients and knot placements into survival functions suitable for HTA CEMs implemented in Excel.
RESULTS: Several packages were identified from the TLR that can be used to derive fp-nMCMs in R, including flexsurv, cuRe, and rstpm2. The rstpm2 statistical package was used to derive knot coefficients in time-to-event data. Coefficients are then implemented into an example CEM developed in Excel, including how uncertainty may be incorporated using a variance-covariance matrix. Knots and their coefficients are used to estimate the relative survival function using the MMULT function in Excel. Expected survival is estimated using Office of National Statistics lifetables and then all cause (overall) survival is estimated by multiplying the relative survival and expected survival functions.
CONCLUSIONS: fp-nMCMs represent a flexible and potentially useful approach to HTA submissions where cure is a possibility, particularly when a “cure” time-point is relevant and when observed hazards have multiple turning points. Clear interpretation of R-generated outputs and transparent integration into CEMs may improve survival extrapolation credibility informing long-term cost-effectiveness.