DYNAMIC SURVIVAL MODELS FOR INCORPORATING EXTERNAL EVIDENCE WHEN EXTRAPOLATING OVERALL SURVIVAL- A CASE STUDY.
Author(s)
Kearns B1, Stevenson MD1, Triantafyllopoulos K1, Manca A2
1The University of Sheffield, Sheffield, UK, 2University of York, Heslington, York, UK
Presentation Documents
OBJECTIVES: Generating extrapolations of overall survival is often a key task in economic evaluations. As extrapolation goodness of fit cannot be assessed, considering clinical plausibility is important. We illustrate this using an existing evaluation (nivolumab for previously treated squamous non-small-cell lung cancer) for which two extrapolation approaches had been used. The first extrapolated a decreasing trend in the hazard of overall survival: this was criticised as implausible, as eventually extrapolations fell below that of the age-matched general population hazard. The second approach extrapolated a constant hazard, which was criticised as ignoring the observed trend in the hazard. Cost-effectiveness estimates were sensitive to the extrapolation method used. This study demonstrates how dynamic survival models (DSMs) can provide plausible extrapolations by incorporating general population mortality. METHODS: DSMs are a flexible approach to modelling survival outcomes that combine survival and time-series methods, with time-varying model parameters. When incorporating general population survival, DSMs may be used to extrapolate the excess (disease-specific) hazard. Using the nivolumab evaluation as a case study, three extrapolation scenarios were modelled: a constant excess hazard, or an excess hazard that decreases to either zero (implying cure) or to a constant value. RESULTS: DSMs were able to provide a good description of the observed data, whilst also providing plausible extrapolations that never fell below the general population hazards. Estimates of cost-effectiveness were sensitive to assumptions about the extrapolated behaviour of the excess hazard, with the most favourable estimates (for nivolumab) when long-term cure was assumed, and the least favourite when a constant extrapolated excess hazard was assumed. CONCLUSIONS: With DSMs, the choice of extrapolating model choice can be based on expert clinical opinion about the likely long-term behaviour of the disease-specific excess hazard. These models can provide a good fit to the observed data whilst incorporating external evidence to provide clinically plausible extrapolations.
Conference/Value in Health Info
2019-11, ISPOR Europe 2019, Copenhagen, Denmark
Code
PCN444
Disease
Oncology