RSSNI - Talk by Dr Chris Jackson

Abstract: Health policy decisions are often informed by survival data from clinical trials. However, since decisions may have long-term consequences, the policy-maker is typically interested in the expected survival over a longer term than the follow-up of the trials. A common practice is to naively "extrapolate" from parametric models fitted to the short-term data. Many methods have been proposed for using longer-term information in a more principled manner, such as population data, registry data or elicited judgements. However these methods all have limitations in terms of the forms of data required, model flexibility, lack of uncertainty quantification, or difficulty of implementation. I will describe a new method and R package that overcomes these limitations. Individual-level shorter-term data can be combined with any number of longer-term aggregate datasets, under a flexible Bayesian parametric model. The hazard is modelled as a penalised spline function, which can represent potential hazard changes at any time. The effects of explanatory variables can be estimated through proportional hazards or with a flexible non-proportional hazards model. Some commonly-used mechanisms for survival can also be assumed: mixture cure models, additive hazards models with known population mortality, and models where the effect of a treatment wanes over time. Through Bayesian estimation, the model automatically adapts to fit the available data, and acknowledges uncertainty where the data are weak. Therefore long-term estimates are only confident if there are strong long-term data, and inferences do not rely on extrapolating parametric functions learned from short-term data. All of these features are provided for the first time in an R package, "survextrap", in which models can be fitted using standard R survival modelling syntax.