Time: 13 February 2019

1 - 2 pm Place: B705

Abstract

Cancer trials pose plenty of statistical challenges, e.g. about missing data, longitudinal tumour measurements, sub-populations and biomarkers, uncontrolled treatment switching, multiple inference and how to optimise the design. We will briefly discuss such aspects and consider where there may be connections to the current research at the Department of Statistics.

 

Of particular importance for drug regulation right now is how to analyse confirmatory clinical survival trials when the relative effect is changing over time. Regulatory authorities have traditionally required logrank tests (sometimes accepting the closely related Cox regression). However, the underlying assumption of proportional hazards is not consistent with many recent trial results. We propose a novel hypothesis test, giving higher power in realistic situations. Furthermore, we discuss wider issues around statistical inference, the role of null and alternative hypotheses and exactly what should be proven for a new drug to be approved.