Beyond the hazard ratio: alternative ways to quantify differences in survival when using flexible parametric survival models
Paul Lambert, Professor of Biostatistics
Department of Health Sciences, University of Leicester
When analysing time-to-event data the most common measure of effect is the hazard ratio from a model that assumes proportional hazards. I will describe flexible parametric survival models with the aim of demonstrating the ease at which alternative measures of effect can be obtained. I will discuss how it is possible to relax the proportional hazards assumption by including interactions between covariates and the effect of time. Rather than reporting hazard ratios I will advocate the use of alternative measures such as differences in hazard rates, the use of restricted mean survival (and differences), differences in survival proportions, the loss in expectation of life and “avoidable” deaths.
Directions from ground floor of Worsley Building: Go through the main entrance and take the lift to level 9. From the lift turn left, go through a set of double doors. Go along the corridor, SR 9.59 is the second room on the right