When: 4 november, 2020, kl. 13-14 Where: This seminar is given online. E-mail Dan Hedlin if you want to attend.
Abstract
In survival analysis, the choice of an appropriate time-scale is essential. Common time-scales are time-since-diagnosis in studies investigating prognosis of a certain disease, or attained-age when studying the incidence of disease. However, sometimes there are several time-scales that act on the rate, for instance attained age and time since pregnancy for the incidence of breast cancer, or time since initial diagnosis and time since relapse when studying death among relapsed cancer patients. Two (or more) time-scales can be implemented in Cox or Poisson models, by splitting the data to create non-overlapping time intervals. This can lead to large datasets and the analysis can be computationally intensive. I will present flexible parametric survival models that incorporate multiple time-scales, where one time-scale is the main time-scale and the other is a function of the first time-scale and included as a time offset. Flexible parametric survival models use splines for modelling the log (cumulative) hazard function, and are a useful alternative to Cox regression since it gives a parametric formulation of the hazard function without forcing a strong parametric assumption. I will show preliminary results from a simulation study where we investigate when multiple time-scales are needed, as well as examples of how results from studies with two time-scales can be presented.