Simultaneous visualisation of a time-to-event outcome and multivariate mixed type covariate data
Featuring Dan Lythgoe, Principal Statistician at PHASTAR
Abstract: 'Simultaneous visualisation of a time-to-event outcome and multivariate mixed type covariate data'.
The ever-increasing complexity and dimensionality of clinical data sets means that the role of effective data visualisations to understand and interpret clinical data is more important than ever. Biplots are a multivariate generalisation of scatter plots that can be used to visualise the key features of clinical data sets in a low-dimensional space. In this presentation, it will be shown how multidimensional scaling biplots can be constructed using Gower's coefficient for mixed-type covariate data, and how time-to-event data can be incorporated. Specifically, it will be shown how a bi plot can be supplemented with a time-to event axis using a log-linear accelerated failure time model, and it will be demonstrated why parametric accelerated failure time models are particularly well-suited to this purpose. As an example, a time-to-event biplot is constructed for a hepatocellular carcinoma data set in which in the relationships between observations, variables and a time-toevent outcome are illustrated simultaneously.
About the speaker:
Dan is a Principal Statistician at PHASTAR with 10 years' experience working as a statistician in clinical trials. Dan recently completed his part-time PhD with the University of Liverpool based on time-to-event modelling with latent variables, is a committee member for the Royal Statistical Society Medical Section and chairs the PHASTAR Statistics Forum.