Non-PH Parametric Survival Modelling
Gilbert MacKenzie (University of Limerick)
Friday 29th May, 2015 15:00-16:00 Maths 204
We trace the development of alternatives to Cox's semi-parametric model. The models are
illustrated by analyzing data from the Northern Ireland Lung Cancer Survival study which identified
and followed-up 855 incident cases.
In particular, we review the derivation of the Generalized Time-Dependent Logistic (GTDL)
regression model which was motivated, in part, with connections to the multiple logistic function.
The GTDL also has deep connections with other well-known statistical distributions.
However, a problem with the GTDL emerged and its resolution via the concept of frailty is demonstrated.
The revised model is shown to perform well in published simulations with competing models. The
Frailty extension leads, unsurprisingly, to multivariate forms of the GTDL.
Finally, the ideas surrounding non-PH parametric models are generalized further using the concept
of Multi-Parameter Regression Survival models (MPRS). The basic ideas are illustrated, briefly, using a
MPRS Weibull model. We anticipate that this class of model will improve practice
in survival analysis.