Non-PH parametric survival modelling is developed within the framework of the mul tiple logistic function. The family considered comprises three basic models: (a) a PH
model, (b) an accelerated life model and (c) a model which is non-proportional haz ards and non-accelerated life. The last model, the generalised time-dependent logistic
model was described first by the author in 1996 and this model gives its name to
the entire family. The family is generalised by means of a Gamma frailty extension
which is shown to accommodate crossing hazards data. A further generalisation is the
inclusion of a dispersion model. These extensions lead naturally to the concept of a
multi-parameter regression model described by Burke and MacKenzie in which the
scale and shape parameters are modelled simultaneously as functions of covariates.
Where possible, we include the MPR extension in the XGTDL family. Following a
simulation study, the new models are used to analyse two sets survival data and the
methods are discussed.
History
Publication
Japanese Journal of Statistics and Data Science; 4, pp. 1227–1245