Carter Phillips
Title: Methods for Variable Selection in Weibull Mixture Cure Models
Date: July 29th 2026
Time: 3:00pm
Location: Zoom
Supervised by: Rachel Altman
Abstract:
Weibull mixture cure models are parametric survival time models that are used when only a fraction of the population is susceptible to the event of interest. Existing variable selection methods in this setting can perform poorly. We therefore introduce adaptations of three methods for variable selection (exact branching, approximate branching, and leaps and bounds) in the Weibull mixture cure model context. The exact branching method is guaranteed to find the best model of each size but can be computationally intensive. The approximate branching method considers only promising models, decreasing the computing time at the cost of occasionally missing the best model. The leaps and bounds method efficiently selects a model by using a quadratic approximation to the log-likelihood function, but its performance is subject to the accuracy of this approximation. We compare the performance of these methods to that of preexisting methods 鈥� and determine factors that impact the performance of our methods 鈥� via a comprehensive simulation study. We use our results to determine important predictors of time to diagnosis of mental health problems in germ cell tumour survivors.