Selected Talks


In this talk I introduce some new developments to the -survsim- package in Stata, going through examples of how to simulate time-to-event data from parametric distributions, custom distributions, competing risk models and general multi-state survival models.

I spoke with Chuck Huber for Stata Happy Hour about my career in biostatistics, academia, and all things Stata

Multi-state models are receiving substantial interest in medical research, as attention is shifting from studying a single trajectory between two particular health states, to describing entire disease pathways and co-morbidities. By modelling transitions between disease states, accounting for competing events at each transition, we can gain a much richer understanding of patient trajectories and how risk factors impact over an entire disease pathway. In this talk, I’ll describe some new developments, with a focus on providing interpretable measures of risk and disease burden, regardless of the underlying transition hazard models. The transition models can be as simple or complex as required for each transition (anything from an exponential to a spline-based approach with time-dependent effects). I will illustrate how to derive analytic transition probability and length of stay predictions for the illness-death and extended illness-death settings, and implement a general simulation algorithm for all other transition matrix structures, providing broad applicability. The simulation approach facilitates derivation of extended predictions such as the probability of ever visiting a state, the population attributable fraction, and more. This provides easily understood predictions to describe results to patients, clinicians, and decision makers, alike. Finally, I will briefly present some ongoing work, including standardising prediction over the observed covariate distribution to obtain marginal estimates, incorporating multiple timescales to the transition hazards, and touch on the issue of interval-censoring. User-friendly Stata software is provided for both estimation and prediction, and will be illustrated throughout the talk through application to a breast cancer example.