Parametric competing risks and multistate models

Parametric competing risks
and multistate models

Prof. Paul Lambert
Department of Health Sciences
University of Leicester, United Kingdom
Department of Epidemiology and Biostatistics
Karolinska Institutet Stockholm, Sweden
https://pclambert.net/

Michael J. Crowther
Lecturer in Biostatistics
Biostatistics Research Group
Department of Health Sciences
University of Leicester, United Kingdom
www.mjcrowther.co.uk

Course description

This course will focus on the use of parametric survival models when analysing data with competing risks and then extending to multi-state models. Multi-state models are increasingly being used to model complex disease profiles. By modelling transitions between disease states, accounting for competing events at each transition, we can gain an improved understanding of patients’ prognosis and how risk factors impact over the whole disease pathway. I will place emphasis on the use of flexible parametric survival models that incorporate restricted cubic splines on the log hazard or log cumulative hazard scale. This will include models with time-dependent effects (non-proportional hazards). We will use an efficient and generalizable simulation method to obtain clinically useful and directly interpretable predictions, which are particularly useful for more complex models. We will also discuss assumptions of the models, including the Markov assumption and how this can be relaxed. The course will be taught using Stata making use of the multistate package. The course will discuss the theory, but emphasis will be placed on applying and interpreting the methods.

Given the limited time there will not be computer lab sessions on the day, but some example questions will be provided to participants together with solutions and Stata code. I will give a demonstration of some of these questions to highlight some of the key features of using these models.

Target audience

The course is aimed at statisticians and epidemiologists with an interest in modelling survival data. The primary focus of the course is on statistical methods, but a degree in statistics or mathematical statistics is not essential. Some previous knowledge of survival analysis would be useful, for example, understanding of survival/hazard functions and experience of using the Cox model and/or the Royston-Parmar flexible parametric survival model. However, I will spend time on application of the methods and explanation of key concepts so that those with a less formal statistical training can gain an understanding of the methods and their interpretation.

Course Timetable

Registration (08:30-08:55)

09:00–10:00 (Paul C. Lambert)

• Introduction and welcome to the course
• Brief review of the Cox model and parametric survival models
• Flexible parametric survival models

10:00–10:30   Coffee Break
10:30–11:30 (Paul C. Lambert) 

• Modelling competing risks
• Estimating cumulative incidence functions
• Example question

11:30–12:30 Lunch

12:30–14:00 (Michael J. Crowther)  

• Introduction to multi-state survival models
• The illness-death model
• The Markov assumption

• Using simulation to obtain clinically useful measures for multi-state models

14:00–14:30    Coffee Break
14:30–16:00 (Michael J. Crowther)    

• Expected length of stay in different states and other extended predictions
• Resetting the clock and semi-Markov models
• Example question
• Course wrap up / summary