Non-Proportional Hazards: Modelling the Restricted Mean Survival Time using R
In survival analysis in medical research, the proportional hazards assumption and the hazard ratio effect measure have been popular for decades, fuelled by extensive application of the log-rank test and the Cox regression model. However, the hazard ratio can be clinically awkward to interpret, or the proportional hazards assumption may not hold, rendering the use of a hazard ratio effect measure questionable at best.
In this course we introduce the restricted mean survival time (RMST), which is a well established, but under-used summary of the survival experience. In recent years there has been a surge of interest in the RMST, particularly in oncology, but also in many other areas. We begin with a review of the definitions of the RMST, approaches to estimation and different RMST-based effect measures which are clinically meaningful alternatives to the hazard ratio and are not based on a proportional hazards assumption.
For the practical analysis of survival data, which includes right-censoring, the course focuses on a non-parametric analysis for comparing treatments and a generalised linear model (GLM) -type modelling approach based on the use of pseudo-values. The latter provides a flexible method for directly modelling the RMST in a regression framework, where a treatment effect may be adjusted for covariates. Model checking is also considered.
The course is a practical introduction to analysing survival data using a RMST-based effect measure. Only essential theoretical aspects of the methodology will be summarised. Examples used will be drawn from applications in medicine and health, particularly clinical trials.
Practical work will be based around the statistical software R; see https://www.r-project.org/.
All training is online and will be delivered live between 09:00 and 17:30 (GMT+1). Delivery platform: Zoom, which may be freely accessed. Questions may be asked using Zoom's chat box. Note our online courses are delivered by a team of two presenters, meaning at least one presenter is always available to provide additional support. During presentations, the team member who is not speaking can take questions in addition to the presenter. We also use Zoom meetings rather than webinars to encourage further interaction during an online course.
Who Should Attend?
Statisticians and data analysts working with survival data in medical research. Participants will be assumed to have a working knowledge of:
How You Will Benefit
You will acquire practical experience in the use of RMST-based effect measures as an alternative to the hazard ratio. You will also be able carry out adjusted as well as unadjusted analyses.
What Do We Cover?
Practical work will be done in R.
Note: For practical work, participants must download and install a number of CRAN packages in R. This must be done prior to the start of the course.
Related courses: Advanced Survival Analysis using R; Advanced Topics in Survival Analysis; Introduction to Survival Analysis; Survival Analysis for Medical and Health Professionals using R.