Survival analysis for clinicians

I took a course in “Survival analysis for clinicians” in Rotterdam, the Netherlands, which was part of the Erasmus Winter Programme in 2015. The majority of the around 30 students taking the course were PhD and master students from Holland, but there were also a couple of Germans, plus two from Norway.

I took a course in “Survival analysis for clinicians” in Rotterdam, the Netherlands, which was part of the Erasmus Winter Programme in 2015. Foto: Magn­hild Torske Oust

The lecturer was Hein Putter, who is Professor at the Department of Medical Statistics and Bioinformatics at Leiden University Medical Center.

Putter was an excellent and enthusiastic lecturer. The first half of each day was devoted to computer practicals, and we had lectures the second half of the day. Finding the “perfect” statistics course is always challenging, because you need to find one which is at the level you are currently at. Not too easy, but also not too advanced. The first two days were quite basic, including drawing Kaplan-Meier plots by hand, which I had already done in my PhD statistics courses at NTNU. Still, we went into more detail in this course than the limited number of lectures in Medical Statistics II allowed, allowing a better understanding of the material.

The main bulk of the course was devoted to Cox regression, which was the method I was primarily interesting in learning about. The course provided a useful introduction to Cox regression, ranging from basics, via testing the assumptions on the model, to more advanced topics such as competing risks and multi-state models. Some topics were only briefly covered – not aiming at providing us with enough knowledge to deal with these issues ourselves, but rather to make sure we can identify the problems should we come across them, and consequently know when to ask for help from an experienced statistician. The course also covered some alternatives to Cox regression, including Poisson regression and accelerated failure time models. The course also covered some aspects of Cox regression that were not included in the statistics course at NTNU, including the stratified Cox model and time-dependent covariates, which I found particularly useful. It was also useful to learn about some of the (probably numerous) pitfalls of survival analysis, including a funny example about Oscar winners (no, contrary to popular belief, Oscar winners don’t actually live four years longer than actors who have “only” been nominated for an Oscar, the authors of that famous paper did not take immortal person time into account).

We used SPSS in the computer practicals, which is probably the easiest solution in a class where the students have varying statistical background and use different statistical software packages. However, quite often we were told that “it is not possible to do this in SPSS”, which indicates that SPSS is probably not the software package of choice when performing survival analysis.

All in all, the course is a useful introduction to survival analysis, and Cox regression in particular. It is probably best suited for PhD students who have taken some basic statistics courses, but have not used survival analysis themselves yet.

I would like to thank the Erasmus Medical Center for hosting the Winter Programme, Hein Putter for holding the course, and EPINOR for funding.

Page administrator: Magn­hild Torske Oust
Last updated: 02.02.2017 15:38