Modern methods for analysing survival and time-to-event data
Dates: December 4-8 2017
Location: University of Oslo
Lecturers: Ørnulf Borgan (UiO), Odd O. Aalen (UiO), Håkon K. Gjessing (Institute of Public Health/UiB)
Credits: 4 ECTS
Course code: IMB9335
Registration: Closed 1 November
Program: and presentations can be viewed here
Please visit the UiO course webpage here for updated practical information.
The analysis of survival data and other types of time-to-event data are central in modern medical research and a number of other fields. A large number of methods for analysing time-to-event data have been developed over the last decades, but many researchers have no knowledge of survival analysis, or they only know the most basic methods, such as the Kaplan-Meier estimator, logrank test and Cox regression. The aim of this course is to improve this situation by giving PhD-students and other researchers in biostatistics, bioinformatics, epidemiology, and related fields an up-to-date overview of statistical methodology for analysing time-to-event data.
The course starts with a broad introduction of the basic concepts and methods in survival and event history analysis, including methods for handling multiple states/outcome such as competing risks. Then topics of special relevance when analysing biobank data and data with high-dimensional covariates are discussed. We also describe alternatives to Cox regression, particularly useful for non-proportional hazards and time-dependent effects. We consider the effect of unobserved heterogeneity (frailty) in survival analysis, and discuss methods for analysing recurrent events and clustered data. The course concludes with a discussion of causality and methods for causal inference for survival data.
After having completed the course, the students should:
- have an overview over the different study designs that are used for survival and time-to-event data and understand their benefits and limitations,
- have knowledge about the various data structures that occur in studies with survival and time-to-event data and their implications for statistical models and methods,
- know the difference between an individual hazard rate and the population hazard rate and understand the implications this has for interpreting empirical finding,
- be able to identify the appropriate method for a given problem with survival and time-to-event data and perform an analysis of the data using the R software,
- be able to understand and critically assess analyses of survival and time-to-event data as they are typically reported in publications.
Required prerequisite knowledge:
Passed exam in an introductory course in statistics and in a more advanced course in statistics, which includes multiple linear or logistic regression.
Recommended prerequisite knowledge:
The students should have a good understanding of the common statistical models, concepts and methods and experience with using statistics in medicine, biology or similar fields. No background in survival and event history analysis is needed, but familiarity with the basic concepts will be useful. Experience with the R software is recommended, but not required.
The course will be given as an intensive one-week long course (Monday to Friday) and consist of a mixture of lectures (about 60 %) and computer exercises (about 40 %). The plan of the five days is as follows:
- Day1: Introduction to survival analysis; statistical methods for one and more samples (Kaplan-Meier and Nelson-Aalen estimators, log-rank type tests).
- Day 2: Cox regression; competing risks and multistate models.
- Day 3: Cox regression for nested case-control and case-cohort data; Cox regression for data with high-dimensional covariates (lasso, ridge, elastic net).
- Day 4: Additive hazards regression; unobserved heterogeneity (frailty); frailty models for recurrent and clustered data.
- Day 5: Mixed effects models for discrete time survival data; causality and causal inference for survival data.
In the computer exercises the students will analyse given survival and time-to-event data using R, and the students should bring their own laptops with the last version of R and RStudio installed. Some R packages will also be needed, and the participants will be informed about this before the start of the course. The students will receive a reading list before the start of the course and are expected to prepare well.
The exam will be a home exam in the form of project work. The students should deliver their written project report within a month after the end of the course.