Course overview
The overall aim of this course is to give an overview of concepts and
methods for estimating causal effects of treatments on time-to-event
outcomes. Topics covered will include time-dependent treatments and
confounding, inverse probability of treatment weighting and marginal
structural models, g-formula, censoring-and-weighting approaches and
competing risks.
The course will combine lectures and computer sessions. There will be
an emphasis on practical implementation, and all analyses and exercises
will be performed using R. At the end of the course, participants should
be equipped to perform analyses to address questions about causal
effects of treatments on their own data.
Content
Over the recent decades we have seen major developments in methods
for causal inference using observational data, but the practical
application of these methods can be challenging, especially in the
context of survival and other time-to-event outcomes.
This course will cover methods for confounding adjustment with
time-to-event outcomes, including inverse probability of treatment
weighting and marginal structural models, g-formula, and
censoring-and-weighting approaches. We will cover settings in which the
treatment of interest is given at a single time point, as well as more
complex situations in which interest lies in time-varying treatments.
Focus will be on the choice of appropriate estimands and links will be
made to the target trial framework. Most of the emphasis will be on
simple survival outcomes, but extensions to competing risks will also be
discussed. Much of the material will be presented with medical and
epidemiological applications in mind, but the methods are equally
relevant in other areas, such as demography, social sciences and
economics.
Who is this course for?
This course is aimed at researchers and students in biostatistics,
statistics, epidemiology and related fields. Some knowledge of methods
for survival analysis will be assumed, such as experience with
estimation of survival curves using the Kaplan-Meier estimator and
fitting regression models to time-to-event data. Participants should
have some prior experience of using R, though not necessarily in the
context of time-to-event analysis or causal inference. Knowledge of
causal concepts would be beneficial, but we will not assume this.
Schedule
All times are in GMT+2.
Day 1
| 09:15 - 09:30 |
Welcome and introductions |
|
Part 1: Introduction and estimation of the effects of point
treatments on survival
| 09:30 - 10:15 |
Lecture: Causal estimands for time-to-event outcomes
and identification using observational data |
Jon Michael Gran |
| 10:30 - 11:15 |
Lecture: Weighting and standardisation for point
treatments |
Ruth Keogh |
| 11:15 - 12:45 |
R practical: Weighting and standardisation for point
treatments |
Ruth Keogh |
Part 2: Extensions to time-dependent treatment
strategies
| 13:45 - 14:30 |
Lecture: Extensions to time-dependent treatment
strategies, and the challenge of time-dependent confounding |
Jon Michael Gran |
| 14:30 - 16:00 |
R practical: Artificial censoring and censor
weighting |
Jon Michael Gran |
Day 2
Part 3: Further methods for time-dependent treatment
strategies
| 09:30 - 10:15 |
Lecture: The sequential trials approach and marginal
structural models |
Ruth Keogh |
| 10:30 - 11:15 |
Lecture: The g-formula |
Ruth Keogh |
| 11:15 - 12:45 |
R practical: Estimation using the more advanced
methods |
Ruth Keogh |
Part 4: Extensions to more complex time-to-event
outcomes
| 13:45 - 14:30 |
Lecture: Estimating causal effects on time-to-event
outcomes under competing risks |
Jon Michael Gran |
| 14:30 - 16:00 |
R practical: Extensions of previously introduced
methods to competing risks |
Jon Michael Gran |