Keynote Speakers
Hein Putter
Hein Putter is Professor of Medical Statistics at the Department of Biomedical Data Sciences of the Leiden University Medical Center (LUMC). He began his academic journey with a PhD in Mathematical Statistics at Leiden University, and before joining LUMC in 2000, he held postdoctoral positions at the University of Amsterdam, Free University Amsterdam, and Cambridge University. His research focuses on statistical methods for complex survival data, including multi-state models, competing risks and frailty models, as well as on methods for dynamic prediction in clinical research.
Understanding time-related bias in survival analysis
Time-related biases are common in longitudinal studies, yet they often appear unrelated and are typically addressed in isolation. Examples of such time-related biases are immortal time bias, delayed or confirmed diagnosis, outcome-dependent inclusion or follow-up, and time-dependent confounding. The aim of this talk is to provide a common framework for understanding, and thereby hopefully also preventing, such biases. We give examples of these biases and show that they all arise from a single underlying issue: a misalignment between the scientific question, the true biological time-to-event process, how events and covariates are observed, and how the statistical model interprets them. We introduce a simple framework that separates these four components and use it to classify time-related biases into three recurring problems: event definitions that do not correspond to the biological onset; risk sets that do not reflect who is genuinely at risk at each moment; and time-varying covariates that influence treatment decisions while also being affected by treatment.
Ingrid Van Keilegom
Ingrid Van Keilegom is Professor of Statistics at the Faculty of Economics and Business of KU Leuven, and part-time Professor at UCLouvain. Her academic career started with a PhD in Statistics at Hasselt University, followed by appointments at Pennsylvania State University, Eindhoven University of Technology and UCLouvain. Her research interests include survival analysis, measurement errors, quantile regression, non- and semiparametric regression techniques, instrumental regression, and mathematical statistics.
Survival analysis under label shift
Let π represent the source population with complete data, containing covariate Z and response π, and π the target population, where only the covariate Z is available. We consider a setting with both label shift and label censoring. Label shift assumes that the marginal distribution of π differs between π and π, while the conditional distribution of Z given π remains the same. Label censoring refers to the case where the response π in π is subject to random censoring. Our goal is to leverage information from the label-shifted and label-censored source population π« to conduct statistical inference in the target population π¬. We propose a parametric model for π given Z in π and estimate the model parameters by maximizing an approximate likelihood. This allows for statistical inference in π and accommodates a range of classical survival models. Under the label shift assumption, the likelihood depends not only on the unknown parameters but also on the unknown distribution of π in π and Z in π, which we estimate nonparametrically. The asymptotic properties of the estimator are rigorously established and the effectiveness of the method is demonstrated through simulations and a real data application. Moreover, our approach remains novel even in the absence of censoring, contributing fresh insight into the study of label shift. This work is the first to combine survival analysis with label shift, offering a new research direction in this emerging topic.
Mats J Stensrud
Mats Julius Stensrud is Associate Professor of Biostatistics at the Institute of Mathematics of the Γcole Polytechnique FΓ©dΓ©rale de Lausanne (EPFL). He is a medical doctor by training and previously worked as a practicing physician. He obtained his PhD in Neuroscience from the University of Oslo and has held positions at the Harvard School of Public Health and the University of Oslo. His research focuses on developing methods for causal inference in settings with longitudinal data, with applications in medicine and epidemiology.
Causal effects conditional on post-treatment variables
Many studies aim to estimate treatment effects on outcomes that are defined only for individuals who experience a post-treatment event. For example, the effect of cancer therapies on quality of life is only well defined among individuals who are alive. Similarly, the effect of vaccines on post-infection outcomes is only of interest among individuals who become infected. A naive comparison of outcomes conditional on such post-treatment events generally lacks a causal interpretation, even when treatment is randomly assigned.
In this talk, I discuss causal contrasts for outcomes conditional on post-treatment events, including principal stratum effects and conditional separable effects. I then derive identification results for these estimands and discuss their interpretation and the conditions under which they might, in principle, be falsified. I illustrate the relevance of these results for clinical trials of cancer therapies and vaccines, and I conclude by revisiting the classical birth weight paradox in epidemiology.