Dieses Tutorial findet virtuell statt.

Rüdiger Pryss

The field of data science has proposed a multitude of new methods, technologies, tools and best practices over the last years. The progress that has been made is utilized in more and more research fields. However, the question emerges very often, which of these findings can help – and how – for a research domain. In this tutorial, two data science methods will be discussed for the field of epidemiology, namely the use of machine learning methods and the utilization of visual analytics. We want to deal with the issue whether these hot and cutting-edge technologies can help in everyday research life of an epidemiologist. The tutorial encompasses a brief introduction and presentation of machine learning methods and visual analytics to show their modes of action. In the context of their operating principles, we will show that a combination of them can be a powerful toolbox for different purposes. On the flip side, it is discussed that they require a lot of training efforts by the citizen developers. The latter term articulates the trend to use powerful IT environments by non-IT experts, which is particularly the case for the topic of the tutorial at hand. On top of the technical discussion, pitfalls and opportunities are presented and evaluated along concrete use cases and conducted studies. At the end, the tutorial tries to conclude with take-home messages and caveats when using machine learning and visual analytics in the field of epidemiology.

Dieses Tutorial findet virtuell statt.

Uwe Sieber (1,2,3,4), Felicitas Kuehne (1), Irene Schmidtmann (5)

1 UMIT – University for Health Sciences, Medical Informatics and Technology, Department of Public Health, Health Services Research and Health Technology Assessment, Hall i.T., Tirol, Austria
2 Division of Health Technology Assessment, ONCOTYROL – Center for Personalized Cancer Medicine, Innsbruck, Austria
3 Center for Health Decision Science, Departments of Epidemiology and Health Policy & Management, Harvard Chan School of Public Health, Boston, MA, USA
4 Program on Cardiovascular Research, Institute for Technology Assessment and Department of Radiology and Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Boston, MA, USA
5 Universitätsmedizin Johannes-Gutenberg-Universität Mainz, Institut für Medizinische Biometrie, Epidemiologie und Informatik (IMBEI), Abteilung Biometrie und Bioinformatik, Mainz, Rhineland-Palatinate, Germany

Whereas “traditional” methods (e.g., stratification, matching, multivariate regression, propensity score), which are appropriate for baseline confounder adjustment, are broadly taught and applied, the more general causal methods (g-methods), which are needed to control for time-varying confounding, are still less known and underused. In addition, emulation of randomized controlled trials and the application of the target trial approach in the planning and conduction of analyses of observational studies gains more and more relevance at the interface between epidemiology, medical decision making and health technology assessment.
This tutorial covers innovative causal inference concepts and methods that are needed for the design and analysis of observational data and pragmatic trials with time-varying exposures or treatments.

We cover the following topics:
1. Introduction to the principles of causation in epidemiology
2. Use of causal diagrams (directed acyclic graphs, DAGs)
3. Brief intuitive illustration of the principles of g-methods: a) g-formula, b) marginal structural models with inverse probability of treatment weighting, and c) structural nested models with g-estimation
4. Application of the target trial emulation concept combined with a counterfactual approach using “replicates” for dynamic treatment regimes
5. Application of g-methods in observational studies and pragmatic trials with post-randomization confounding and selection bias (treatment switching/non-adherence/2nd-line-treatment etc.)
6. Case examples from oncology, cardiovascular disease, HIV, nutrition and other disease areas, illustrating the bias when using “traditional” methods for time-varying confounding

The tutorial will consist of lectures, exercises, case examples drawn from the published literature, and interactive discussion. The intended audience includes researchers from all substance matter fields interested either in methods of causal design/analysis or in the mere interpretation of observational study results. We also give examples and didactical elements guiding those who teach epidemiologic methods and causal principles.
OPTIONAL: Afternoon session for participants bringing their own causal inference cases to be discussed.

Instrumente zur psychosozialen/physischen Belastungen und Beanspruchungen im Kontext der Digitalisierung