Causal Inference & AI for Health-meeting

Causal Inference & AI and its applications in the health domain

Large observational datasets combined with complex models in statistics and machine learning hold promise to inform decision making in the health domain. In many health applications randomizing patient to different treatment to find out what is the best decision is not feasible. Evidence needs to come from big data sources such as electronic health records or patient registries. For instance, when deciding on a new allocation system for donor organs, we want to predict outcomes of patients if they would or would not receive a certain donor organ. To answer such ‘what if’ questions from a registry of patients who were transplanted in the past, a purely predictive AI algorithm does not suffice. We need to combine it with causal inference to find the effects of different actions. Especially in healthcare, it is important that the use of such causal models combined with large datasets leads to intelligible, robust and trustworthy insights and support tools.

Researchers at TU Delft, Leiden University and Erasmus University are joining forces and build collaborations on the emerging topic of causality in AI and its applications in the health domain. Their goal is to bring together researchers on methodology of causal inference across LDE, to see where interests intersect, to get to know each other better and explore opportunities for joint research and joint research funding. This effort is led by researchers Jesse Krijthe (TU Delft), Nan van Geloven (LUMC), Jeremy Labrecque (Erasmus MC) and Marcel Reinders (TU Delft & LUMC).

Jesse Krijthe: “In a medical setting, decisions about treatments and medicines are made every day. We are trying to develop methods to support making better decisions. While the methods apply to many fields, we consider the health domain as an area where this has a very clear added value, and where we can make an important impact.”

Some relevant open issues are: 1. Optimizing complex models for the estimation of individualized treatment effects, and uncertainty quantification of these estimated effects. 2. Constructing procedures that incorporate the sensitivity of inferential claims to the assumptions required for causal conclusions. 3. Developing procedures that combine information from observational with experimental data 4. Studying the role of causality in building stable predictive models, when these predictive models are used to inform decision making. 5. Understanding how causal assumptions can help to construct models that generalize to new environments, and remain stable over time.

If you are a researcher working on causality-research and you find it interesting to attend the first Causal Inference & AI for Health-meeting at 22 November 2022, from 14:30 to 17:00 in Leiden, or you like to stay informed, please contact: