CLUE: Causal Learning with Unstructured Events
A human life is composed of an astronomical number of events unfolding over time and we unconsciously use abstractions and models on a daily basis to structure and derive meaning from these to navigate in life. Similarly, our health and welfare systems systematically record events with the purpose of prevention, diagnosis, prognosis and treatment to optimize health and well-being. Unfortunately, passively observing how events unfold does not answer the fundamental question of causality among events. Why does prostate enlargement, for instance, evolve into cancer for some and not others? The statistical proverb “correlation is not causation” expresses that the path to answering the why-question from observational data is fraught with difficulties. My project aims to go beyond the current state-of-the-art in causal learning by using all available clues from unstructured event data. The new methods will be particularly relevant to epidemiology, patient trajectories, co-morbidities, and problems within the social sciences.