Reducing waiting times and crowding in hospital emergency departments using Machine Learning

Emergency department physicians receive patients in a wide range of conditions and must be able to take sensitive decisions in a small amount of time. Lack of resources, in the form of medical personnel, diagnostic tools, and beds, as well as improper emergency room access, often translate to high emergency room crowding. In addition to the lower perceived quality of service, emergency room crowding and, in particular, excessive waiting times, are linked to major risks for the patient health (complications, readmissions, leaves without being seen, greater hospital length of stay etc) and higher mortality rate. Analyzing the case of the \textit{Consorci Sanitari de Terrassa}, in agreement with which this work has been developed, we propose a methodology to predict hospital admission from the emergency department using machine learning, right after triage. Being able to anticipate hospitalizations at a so early stage could theoretically allow eliminating waiting times between the physician final decision and the moment the patient is actually admitted in a hospital room. Starting from a baseline which only uses data produced in the emergency room, this thesis shows that integrating data from the patient's medical history considerably improves the model's predictive power. A first methodology to remove trivial cases is also discussed. The obtained results show that a prediction could be obtained with a nontrivial accuracy even when using highly interpretable models.