Hospital care accounts for a large part of ever increasing healthcare expenditures in Switzerland. Consequently, expectations for hospital management to decrease costs have been high. Simultaneously, hospitals are expected to continuously improve the quality of care they provide. Research suggests that optimizing inefficient hospital processes is a key lever to meet both challenges.
Key hospital process areas, such as the intensive care unit (ICU), are process bottlenecks prone to inefficiencies. Key reasons for this are uncertainties in demand (e.g., number of patients arriving at an ICU in a given time period) and supply (e.g., process throughput times). While the topic of hospital process bottleneck management has been researched comprehensively, conventional operations research models are limited in dealing with process uncertainty, their algorithms lack flexibility, and they are limited in their general applicability.
In this project, we aim to address hospital process bottleneck management in a novel way, i.e., with machine learning techniques from the class of doubly robust learners and (offline) policy learners, e.g., causal forest and policy tree. These machine learning techniques can effectively address shortcomings of current state of the art models.
Together with our partner, the Cantonal Hospital of St. Gallen (Kantonsspital St. Gallen, KSSG), we are working on training a model to supply individual ICU readmission risks supporting clinical decision making as a first application example. Specifically, we aim to reduce ICU readmissions by individualizing discharge decisions using conditional average treatment effects supplied by causal forest.
Dr. Justus Vogel, Johannes Cordier, Daria Bukanova-Berend, David Klug
Kantonsspital St. Gallen (KSSG)
HSG Stepping Stone Program
15 months (01.05.2024 – 31.07.2024)