A Machine Learning Approach for Predicting Inpatient Discharge at the Central Maine Medical Center
Abstract
Operating with a finite quantity of beds, medical resources, and physicians, hospitals are constantly allocating resources under conditions of scarcity. Misallocation of resources and operational inefficiencies are a substantial driver of the United States’ strikingly high healthcare costs. Accurately forecasting the duration which a specific patient will stay in a hospital, also known as a patient’s length of stay, could assist hospital decision makers in optimizing their workflow and allocating their resources efficiently. This paper demonstrates the superiority of a survival random forest approach over classical econometric techniques and current practice at the Central Maine Medical Center. Included in the discussion is an assessment of the strengths and weaknesses of the model, with the hope of informing the application of machine learning methods in the real world.