A Methodology for Estimating Hospital Intensive Care Unit Length of Stay Using Novel Machine Learning Tools
Roberto Williams Batista, Reinaldo Sanchez-Arias · 2020
In this work we investigate the use of electronic health records (EHR) to explore length of stay (LOS) prediction models using machine learning methods. We use the Medical Information Mart for Intensive Care, version 3 (MIMIC-III) dataset, developed by the Massachusetts Institute of Technology (MIT) Lab for Computational Physiology. In this study we focus on the implementation of supervised learning algorithms that classify the patients in three different LOS ranges considering the underlying relationship between the patient hospital stay and several other attributes available in the MIMIC-III dataset. The use of a modern, consistent, and unifying framework for machine learning modeling allows for fair comparison and stable implementations of the different supervised learning techniques considered for LOS estimation. Tree-based models and maximum margin classifiers show reasonable accuracy when predicting patient LOS in a subset of observations for a group of patients with a respiratory disease.