Predicting Hospital Length of Stay in Intensive Care Unit

Abstract

In this thesis, we investigate the performance of a series of classification methods for the Prediction of the hospital Length of Stay (LoS) in Intensive Care Unit (ICU). Predicting LOS for an inpatient in an hospital is a challenging task but is essential for the operational success of a hospital. Since hospitals are faced with severely limited resources including beds to hold admitted patients, prediction of LoS will assist the hospital staff for better planning and management of hospital resources. The goal of this project is to create a machine learning model that predicts the length-of stay for each patient at the time of admission. MIMIC-III database has been used for this project due to detailed information it contains about ICU stays. MIMIC is an openly available dataset developed by the MIT Lab for Computational Physiology, comprising de-identified health data associated with ~40,000 critical care patients at Beth Israel Deaconess Medical Centre. It includes demographics, vital signs, laboratory tests, medications, and more. Different machine learning techniques/classifiers have been investigated in this thesis. We experimented with regression models as well as classification models with different classes of varying granularity as target for LoS prediction. It turned out that granular classes (in small unit of days) work better than regression models trying to predict exact duration in days and hours. The overall performance of our classifiers was ranging from fair to very good and has been discussed in the results. Secondly, we also experimented with building separate LoS prediction models built for patients with different disease conditions and compared it to the joint model built for all patients.