Forecasting Mortality Risk for Patients Admitted to Intensive Care Units Using Machine Learning

Providing an accurate prediction of mortality risk for patients in the health care system as early as possible could help improve care quality and reduce costs. The recent adoption of Electronic Health Records (EHRs) has created an opportunity to improve prediction accuracy by obtaining more detailed data and applying more advanced algorithms. In this work, we applied gradient boosted trees and deep neural networks to estimate the mortality risk of patients admitted to a single institution’s Intensive Care Units (ICUs). Unlike the prior studies that utilize a rich set of features usually available at discharge, we used commonly available data in EHRs at the time of admission. We used a subset of MIMIC III of 4,440 admissions between 2001 and 2012 to extract demographic information (age, gender, marital status, etc.), diagnosis and medical (ICD) codes. The model achieved 87.30 AUC on the 10% test set. The excellent performance of the model highlights the usability of models with few features for building real-world prediction engines.