A Hybrid Machine Learning Framework to Predict Early Risk of Mortality in Paralytic Ileus Patients using Electronic Health Records

Abstract

AbstractBackground and ObjectiveParalytic Ileus (PI) is the pseudo-obstruction of the intestine secondary to intestinal muscle paralysis. PI is caused by several reasons such as overuse of medications, spinal injuries, inflammation, abdominal surgery, etc. We have developed an early mortality prediction framework that can help intensivist, surgeons and other medical professionals to optimize clinical management for PI patients in terms of optimal treatment strategy and resource planning.MethodsWe used publicly available ICU database called MIMIC III v1.4, extracted patients that had paralytic ileus as primary diagnosis over the age of 18 years old. We developed FLAIM Framework a two-phase model (Phase I: Statistical testing and Phase II: Machine Learning application) that was compare to traditional methods of machine learning. We used five different machine learning algorithms to test the validity of our Framework. We evaluated the effectiveness of the proposed framework by comparing accuracy, sensitivity, specificity, Receiver Operating Characteristic (ROC) curves, and area under the curve (AUC) for each model.ResultsThe highest improvement in AUC of 7.78% was observed due to application of the proposed FLAIM method. Additionally, almost for all the machine learning models, improvement in accuracy was also observed. With the FLAIM framework, we recorded an accuracy of 81.30% and AUC of 81.38% under support vector machine (with RBF kernel) model in predicting mortality during a hospital stay for the PI patientsDiscussionOur results show promising clinical outcome prediction and application for individual patients admitted to the ICU with paralytic ileus after the first 24 hours of admission.