PROJECT TITLE :

Improving the In-Hospital Mortality Prediction of Diabetes ICU Patients Using a Process Mining/Deep Learning Architecture

ABSTRACT:

Patients in the intensive care unit (ICU) who have diabetes have an increased risk of complications that could lead to death while they are still in the hospital. Due to the large number of factors that can have an effect on the outcome, calculating the probability of dying is a difficult and time-consuming task. ICU patients who are considered to be at a higher risk are of particular interest to healthcare providers because this allows for the possibility of risk factors being reduced. Although such severity scoring methods do exist, they are typically based on a snapshot of a patient's health conditions taken during their stay in the intensive care unit (ICU), and they do not take a patient's prior medical history into specific consideration. In this article, a process mining and deep learning architecture is proposed with the goal of enhancing previously established severity scoring methods by incorporating the medical histories of diabetes patients. First, the patient's medical history from their previous visits to the hospital is converted into event logs that can be used for process mining. After that, the event logs are used to find a process model that describes the previous hospital encounters that patients have had. It has been suggested that an adaptation of Decay Replay Mining be used to combine medical and demographic data with previously established severity scores in order to predict the in-hospital mortality of diabetes intensive care unit patients. Using the Medical Information Mart for Intensive Care III dataset, significant performance improvements are demonstrated in comparison to previously established risk severity scoring methods and machine learning approaches.