9/27 – Houshang Darabi, University of Illinois Chicago
September 27, 2021
12:00 PM - 1:00 PM
CalendarDownload iCal File
Title: Process Mining/Deep Learning of Electronic Health Records for Next Event Prediction: Modeling, Challenges, and Accomplishments
Abstract: What is the mortality probability of a diabetes patient after admission to the ICU? What is the likelihood of the 30-day unplanned readmission of a heart failure patient after being discharged from the hospital? What is the 72-hour mortality risk of a patient with acute kidney injury after admission to the ICU? These and many other similar questions are common in patient workflow settings, and are classified as next-event prediction problems. Patient workflow settings exist in almost every healthcare facility. Several tools/techniques such as simulation, severity score calculators, machine learning, and classical optimization methods have been used to solve the next event prediction problems associated with patient workflow systems. However, these techniques have major drawbacks that may reduce their prediction efficiency and effectiveness. Ignoring the complete medical history of patients available through health records, overlooking the time dimension of events, and inability to incorporate sequential and prerequisite rules in the event prediction algorithms are some of these drawbacks. In this presentation, we propose a general framework for the next event prediction problem in patient workflow systems. The framework allows for the integration of several proven-to-work tools such as process mining, deep learning and information reduction algorithms. We show the superior performance of the proposed framework over the existing techniques on multiple real world next event prediction problems in patient workflow systems. We also show the challenges associated with the proposed framework and future research opportunities that might be pursued from here.
Sep 14, 2021
Sep 14, 2021