Deep learning-based recurrent prediction of delirium in the intensive care unit
Joon Lee, PhD
Associate Professor of Health Data Science |
Depts of Cardiac Sciences and Community Health Sciences | Cumming School of Medicine | University of Calgary
WHERE: Hybrid | 2001 McGill College, Rm 1140 | Zoom
Note: Joon Lee will be presenting from U of Calgary
Delirium is common in the intensive care unit (ICU) and associated with longer ICU and hospital stays as well as worse patient outcomes including long-term cognitive impairment. Early prediction of impending delirium can lead to efficient allocation of ICU resources and improved patient outcomes via preventive care. Based on rich clinical data from over 43,000 ICU admissions in Alberta, we developed deep learning-based models capable of predicting delirium in the next two 12-hour windows, with new predictions generated every 12 hours. Our best model based on gated recurrent units resulted in areas under the receiver operating characteristic curve around 0.9.
By the end of this session, attendees will:
- Understand delirium as a major clinical problem in the intensive care unit;
- Learn how recurrent deep learning can be utilized to predict impending delirium in the intensive care unit;
- Appreciate the challenges surrounding feature engineering and hyperparameter tuning when granular electronic health record data are used.
Dr. Joon Lee is the Director of the Data Intelligence for Health Lab and an Associate Professor of Health Data Science in the Departments of Cardiac Sciences and Community Health Sciences, Cumming School of Medicine, University of Calgary. He holds a PhD in Biomedical Engineering from the University of Toronto and a BASc in Electrical Engineering from the University of Waterloo. He also completed a Postdoctoral Fellowship in Medical Data Science at MIT. His research applies data science, machine learning, and artificial intelligence to a variety of problems in medicine and public health including intensive care, cardiology, public health surveillance, and food marketing.