Artificial intelligence might prove useful for predicting which hospital patients will soon be discharged, aiding hospital efficiency and patient throughput.
Electronic health record audit logs capture user interactions with patients’ records at a granular level. A first demonstration of how these logs can help power machine learning for 24-hour discharge predictions is reported in the Journal of the American Medical Informatics Association by Xinmeng Zhang, Chao Yan, You Chen, PhD, and colleagues.
The project used data from more than 26,000 adult hospital stays. The team’s machine learning algorithm used information available as of 2 p.m. to predict with 88% accuracy who would and would not be discharged over the next 24 hours (AUC 92%).
Of the 20 predictive factors identified through machine learning as most influential, half were derived from EHR audit logs. The algorithm also used admission diagnoses, historical diagnoses, age, heart rate, BMI, length of stay at prediction point and day of the week, among other factors.
Also on the study were Bradley Malin, PhD, and Mayur Patel, MD, MPH. The study was supported by the National Institutes of Health (GM120484).
