A team of researchers led by Yonghui Wu, Ph.D., in the Department of Health Outcomes & Biomedical Informatics (HOBI) ranked third in a clinical natural language processing challenge for detecting medication and adverse drug events (MADE) from electronic health records. Other members of the team included Jiang Bian, Ph.D., an assistant professor, and Xi Yang, Ph.D., who are both on the biomedical informatics team in HOBI.
The competition, organized by the University of Massachusetts Medical School in Worcester, was designed to āfurther advance adverse drug event detection techniques to improve patient safety and health care quality.ā
Adverse drug events (ADEs) are common, occurring in approximately two to five percent of hospitalized adult patients. Each ADE is estimated to increase health-care costs by more than $3,200. Severe ADEs rank among the top five or six leading causes of death in the United States.
āAdverse drug events are critical for patient safety,ā Wu said. āEarly detection of adverse drug events from electronic health records provides an efficient way of conducting drug safety surveillance.ā
The HOBI biomedical informatics team collaborates with researchers and clinicians throughout the UF Health system and UF to conduct research using clinical natural language processing to extract information from unstructured clinical text in EHR data.
The MADE challenge consisted of three tasks: (1) a named entity recognition (NER) challenge to detect mentions of a medication name and its attributes, and mentions of ADEs, drug indications, and other signs and symptoms from clinical notes; (2) a relation extraction challenge to identify relations between a medication and its attributes, relations between medications and ADEs, indications and other sign & symptoms; and (3) an end-to-end task that combines tasks 1 and 2.
A total of 23 natural language processing systems were developed by teams from 10 universities and research institutes, including the University of Florida, University of Utah, University of Arizona, IBM Research, the University of Nice Sophia Antipolis, the Australian e-Health Research Centre, and others.
The deep learning-based natural language processing system developed by the HOBI team ranked among the top three in Task 1.