Assistant Professor, Department of Surgery; Core Faculty, Institute for Health Informatics; University of Minnesota, Twin Cities
“Leveraging Natural Language Processing in Health Sciences.”
Wednsday, February 22
Clinical and Translational Research Building
Room 2161, 11 a.m.-12 p.m.
Read more on upcoming seminars from the Department of Health Outcomes & Policy here
A large amount of information in biomedicine and healthcare is recorded as free text, such as clinical notes and biomedical literature. Natural language processing (NLP) techniques provide methods to analyze automatically extract data from text. In biomedicine and healthcare, important NLP applications include secondary use of data from text for research, quality, and literature discovery.
The widespread adoption of Electronic Health Record (EHR) brings rapid text proliferation. The practice of “copying and pasting” of text between notes and incorporation of large amounts of data within clinical notes located in other parts of the medical record results in a large amount of redundant clinical information within clinical records. Redundant information (especially outdated and incorrect information) in clinical notes increase both cognitive burden and decision-making difficulties for clinicians. In a series of phased experiments, redundant information was characterized and methods were developed to help clinicians identify and visualize new information in clinical notes, as well as navigate between notes with more relevant new information.
NLP techniques were also used for generating new knowledge from the large textual data in biomedical literature, including drug-drug interactions (DDIs), drug-supplement interactions (DSIs) and potential prostate cancer drugs. DDIs/DSIs are a serious concern in clinical practice, and many DDIs/DSIs resulting from biomedical pathways are not widely known. Such interactions may be indirectly derived from the scientific literatures. A system was developed to find potential DDIs/DSIs from about 25 million biomedical abstracts. It is also demonstrated that the potential prostate cancer drugs can be exploited by appropriate linking of semantic relationships extracted from biomedical literature.