Natural language processing (NLP) is very widely used in clinical informatics research, but its utility in the operations world stands in doubt. While it has been proven time and again that a significant amount of clinical information is locked in narrative reports, and many clinical systems and processes (from coding and billing, to clinical decision support, to quality metrics monitoring, and so forth) can drastically benefit from converting this information into computable form, in reality the road from narrative reports to structure information is often perceived as too long and convoluted, and examples of live NLP systems that are used in routine operation are few and hard to generalize.
In this webinar, we will share our experiences on utilizing various NLP tools, including rule-based and statistical systems, and will discuss when and how NLP tools can be very easy to use (much more so than most believe) and when they cannot meet our expectations.
After participating in this activity, the learner should be better able to:
- Explain the main components of natural language processing (NLP)
- Explain major limitations of NLP in clinical domain
- Critically appraise different NLP systems based on their need
Hojjat Salmasian, MD, MPH, PhD - Program Director for Research Science - Value Institute, NewYork Presbyterian Hospital
Rimma Pivovarov, PhD - Value Institute, NewYork Presbyterian Hospital
Dr. Hojjat Salmasian uses his training as a clinician and informatician to design and evaluate solutions focused on measuring and improving quality and safety of health care. His informatics career started at Columbia University where he focused on developing informatics methods for quantifying and preventing medication overuse. During this period, he worked with Dr. Carol Friedman, an internationally renowned researcher in the area of biomedical natural language processing (NLP). He has since expanded his focus to other areas of medication and prescription safety, and patient safety as a whole. He currently works at NewYork-Presbyterian Hospital’s Value Institute as program director for research science, and as a lecturer in biomedical informatics at Columbia University.
Dr. Rimma Pivovarov received her PhD training at Columbia University under Dr. Noemie Elhadad’s supervision. While at Columbia, she worked with both narrative and structured data, developing a probabilistic EHR phenotyping model. Since her graduation, she has moved into the realm of designing and evaluating hospital interventions, working with Dr. Salmasian in the Value Institute at NewYork-Presbyterian Hospital.
The presenters have been quoted, appeared in media outlets, and/or published in:
The presenters have published in Journal of American Medical Informatics Association, Journal of Biomedical Informatics, JAMA Internal Medicine, Annals of Emergency Medicine, Pharmacoepidemiology and Drug Safety and made several presentations at the AMIA Annual Symposium.