Send the link below via email or IMCopy
Present to your audienceStart remote presentation
- Invited audience members will follow you as you navigate and present
- People invited to a presentation do not need a Prezi account
- This link expires 10 minutes after you close the presentation
- A maximum of 30 users can follow your presentation
- Learn more about this feature in our knowledge base article
Event detection using natural language processing
Transcript of Event detection using natural language processing
Self-reporting by physicians very low
Need to find real rate of adverse occurrences Background Lindsay Lee
Vanderbilt University Medical Center
Department of Anesthesia What is the answer? Natural Language Processing! Using computer systems to analyze human-generated text
Levels of language processing:
Ambiguity present in each level NLP "I saw the man with the telescope."
"The city councilors refused the demonstrators a permit because they feared violence" Ambiguity How can NLP be used to analyze free-text case comments in order to look for adverse events? My project Python Natural Language Toolkit (NLTK)
Checks for words in comments Components of program Import comments from spreadsheet Make everything lower case
and split the string into list For each comment,
is a keyword in the comment? No? Then move on
to next comment Yes? Is there a difficulty indicator
in a threshold around the keyword? Is that difficulty indicator
not negated and referring
to the keyword? Then that's a comment with
difficult intubation! Outputs list of anesthesia case numbers
Runs as hoped up to 1800 case comments
Comment output matches manual data review Results This Python program can pick up on difficult intubations that manual data review misses.
Difficult intubations are under-reported
May be uncertainty of the definition of "difficult" Conclusion Make program more efficient
Increase breadth of applicability of program
encompass all expressions of difficult intubation
allow conversion to search for other adverse events Future plans Dr. Paul St. Jacques and Dr. Jesse Ehrenfeld
Department of Anesthesiology
Perioperative Data Systems Research group
Marnie McNamara and Vanderbilt Summer Science Academy: Short Term Training Program
Department of Biomedical Informatics Acknowledgments Sample set of approximately 5500 case comments
Two weeks worth of data
One anesthesia case number can correspond to multiple case comments Data Search in Excel for words "intubat," "ett", "DL"
Of those listed, mark if difficult or not
difficult: more than one attempt needed or explicitly stated difficulty
not difficult: explicitly stated ease or no mention of difficulty or multiple attempts
In order to have information to compare program against Manual review of data Allen, James F. "Natural Language Processing."
Liddy, E. D. "Natural Language Processing." Encyclopedia of Library and Information Sciences, 2nd Ed. Marcel Decker, Inc.
Murff HJ, Patel VL, Hripcsak G, Bates DW. Detecting adverse events for patient safety research: a review of current methodologies. Journal of biomedical informatics. 2003;36(1-2):131-43. Epub 2003/10/14.
Weinger MB, Slagle J. Human factors research in anesthesia patient safety. Proceedings / AMIA Annual Symposium AMIA Symposium. 2001:756-60. Epub 2002/02/05. References Our adverse event: difficult intubation Of 18 difficult intubations:
16.7% checked "difficult" (3 cases)
44.4% checked "easy" (8 cases)
38.9% not checked either way (7 cases)
88.9% cite multiple attempts (16 cases) Results (cont.)