Rapid-response (RR events) teams can improve clinical outcomes but their triggering criteria are sometime subjective, hindering their identification in free-text notes. We used unsupervised method, latent Dirichlet allocation, to identify clinically interpretable topics from nursing notes and evaluated their specificity to RR events compared to healthier patients, demonstrating the feasibility of unsupervised and low-effort method to analyze free-text nursing notes.

Describe the new knowledge and additional skills the participant will gain after attending your presentation.: Participants will gain new insights about the use of automatic (unsupervised) machine-learning methods to explore the medical recod of specific patient populations based on clinical condition. The presentation provides a demonstration of the process, the results, the gains and challenges of such methods. These insights may help the participants design and conduct EHR-based clinical studies (e.g. descriptive and predictive analyses) at lower domian-expert effort.


Zfania Tom Korach (Presenter)
Brigham and Women's Hospital

Kenrick Cato, Columbia University
Sarah Collins, Columbia University
Min Jeoung Kang, Brigham and Women's Hospital
Christopher Knaplund, Columbia University
Patricia Dykes, Brigham and Women's Hospital
Liqin Wang, Brigham and Women's Hospital
Kumiko Schnock, Brigham and Women's Hospital
Jose Garcia, Brigham and Women's Hospital
Haomiao Jia, Columbia University
Frank Chang, Brigham and Women's Hospital
Jessica Schwartz, Columbia University
Li Zhou, Brigham and Women's Hospital

Presentation Materials: