San Francisco, CA
May 15-16, 2017

Mining Physician Documentation to Identify and Treat at-Risk Patients

May 16, 2017
2:40pm - 3:15pm
Grand Ballroom

Identifying patients with a stroke as early as possible can help health systems initiate effective treatments in a timely fashion. Unfortunately, patients with ambiguous presentations can delay a stroke diagnosis and subsequent treatment. Some healthcare systems manually review patient records to determine if a patient had a stroke that was missed by the primary team. Electronic medical records allow for computer review of hundreds of charts with Natural Language Processing (NLP) to identify those patients whose presentation merit further evaluation for stroke. We describe one healthcare system’s approach to use NLP to reduce the manual review required to identify patients who may have suffered a stroke undetected by the primary team.

The process occurs in near-real time as notes are reviewed on a daily basis. The stroke detection algorithm is updated regularly as the human reviewers provide feedback around identified cases.

The project highlights NLP’s ability to provide near real-time analytics and process efficiencies for this specific use case with potential applications for other disease states. 

Key takeaways: 

  • NLP can help identify specific clinical conditions before formal diagnosis.
  • NLP is vulnerable to bias based on what clinicians enter in their documentation.
  • NLP can reduce human review of clinical documentation to achieve specific clinical outcomes.
Off White X Max 180-90


Vice President, Clinical Information
Advocate Health Care

Get Updates

Sign up to get the latest information on upcoming events.