Skip to primary content

Detection and presentation of bleeding events in the electronic health record

Bleeding and venous thromboembolism are frequent complications with hospitalised patients. Clinicians should be able to easily identify high risk cases when assessing risk of bleeding prior to surgery, preventive treatment for blood clots etc.

Project period

Start: April 2021
End: March 2022

Aim

The goal of this project was to develop AI algorithms based on Natural Language Processing to find bleeding episodes reported in the patients' electronic health records. It was developed based on extraction of relevant text from more than 300,000 electronic health records (unstructured data). When implemented, the algorithm could extract information on bleeding episodes from the patient's medical record and the results would be reported to the medical doctor and patient in an easily understandable and relevant way.

RESULTS

The first part of the project proved that an artificial neural network can detect 98% of the bleeding episodes, and the second part of the project aimed to develop a viable solution which could become part of daily clinical practice. The solution should be able to find indications of bleeding episodes, present a summary for the clinician, and suggest appropriate measures.

Participants

This project was a subproject of the project The Intelligent Health Record, which included several other subprojects on the use of AI to find information in the electronic health record.

Read more about the project The Intelligent Health Record at ipj.nu (website only available in Danish).

Pernille Just Vinholt

Pernille Just Vinholt

Physician, Associate Professor

Odense University Hospital, Department of Clinical Biochemistry


(+45) 2964 8694
Thiusius Rajeeth Savarimuthu

Thiusius Rajeeth Savarimuthu

Research Manager - Professor

Centre for Clinical Robotics (CCR). University of Southern Denmark, Maersk Mc-Kinney Moller Institute


(+45) 24 40 95 45
APPFWU02V