Deep learning detects and visualizes bleeding events in electronic health records

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Data Analysis

Background:

• Bleeding events significantly increase patient morbidity and mortality.
• Identifying bleeding in electronic health records is challenging due to unstructured text.
• Manual review of clinical notes for bleeding is time-consuming and prone to error.

Purpose of the Study:

• To develop a deep learning model for detecting bleeding events in electronic health records.
• To enable visualization of bleeding event information within clinical notes.
• To improve the systematic assessment of bleeding risk.

Main Methods:

• Extracted 300 electronic health records with bleeding or leukemia diagnosis codes.
• Annotated sentences within records as positive or negative for bleeding.
• Developed and evaluated a deep learning model for sentence and note-level bleeding detection.

Main Results:

• The deep learning model achieved high performance metrics (sensitivity, specificity, NPV) on sentence and note-level test sets.
• The model demonstrated the ability to visualize specific sentences indicating bleeding events.
• Consistent performance was observed across various types of bleeding events.

Conclusions:

• Deep learning models can effectively detect and visualize bleeding events in unstructured EHR text.
• This technology can enhance the systematic assessment of bleeding risk.
• Improved bleeding risk assessment can lead to optimized patient care and safety.