A Machine Learning Approach to Reclassifying Miscellaneous Patient Safety Event Reports

Area of Science:

• Health Informatics
• Medical Error Analysis
• Natural Language Processing

Background:

• Medical errors are a significant cause of death in the U.S.
• Current patient safety reporting systems struggle with complex event classification taxonomies.
• Frontline reporters often misclassify reports as "miscellaneous," hindering analysis.

Purpose of the Study:

• To develop a machine learning model to reclassify "miscellaneous" medical error reports.
• To assist patient safety analysts in efficiently analyzing unstructured safety data.
• To integrate the model into a clinical workflow dashboard for practical application.

Main Methods:

• Developed an ensemble machine learning model using natural language processing.
• Applied the model to reclassify "miscellaneous" event reports.
• Integrated the model into an interactive clinical workflow dashboard and collected user feedback.

Main Results:

• The ensemble model and interactive visualization aided analysts in reviewing "miscellaneous" reports.
• Analysts demonstrated varying thresholds for model-driven reclassification.
• Role and experience influenced analysts' acceptance of model reclassifications.

Conclusions:

• Machine learning and visualization tools can enhance the analysis of unstructured patient safety data.
• Individual user differences must be considered when implementing automated classification systems.
• Further refinement of the model and user interface may improve adoption and consistency.