Automatic Patient Fall Outcome Extraction Using Narrative Incident Reports.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Healthcare Data Analysis

Background:

• Patient outcomes are crucial in incident reporting for analyzing in-hospital falls.
• Automating the extraction of patient fall outcomes aids in understanding consequences and developing mitigation strategies.
• Current methods for extracting patient outcomes from narrative reports are limited.

Purpose of the Study:

• To automate the extraction of patient outcomes from narrative fall incident reports.
• To decompose the extraction process into three classification subtasks: injured or not, injury types, and number of injuries.
• To validate an incident report classification (IRC) framework for imbalanced classification problems in patient outcome extraction.

Main Methods:

• Implementation of an existing incident report classification (IRC) framework.
• Decomposition of patient outcome extraction into three distinct classification subtasks.
• Application of oversampling techniques and structured features for improved classification performance.

Main Results:

• Demonstrated effectiveness of oversampling and structured features across all three classification subtasks.
• Achieved better overall performance in classifying patient injury status, types, and counts.
• Validated the IRC framework's capability to handle imbalanced datasets in patient outcome classification.

Conclusions:

• The study successfully automated patient outcome extraction from fall incident reports.
• The IRC framework is effective for imbalanced classification tasks in healthcare.
• This advancement contributes to the science of automatic patient outcome extraction, enhancing patient safety analysis.