Automated categorisation of clinical incident reports using statistical text classification

Area of Science:

• Health Informatics
• Natural Language Processing
• Machine Learning

Background:

• Clinical incident reporting systems are crucial for patient safety.
• Manual categorization of incident reports is time-consuming and prone to errors.
• Automated methods are needed to efficiently process large volumes of reports.

Purpose of the Study:

• To assess the feasibility of using statistical text classification for automatic categorization of clinical incident reports.
• To evaluate the performance of Naïve Bayes and Support Vector Machine algorithms in identifying specific incident types.

Main Methods:

• Trained and tested Naïve Bayes and Support Vector Machine (SVM) classifiers on hospital incident reports.
• Classifiers were designed to identify two key incident types: inadequate clinical handover and incorrect patient identification.
• Evaluated classifier performance using accuracy, precision, recall, F-measure, and Area Under the Curve (AUC).

Main Results:

• Both Naïve Bayes and SVM demonstrated strong performance in categorizing clinical incidents.
• Naïve Bayes excelled in identifying handover incidents (AUC=0.97 for expert-classified reports).
• SVM achieved high accuracy for patient identification incidents (AUC=1.00 for reporter-classified reports), with small training sets proving adequate.

Conclusions:

• Statistical text classification techniques are feasible for the automated categorization of clinical incident reports.
• Automated systems can significantly improve the efficiency and accuracy of incident report analysis.
• This approach supports enhanced patient safety by enabling faster identification of critical incident types.