Development of an Automated Classification System for Medication-Related Incident Factors: A Practical Approach to...

Area of Science:

• Health Informatics
• Natural Language Processing
• Patient Safety

Background:

• Analyzing medication-related incident reports is vital for patient safety.
• Systematically extracting contributing factors from these reports is a significant challenge.

Purpose of the Study:

• To develop and evaluate a multi-label classifier for automatic identification of incident factors in drug-related reports.
• To assess the feasibility of NLP techniques for systematic incident factor analysis.

Main Methods:

• Utilized Bidirectional Encoder Representations from Transformers (BERT) and its derivatives for classification.
• Developed a multi-label classifier on 1,212 drug-related incident reports.
• Employed the P-mSHELL model framework and evaluated models using five-fold cross-validation.

Main Results:

• The multi-label classifier successfully identified incident factors across seven distinct categories.
• Most evaluated models achieved macro F1 scores above 0.6.
• A Lite BERT demonstrated performance comparable to BERT.

Conclusions:

• Natural language processing techniques are practically feasible for systematic incident factor analysis.
• This methodology can support enhanced patient safety management and reduce medication errors.