A large dataset of annotated incident reports on medication errors

Area of Science:

• Medical Informatics
• Natural Language Processing
• Patient Safety

Background:

• Medication errors are critical for patient safety, but unstructured text in incident reports hinders analysis.
• Automated analysis requires large, annotated datasets for natural language processing (NLP) model development.

Purpose of the Study:

• To present a large, machine-annotated corpus of medication error incident reports.
• To facilitate the development of NLP models for automated information extraction and incident learning.

Main Methods:

• Developed a machine annotation process for 58,658 medication error reports.
• Performed cross-validation for named entity recognition (NER) and intention/factuality analysis.
• Detailed annotation workflow and technical validation procedures.

Main Results:

• Achieved F1-scores of 0.97 for NER and 0.76 for intention/factuality analysis.
• Annotated dataset includes 478,175 named entities.
• Successfully differentiated incident types by recognizing intended vs. actual occurrences.

Conclusions:

• The presented corpus enables advanced information extraction for medication error analysis.
• The dataset and machine annotator support future development in patient safety learning.
• Automated analysis of medication error reports can be significantly improved with this resource.