Medication-rights detection using incident reports: A natural language processing and deep neural network approach

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Patient Safety

Background:

• Medication errors frequently arise from violations of the 'rights' of medication administration: right patient, drug, time, dose, and route.
• Automating the identification of these errors from free-text incident reports is crucial for enhancing patient safety.

Purpose of the Study:

• To develop a medication-rights detection system leveraging natural language processing (NLP) and deep neural networks (DNNs).
• To automate the identification of medication incidents from free-text reports within the Advanced Incident Reporting System (AIRS).

Main Methods:

• Utilized NLP and DNNs to classify medication incidents based on free-text reports.
• Compared DNN model performance against traditional classifiers like logistic regression, support vector machines, and decision trees.
• Evaluated the impact of various DNN hyperparameters (layers, neurons, regularization) on prediction accuracy.

Main Results:

• DNN models achieved high accuracy (≥0.9) across different settings and algorithms.
• Average accuracy and area under the curve (AUC) were 0.940 (SD: 0.011) and 0.911 (SD: 0.019), respectively.
• DNNs outperformed other classifiers in identifying all tested medication rights violations (wrong patient, drug, time, dose, route).

Conclusions:

• The developed medication-rights detection system effectively uses NLP and deep learning for classifying patient safety incidents.
• This approach shows promise for transferability to other incident reporting systems globally.
• Automated detection of medication errors can significantly enhance patient safety protocols.