Extracting medication changes in clinical narratives using pre-trained language models.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Clinical Data Mining

Background:

• Accurate patient medication records, including changes, are vital for effective healthcare delivery.
• Medication changes offer insights into patient health status and treatment rationale.
• Automating the extraction of medication change details from clinical notes is a significant challenge.

Purpose of the Study:

• To develop and evaluate high-performing systems for automatically extracting medication change information from free-text clinical notes.
• To leverage the Contextual Medication Event Dataset (CMED) for training and testing these systems.
• To improve the classification performance of medication change attributes.

Main Methods:

• Utilized the Contextual Medication Event Dataset (CMED), a corpus of annotated clinical notes.
• Developed three novel BERT-based systems for identifying medication mentions and their associated change characteristics.
• Evaluated system performance on attributes such as change type, initiator, temporality, likelihood, and negation.

Main Results:

• The proposed BERT-based systems demonstrated improved performance in classifying medication change characteristics compared to previous methods.
• Successfully identified medication mentions and resolved complex change-related attributes within clinical text.
• Achieved high performance in extracting detailed medication change information.

Conclusions:

• Automatic extraction of medication change information from clinical notes is feasible and can be significantly enhanced with advanced NLP models.
• The developed BERT-based systems offer a promising approach for improving the accuracy and detail of patient medication histories.
• This work contributes to better clinical decision-making by providing more reliable medication change data.