Automatically extracting information needs from complex clinical questions.

Area of Science:

• Natural Language Processing
• Medical Informatics
• Artificial Intelligence in Healthcare

Background:

• Clinicians frequently encounter complex clinical questions requiring timely answers to enhance patient care quality.
• Existing methods for extracting information needs from clinical questions are often manual and time-consuming.
• The AskHERMES system aims to provide a comprehensive solution for answering clinical questions.

Purpose of the Study:

• To develop and evaluate natural language processing (NLP) models for automatic topic assignment and keyword identification in clinical questions.
• To effectively extract information needs from ad hoc clinical questions to support the AskHERMES question-answering system.
• To improve the efficiency and accuracy of information retrieval for clinical decision-making.

Main Methods:

• Developed supervised machine-learning systems for automatic assignment of predefined general categories (e.g., etiology, diagnosis) to clinical questions.
• Explored supervised and unsupervised systems for automatic identification of keywords that represent the core content of clinical questions.
• Evaluated system performance on a dataset of 4654 annotated clinical questions.

Main Results:

• Achieved an F1 score of 76.0% for general topic classification.
• Achieved an F1 score of 58.0% for keyword extraction.
• The developed systems were successfully integrated into the AskHERMES question-answering system.

Conclusions:

• The developed NLP systems can automatically extract information needs from diverse clinical questions, regardless of length or structure.
• Performance in topic classification and keyword extraction can be further enhanced with more consistently annotated training data.
• These automated tools hold potential for improving clinical decision support and patient care.