Collaborative and privacy-enhancing workflows on a clinical data warehouse: an example developing natural language...

Area of Science:

• Medical Informatics
• Computational Linguistics
• Artificial Intelligence in Healthcare

Background:

• Clinical notes contain valuable patient data.
• Accurate extraction of medical conditions is crucial for research and patient care.
• Existing methods for condition detection in clinical notes can be limited.

Purpose of the Study:

• To develop and validate a natural language processing (NLP) pipeline for detecting 18 conditions in French clinical notes.
• To include 16 comorbidities from the Charlson index.
• To explore a collaborative and privacy-enhancing workflow for AI model development.

Main Methods:

• Utilized a hybrid approach combining rule-based and machine learning algorithms for named entity recognition and entity qualification.
• Employed a large language model pre-trained on millions of clinical notes.
• Leveraged annotated data from oncology, cardiology, and rheumatology cohort studies.
• Designed a workflow to foster inter-study collaboration while ensuring data privacy.

Main Results:

• Achieved a macro-averaged F1-score of 95.7%, positive predictive value of 95.4%, sensitivity of 96.0%, and specificity of 99.2%.
• Demonstrated superior performance compared to alternative technologies and non-collaborative settings.
• Successfully shared validated models through a secured registry.

Conclusions:

• A collaborative approach among investigators using a common clinical data warehouse enables efficient and secure development, validation, and deployment of AI models.
• The developed NLP pipeline is efficient and robust for detecting conditions in clinical notes.
• This work highlights the potential of privacy-preserving AI collaboration in medical research.