Assertion Detection in Clinical Natural Language Processing using Large Language Models.

Area of Science:

• Clinical Natural Language Processing (NLP)
• Artificial Intelligence in Healthcare
• Medical Informatics

Background:

• Assertion detection is crucial for understanding medical concepts in clinical notes, influencing patient care.
• Traditional methods for assertion detection are labor-intensive and may miss nuances.
• Accurate assertion detection requires identifying certainty, temporality, and experiencer for medical concepts.

Purpose of the Study:

• To develop and evaluate a novel methodology for assertion detection using Large Language Models (LLMs).
• To enhance LLM-based assertion detection with advanced reasoning techniques and fine-tuning.
• To improve the accuracy and generalizability of assertion detection in clinical NLP tasks.

Main Methods:

• Utilized pre-trained Large Language Models (LLMs) on extensive medical data.
• Integrated advanced reasoning techniques: Tree of Thought (ToT), Chain of Thought (CoT), and Self-Consistency (SC).
• Applied Low-Rank Adaptation (LoRA) for model fine-tuning.

Main Results:

• Achieved a micro-averaged F-1 score of 0.89 on the i2b2 2010 assertion dataset, an 0.11 improvement.
• Demonstrated generalizability with an F-1 score of 0.74 on a local sleep concept dataset, a 0.31 improvement.
• LLMs show significant potential for advancing assertion detection in clinical NLP.

Conclusions:

• Large Language Models offer a viable and effective approach for assertion detection in clinical NLP.
• The proposed LLM-based methodology significantly outperforms previous methods in accuracy and generalizability.
• This approach can be integrated with other LLM-based models for comprehensive clinical NLP solutions.