Large Language Models for Healthcare Data Augmentation: An Example on Patient-Trial Matching

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Trial Management

Background:

• Patient-trial matching is crucial for medical research and patient care.
• Current methods struggle with data standardization, ethical issues, and EHR interoperability.
• Large language models (LLMs) offer potential for improved patient-trial matching.

Purpose of the Study:

• To explore LLMs for enhancing patient-trial matching.
• To propose a privacy-aware data augmentation method for LLM-based patient-trial matching (LLM-PTM).
• To address challenges in EHR and clinical trial criteria compatibility.

Main Methods:

• Utilizing LLMs' natural language generation for EHR-clinical trial compatibility.
• Implementing a privacy-aware data augmentation strategy for LLM-PTM.
• Conducting experiments to evaluate the LLM-PTM method's performance and generalizability.

Main Results:

• Achieved an average performance improvement of 7.32% with the LLM-PTM method.
• Enhanced generalizability to new data by 12.12%.
• Demonstrated effectiveness through case studies.

Conclusions:

• LLM-PTM offers a promising solution for efficient and secure patient-trial matching.
• The privacy-aware approach effectively balances LLM benefits with data confidentiality.
• The method shows significant improvements in performance and generalizability for clinical trial recruitment.