A Hybrid Natural Language Processing Platform for Multi-Site RWD Studies.

Area of Science:

• Health Informatics
• Computational Linguistics
• Data Science

Background:

• Real-world data (RWD) from electronic medical records is crucial for healthcare research.
• Integrating unstructured free-text clinical data presents significant challenges.
• High computational costs and privacy concerns hinder natural language processing (NLP) implementation in multi-site RWD studies.

Purpose of the Study:

• To propose a hybrid platform for efficient and privacy-preserving information extraction from free-text clinical data.
• To facilitate effective data integration across multiple institutions in RWD studies.
• To address the computational and privacy challenges associated with NLP in healthcare research.

Main Methods:

• Development of a hybrid platform integrating centralized NLP processing with privacy protection.
• Comparative experiments evaluating the hybrid platform against a fully distributed method.
• Utilizing 500 sample reports for performance assessment.

Main Results:

• The central NLP processing server demonstrated significantly superior performance, processing reports in 0.12 seconds compared to 64.23 seconds for on-site servers.
• The central server showed minimal and consistent increases in processing time irrespective of report length, indicating efficiency and scalability.
• The hybrid platform effectively addressed privacy and data governance issues.

Conclusions:

• The developed hybrid platform enhances computational efficiency for NLP tasks in RWD studies.
• The platform successfully tackles privacy and data governance challenges in multi-site research.
• This approach facilitates effective information extraction from free-text clinical data across institutions.