Applications of Federated Large Language Model for Adverse Drug Reactions Prediction: Scoping Review.

Area of Science:

• Health Informatics
• Artificial Intelligence in Medicine
• Computational Linguistics

Background:

• Adverse drug reactions (ADRs) pose significant healthcare challenges, necessitating early detection for effective treatment and patient safety.
• Traditional supervised learning methods struggle with heterogeneous, unstructured healthcare data due to privacy and access restrictions.

Purpose of the Study:

• To explore the potential of combining federated learning (FL) with natural language processing (NLP) and large language models (LLMs) for enhanced ADR prediction.
• To review current advancements and identify use cases for FL and LLMs in ADR prediction.

Main Methods:

• A scoping review of peer-reviewed publications from 2019-2024 was conducted using Google Scholar and Semantic Scholar.
• The review followed the PRISMA protocol, analyzing 145 articles and selecting 12 for in-depth examination of FL and LLM applications in ADR prediction.

Main Results:

• Synthesized data sources for ADR prediction, including structured and unstructured types.
• Examined use cases of FL integrated with NLP for ADR identification and prediction.
• Focused on unstructured ADR prediction using federated learning with large language models, covering development, deployment, and evaluation.

Conclusions:

• The integration of FL and LLMs for ADR prediction is nascent, with limited documented use cases.
• Highlighted advancements in privacy-preserving, scalable ADR prediction systems using FL and LLMs.
• Identified key areas for future research, including algorithm refinement, fairness, and real-world deployment strategies.