Large Language Models for Adverse Drug Events: A Clinical Perspective

Area of Science:

• Pharmacovigilance and Clinical Informatics
• Artificial Intelligence in Healthcare

Background:

• Adverse drug events (ADEs) pose significant risks to patient safety and healthcare outcomes.
• Manual detection of ADEs from clinical narratives is inefficient and costly.
• Large language models (LLMs) offer automated solutions for ADE extraction.

Purpose of the Study:

• To review the current applications of LLMs for ADE detection in clinical settings.
• To categorize LLM-based ADE detection studies.
• To identify challenges and future directions for LLM implementation in pharmacovigilance.

Main Methods:

• A narrative review of 39 articles was conducted.
• Studies were categorized based on LLM applications in ADE detection.
• Focus on clinical decision support, immune-related ADEs, cancer-related ADEs, and personalized systems.

Main Results:

• LLM-driven methods demonstrate strong performance in ADE detection, often surpassing traditional approaches.
• Applications span decision support tools, surveillance, and personalized medicine.
• Identified limitations include domain variability, interpretability, data concerns, and infrastructure needs.

Conclusions:

• LLMs hold significant potential to enhance pharmacovigilance and patient safety.
• Addressing identified limitations is crucial for effective clinical workflow integration.
• Future research should focus on improving model robustness, interpretability, and data security.