Artificial Intelligence for Drug Safety Across the Lifecycle and Decision Type: A Scoping Review

Area of Science:

• Pharmacovigilance and drug safety research
• Application of artificial intelligence in medicine
• Health informatics and data science

Background:

• Artificial intelligence (AI) is increasingly used in drug safety evaluation, but its applications are fragmented across different stages and tasks.
• Existing research lacks a comprehensive overview of AI's role in drug safety decision-making throughout the drug lifecycle.
• Understanding the evaluation strategies for AI models is crucial for their reliable clinical and regulatory use.

Purpose of the Study:

• To map the application of AI in supporting safety and treatment decisions across the entire drug lifecycle.
• To examine the evaluation strategies employed to assess the reliability of AI models for clinical and regulatory purposes.
• To identify patterns and gaps in AI utilization for drug safety and treatment optimization.

Main Methods:

• A scoping review was conducted using the Arksey and O'Malley framework.
• Searched a major database for studies from the past decade on AI/machine learning in drug safety or medication decisions.
• Extracted data on lifecycle stage, decision type, AI methods, data sources, and evaluation strategies, constructing a lifecycle-decision matrix.

Main Results:

• AI applications are concentrated in patient-level safety prediction (clinical care) and safety surveillance (post-marketing), utilizing EHRs, spontaneous reporting systems, and clinical text.
• Common AI methods include gradient boosting, deep neural networks, graph neural networks, and natural language processing.
• Limited evidence exists for AI in treatment optimization and regulatory decision modeling; most studies rely on internal validation, with external validation and real-world deployment being uncommon.

Conclusions:

• AI holds significant potential for enhancing drug safety, especially in risk prediction and pharmacovigilance.
• Uneven application across the drug lifecycle and limited external validation/real-world testing hinder widespread clinical and regulatory adoption.
• Further advances in AI for drug safety necessitate robust external validation and real-world testing to bridge the translational gap.