Beyond black boxes: using explainable causal artificial intelligence to separate signal from noise in...

Area of Science:

• Pharmacovigilance
• Artificial Intelligence
• Machine Learning

Background:

• Machine learning (ML) models show promise in improving pharmacovigilance (PV) tasks like case triage and signal detection.
• Current ML models often function as "black boxes," limiting transparency and trust in their decision-making processes.
• Traditional ML may perpetuate biases present in spontaneous reporting systems, such as under-reporting and missing data.

Purpose of the Study:

• To explore the potential and challenges of integrating AI into PV.
• To advocate for a transition towards causally informed and interpretable AI models in PV.
• To highlight the need for AI that enhances, rather than replaces, expert judgment in drug safety.

Main Methods:

• Review of current AI applications in PV.
• Discussion of limitations of traditional ML, including lack of explainability and bias amplification.
• Exploration of explainable AI (XAI) and causal AI as potential solutions.

Main Results:

• AI can accelerate the identification of potential drug safety issues.
• Lack of transparency and potential for bias are significant challenges with current AI in PV.
• Explainable and causal AI methods offer more interpretable and reliable outputs but require further development and validation.

Conclusions:

• A shift towards causally informed, interpretable AI models is crucial for credible and ethical PV.
• Key priorities include integrating causal inference, developing benchmark datasets, aligning outputs with clinical logic, and establishing rigorous validation.
• The goal is to enhance expert judgment with transparent and reliable AI tools for improved drug safety surveillance.