After three days at the World Drug Safety Congress 2025, I left with many thoughts about where AI is heading in pharmacovigilance (PV). Here are some of my key takeaways and beliefs about what’s next:
1. Broader adoption of AI: More companies are now applying AI to streamline operations, automate workflows, and generate reports efficiently.
2. Prompt-based LLMs on the rise: We’ll likely see more prompt-based models used for document triage and even AE extractions.
3. Benchmark datasets are coming: Once standardized benchmark data become available, we’ll finally be able to compare model performance objectively. This will accelerate real progress in the field.
4. Smarter signal detection: With richer, more integrated data from multiple sources, AI and ML methods for signal detection will become more sophisticated and reliable.
5. Personalized safety prediction: As data diversity grows, personalized AE detection and prediction will become a focus area in the next few years.
6. Genomics-driven insights: Genomic data could help us infer individual adverse events, enabling more personalized risk understanding.
7. Smaller, domain-specific models: Expect to see more distilled, task-specific models developed to reduce the high costs of commercial LLMs. Benchmark datasets will again play a key role here.
8. Quantifying real-world efficiency/cost gains: We’ll start to see clearer quantification of AI’s impact on efficiency and cost reduction, factoring in the often-overlooked costs of deployment, integration, and infrastructure.
AI is transforming PV faster than ever. The key challenge is ensuring that innovation goes hand-in-hand with validation, transparency, and usability.
What do you think will drive the next major breakthrough in AI-powered PV?
#Pharmacovigilance #DrugSafety #ArtificialIntelligence #MachineLearning #WDSC2025