Session Outline
Since 2021, the Swedish Medical Products Agency has been building AI competence and capacity to meet regulatory needs and enable in-house intelligent automation. To support the harmonization of medicinal product information, we have used NLP models for sentence-level semantic clustering of the complete corpus of product information for centrally approved drugs in the EU. To facilitate the assessment of adverse event reports (AER) related to medicinal products, which have increased greatly in number during the covid pandemic, we have developed PhaVAI – an ML-generated NLP model ensemble for AER severity classification.
A core data science unit including a PhD candidate in applied AI is now a part of the Agency’s strategic planning for the future.
Key Takeaways:
- Low-hanging ML fruit is abundant when introducing AI to new domains such as pharmaceutical regulatory science.
- Detailed domain knowledge integrated into the data science team is essential to ensure the correct acquisition, curation, and modelling of data.
- Deployment and life-cycle management in real-life regulatory settings provide challenges not easily simulated during modelling.
