Practical, pragmatic and predictable AI.
The European Medicines Agency (EMA) recently published a reflection paper on the use of artificial intelligence (AI) and machine learning (ML) in the medicinal product lifecycle, providing key guidance on how AI technologies can be integrated responsibly throughout drug development [1]. This reflection paper highlights AI’s potential to enhance drug discovery, clinical trials, regulatory submissions, and post-authorization monitoring, while also addressing critical technical, ethical, and regulatory considerations.
Some of the key recommendations include:
1. Data Quality and Bias Mitigation: The EMA emphasizes the importance of addressing potential biases in data used for AI/ML models. This involves documenting data sources and processing activities in a traceable way, ensuring compliance with Good Practices (GxP) requirements, and taking steps to mitigate biases in the training and testing phases of AI models.
2. Model Development and Validation: AI/ML models must be robust, generalisable, and subject to rigorous validation processes. Performance assessments and the metrics used to evaluate models are central to ensuring AI systems are reliable, especially when they contribute to clinical decision-making or marketing authorisation applications.
3. Explainability and Transparency: The paper acknowledges that while “black box” AI models (those whose internal workings are not easily interpretable) may be acceptable in certain scenarios, developers must justify their use. In general, the EMA prefers transparent models and encourages using explainable AI techniques where possible.
4. Regulatory Engagement: Developers are advised to engage with regulatory authorities early in the process when AI/ML models may impact the benefit-risk balance of a medicine. This is particularly important for applications where AI plays a significant role in data analysis or decision-making.
5. Ethical Considerations and Data Protection: The EMA underscores the importance of aligning AI/ML systems with ethical principles and data protection laws. Developers must ensure that AI applications respect human rights and fundamental freedoms, following established guidelines for trustworthy AI.
The paper reflects what the EMA’ proposes is a proactive approach to managing the growing role of AI in medicine, highlighting the need for rigorous oversight to ensure patient safety, regulatory compliance, and ethical standards. However, for now it is little more than a positioning piece. The agency plans to update the reflection paper based on feedback from stakeholders, ensuring it remains relevant as AI technologies evolve. If you would like a more proactive approach you might like to read our recent Insider's Insight into how to manage the runaway use of large language model software, like Chat GPT, that is already going on in your company - like it or not [2].
Recommended by LinkedIn
These insights can guide pharmaceutical companies and regulators as they navigate the complex integration of AI into medical research and drug development, ensuring that the technologies are used responsibly and safely across all phases of a product’s lifecycle. However, handle with care or run the risk of opening Pandora's box [3]!
References
AI & Data Solutions Expert | Driving Business Growth with Custom AI Models, Data Analysis, Workflow Automation & Intelligent Chatbots | Founder of AI Data House
3moGreat step by EMA we have used AI to streamline healthcare claims processing which has improved accuracy and speed. AI's potential in drug discovery and trials is exciting. How do you see AI shaping post-approval monitoring while keeping regulations in check?
Insightful and informative
Useful info from the EMA
Cheers
Thanks