Assessment and Compliance of Personalized Machine-Learning Pharmacokinetic Models in the European Regulatory Environment

The transformative potential of Deep Learning for the field of pharmacometrics is being extensively investigated by the scientific community. Latest models provide the innovative capability of differentiating the predictions at the level of individual patients, greatly fostering the personalization of the therapy. Still, due to the high-risk context of their application, there’s the possibility that such technologies will not be employed soon in clinical settings at large scale. The European Union, through regulations such as the Artificial Intelligence Act and Medical Device Regulation, subjects the adoption of AI-based systems to careful risk assessment procedures, quality management, post market monitoring and, possibly, explicit safety thresholds. The safety of the individual patient cannot be exclusively derived from suitable and performant models; instead, it emerges from the interaction between the developer, the deployer and the user of the application. Accordingly, we present a harmonized methodology that explicitly links technical design decisions to sound legal reasoning. Developed through the interdisciplinary collaboration advocated in the literature and built on the tradition of jurimetrics, our proposed standard combines empirical performance metrics and theoretical engineering frameworks with formally verifiable compliance criteria. By basing technical requirements on measurable legal approximations, jurimetrics represents an epistemic bridge between pharmacometrics and regulatory oversight, guiding developers towards legal-by-design strategies that allow the seamless transition of technology from the research lab to routine care, and provides auditable evidence that protects stakeholders from liability, creating a common benchmark for the safe, compliant and equitable adoption of AI-driven pharmacometrics systems in healthcare.