VDI/VDE-EE 3516 Blatt 7
Validation in GxP area - Usage of machine learning methods in pharmaceutical industry
At a glance
- German title
-
Validierung im GxP-Umfeld - Einsatz von Verfahren des maschinellen Lernens in der pharmazeutischen Industrie
- Publication date
- 2024-12
- Publisher
- Engl. VDI/VDE-Gesellschaft Mess- und Automatisierungstechnik
- Related manuals
- Number of pages
- 39
- Available in
- German
- Abstract
-
The primary requirement for the quality of pharmaceutical products is to ensure patient safety. To this end, authorities formulate regulations and laws that require processes for compliance with product quality requirements and continuous improvement of safety. In conventional computerised systems, procedures such as quality risk management and the validation of systems and processes have become established in order to achieve a high level of safety. As these processes have not yet explicitly considered the use of models, an interpretation and continuation of these processes is required for computer-aided systems with machine learning components.The transition from classic algorithms based on explicit programming of rules and controls to models characterised by training based on historical data poses challenges for the industry in terms of the trustworthiness and reliability of these computer-aided systems. In particular, the fact that models are subject to statistical uncertainties must be emphasised.This VDI Expert Recommendation defines a practicable procedure model for the identification, use and operation of ML-supported systems that promotes trustworthiness in these systems to the same extent as in conventional systems. Existing, typically generalised regulations in the field of pharmaceutical manufacturing are appropriately interpreted in terms of patient safety. This VDI Expert Recommendation thus promotes a common understanding and general acceptance among authorities, pharmaceutical companies and manufacturers of ML-supported systems and their safe use in suitable fields of application.