The life science regulated industries, such as quality control, pharmacovigilance, production,laboratory diagnostics, and clinical services, have a significant presence of Artificial Intelligence(AI), and Machine Learning (ML). However, Computer System Validation (CSV) hastraditionally been applied to deterministic algorithms whose requirements are known and whoseoutput can be predicted. Therefore, CSV does not cover all the lifecycle risks associated with theuse of AI-based systems. Such risks include probabilistic behavior, model drift, retraining,changes in data distribution, and evolution of operational context. This paper presents a risk-based Computer Software Assurance (CSA) approach for validatingAI-based systems within GxP. This proposed framework combines a clear definition of the usecase, AI risk classification, data governance, evidence-based testing, model governance, and real-timemonitoring, and periodic reviews to form a single, lifecycle-structured method. Theframework also complies with the FDA's CSA principles, GAMP 5 risk-based thinking, 21 CFR Part11 expectations, and EU Annex 11 lifecycle controls. This paper provides a practical and applicable AI-GxP validation framework that relatesgeneral regulatory principles into actionable decisions for monitoring AI model drift, when torevalidate AI models, and managing audit-ready evidence. The intention behind this frameworkis to enable regulated companies to utilize AI more safely and efficiently.
Sivakumar Kalidoss (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: