From Traditional Statistics to Adaptive Multivariate Models: Exploring the Role of AI in Modern Pharmaceutical...

Modern pharmaceutical manufacturing increasingly relies on high-frequency, high-dimensional and dynamically evolving data. Traditional statistical tools-such as univariate Statistical Process Control (SPC), Shewhart logic and Nelson's rules-remain valuable for routine monitoring, but are constrained by assumptions of independence, stationarity and linearity. These assumptions are frequently violated in contemporary bioprocesses, leading to inconsistent diagnostics and elevated false-alarm rates. Such limitations have been documented in data-rich manufacturing environments and confirmed in continued process verification initiatives, where repeated univariate rule violations were shown to lack biological relevance.In parallel, regulatory frameworks-including FDA Process Analytical Technology (PAT), ICH Quality by Design (QbD) and the draft EudraLex Annex 22 on Artificial Intelligence (AI)-increasingly emphasize scientifically justified, model-based approaches. This manuscript examines the boundaries of classical statistical methods, the inherently multivariate and dynamic nature of modern pharmaceutical processes, and the role of fixed deterministic and fixed probabilistic AI-based models in addressing analytical gaps. A structured scientific exploration framework is proposed to support evidence-based regulatory acceptance of advanced modelling approaches while improving process understanding, product quality and manufacturing efficiency.