ЗАМКНЕНИЙ ЦИКЛ УПРАВЛІННЯ ЯКІСТЮ СИСТЕМИ ПІДТРИМКИ ПРИЙНЯТТЯ КЛІНІЧНИХ РІШЕНЬ: ФОРМАЛІЗАЦІЯ В ARCHIMATE ТА ОЦІНЮВАННЯ ПЕРЕХОДУ ЗА МОДЕЛЛЮ ЗРІЛОСТІ

The problem of closing the quality management loop for machine-learning-based clinical decision support systems within a medical center enterprise architecture is considered. Deployed clinical models lose accuracy because of data shift and concept drift, while machine learning operations remain poorly studied specifically in health care settings. It is shown that existing research treats the technical composition of machine learning pipelines and the medical enterprise architecture separately, so a cross-cutting description of how to close the clinical recommendation quality control loop, aligning motivation, roles, processes, applications, and data, is lacking. The current state of a representative architecture formalized with an enterprise architecture modeling tool is analyzed, and two gaps are identified: the absence of systematic auditing of recommendation effectiveness and the absence of a model update loop. A motivational model turning the gaps into requirements is built, and a target architecture whose components are derived from these requirements is designed; it adds an analytical circuit, a model update platform, and a model gateway as the integration point with the clinical system. The physician feedback loop already exists in the current state, so the loop is closed without rebuilding the clinical circuit. The transition is evaluated analytically, without a prototype implementation. On an external maturity scale the current state is low and the target state partial, since retraining is triggered manually. A traceability matrix confirms the coverage of requirements by their implementers and reveals elements added outside the justification chain, which a direct coverage check does not register. The limitations of the model-based evaluation are analyzed: the current state is a representative architecture rather than a surveyed installation, so a pilot instantiation is required.