An intelligent framework for dynamic modeling of therapeutic response using clinical compliance data.

Introduction: The increasing availability of real-world clinical compliance data provides unprecedented opportunities to model medication behaviors dynamically and personalize treatment strategies. However, the complex, heterogeneous, and often incomplete nature of these data presents significant modeling challenges, particularly for capturing medication nonadherence, patient-specific therapeutic dynamics, and drug interaction effects. Existing approaches, including statistical regression models and rule-based decision systems, often fail to capture the high-dimensional, temporally-evolving, and probabilistic characteristics inherent in medication trajectories, limiting their effectiveness in precision medicine and policy simulation contexts. Methods: To address these limitations, we propose a novel intelligent computing framework that unifies probabilistic graphical modeling, deep temporal inference, and domain-informed strategy design. Our approach is instantiated in the Hierarchical Therapeutic Transformer (HTT), a Bayesian transformer-based model that captures therapeutic state transitions via structured latent variables and medication-aware attention mechanisms. Furthermore, we introduce the Pharmacovigilant Inductive Strategy (PIS), a training paradigm that integrates pharmacological priors, adaptive quantification, and entropy-driven curriculum learning to enhance robustness and generalizability. Our method effectively models dose-response variability, accounts for clinical data missingness, and generalizes across cohorts through a hierarchical latent prior framework. Results and discussion: Experimental evaluations demonstrate that our system achieves state-of-the-art performance in predicting adherence patterns and clinical outcomes across diverse datasets, aligning with current advances in medication adherence modeling and probabilistic health informatics. This work provides a rigorous, interpretable, and scalable foundation for real-time decision support in pharmacotherapy, contributing to the broader goals of personalized medicine, drug safety monitoring, and computational clinical reasoning.