Population Pharmacokinetics and Pharmacodynamics under Treatment Evolution, Time-Varying Covariates, Informative Sampling, and Changing Disease State during Clinical Care

Population pharmacokinetic and pharmacodynamic models commonly support inference by relating dose, exposure, response, and patient characteristics, yet their application during clinical care can become problematic when treatment, physiology, disease severity, observation intensity, and clinical decisions change together. This article develops an original longitudinal modelling framework that treats clinical PK/PD as a coupled system of biological states, treatment events, observation processes, and model-mediated decisions. The framework distinguishes latent pharmacokinetic and pharmacodynamic states from measured covariates, recorded doses, laboratory observations, and clinician actions. Treatment is represented as a timestamped history rather than a fixed regimen, while physiological variables and disease markers are treated according to their temporal availability and their possible roles as predictors, mediators, treatment consequences, or imperfect measurements of latent disease. Missing data are positioned within the observation process because absence may arise from visit timing, test-ordering decisions, documentation failure, assay limitations, or selective clinical attention. Dynamic prediction is defined as sequential updating from accumulating patient information, whereas population-model relearning is treated as a separate, governed activity. Evaluation therefore requires more than conventional goodness-of-fit assessment: calibration, uncertainty coverage, temporal validation, external transportability, observation-process sensitivity, update stability, and prospective workflow evaluation must be considered. The original contribution is an integrated state–event–observation architecture connecting treatment evolution, changing disease, time-varying covariates, informative sampling, and model updating without assuming that any component is sufficient alone. The framework remains conceptual and requires drug-specific, population-specific, computational, prospective, and implementation validation before decision use.