Clinical Pharmacokinetic and Pharmacodynamic Modelling for Precision Therapeutics: A Methodological Review of Identifiability, Transportability, and Decision Use

Clinical pharmacokinetic and pharmacodynamic modelling can support precision therapeutics by connecting dose, exposure, response, patient characteristics, and uncertainty. However, a model that describes observed data adequately may still provide unreliable individual predictions or unsuitable dose recommendations when its parameters are non-identifiable, its evaluation is restricted to development data, or its assumptions do not transfer across populations and clinical settings. This methodological review examines the principal model classes used in clinical PK/PD research and evaluates how their assumptions, data requirements, identifiability, estimation procedures, validation strategies, calibration, transportability, and decision consequences affect their suitability for precision therapeutics. The proposed taxonomy distinguishes population and exposure–response models, semi-mechanistic and mechanistic models, physiologically based pharmacokinetic and systems models, Bayesian individual-prediction methods, and emerging data-adaptive or hybrid approaches. These classes are compared according to the questions they can answer rather than their mathematical complexity. Structural identifiability concerns whether model parameters are theoretically recoverable under defined observations, whereas practical identifiability concerns whether available clinical data support sufficiently precise and stable estimation. Internal diagnostics are distinguished from independent external validation, uncertainty calibration, and prospective evaluation of dose decisions. Transportability is treated as a conditional property determined by population characteristics, clinical context, measurement processes, therapeutic targets, and workflow rather than by model performance in the development dataset alone. The review argues that precision dosing requires an evidentiary chain linking interpretable model structure, stable estimation, externally evaluated prediction, calibrated uncertainty, and clinically meaningful decision consequences. Models may support therapy only within clearly defined contexts of use and should not be considered clinically ready merely because they fit data, reproduce expected simulations, or generate individualized recommendations.