TY  - JOUR
T1  - Artificial Intelligence in Clinical Pharmacokinetics and Pharmacodynamics: A Realist Review of Contexts, Mechanisms, Outcomes, and Failure Modes
A1  - Carlos Ramirez
A1  - Elena Torres
A1  - Pablo Ortega
A1  - Sofia Mendes
JF  - International Journal of Pharmaceutical Research and Allied Sciences
JO  - Int J Pharm Res Allied Sci
SN  - 2277-3657
Y1  - 2026
VL  - 15
IS  - 2
DO  - 10.51847/Ppht0M5a9H
SP  - 61
EP  - 72
N2  - Artificial intelligence is increasingly used to predict pharmacokinetic parameters, characterize exposure–response relationships, select population models, interpret therapeutic drug-monitoring data, and support individualized dosing. However, analytical performance alone does not explain why apparently similar systems succeed in one setting, fail in another, or create unintended clinical consequences. This realist review examines how data, model, patient, organizational, and decision contexts influence the mechanisms and outcomes of artificial-intelligence-enabled pharmacokinetic and pharmacodynamic applications. The initial programme theory proposes that these applications are most likely to support useful decisions when a clearly defined clinical task is matched to representative longitudinal data, appropriate pharmacological structure, calibrated uncertainty, explicit therapeutic targets, and workflows that allow clinicians to interrogate or reject recommendations. Evidence was identified through purposive and iterative searching, appraised for relevance to programme-theory development and rigour of the supporting inference, and synthesized through context–mechanism–outcome configurations, negative cases, demi-regularities, and rival explanations. The reviewed evidence suggests that performance may improve when machine learning captures residual heterogeneity, integrates complex covariates, or reduces dependence on a single misspecified population model. Conversely, incomplete sampling, biased labels, inappropriate priors, weak validation, miscalibration, distribution shift, and automation-related reliance may amplify errors as predictions are converted into doses. The refined programme theory therefore treats artificial intelligence not as an autonomous dosing intervention but as a context-dependent resource embedded within pharmacological models, measurement systems, clinical reasoning, governance, and monitoring. Its clinical value remains conditional on alignment across these components and cannot be inferred from computational performance alone. 
UR  - https://ijpras.com/article/artificial-intelligence-in-clinical-pharmacokinetics-and-pharmacodynamics-a-realist-review-of-conte-6xg6vkqsvzm7yjp
ER  -