Optimized therapeutic drug monitoring: the role of machine learning models

Traditional therapeutic drug monitoring (TDM) faces limitations in accuracy and adaptability, often failing to optimize therapy for complex patients. Machine learning (ML) is emerging as a powerful tool to overcome these challenges, offering a data-driven paradigm to enhance therapeutic outcomes and minimize toxicity for drugs with narrow therapeutic indices.

This review synthesizes the evolution of ML in TDM. We cover foundational models that predict drug exposure from sparse data using either real-world or simulation-based training. We then explore the extension of these techniques to proactive first-dose optimization and the recent development of hybrid models, which integrate the physiological interpretability of population pharmacokinetic frameworks with the corrective power of ML.

The future of TDM lies not in replacing mechanistic models, but in their convergence with ML. While promising, clinical translation requires overcoming critical barriers in data access, model interpretability, and workflow integration. The long-term trajectory points toward dynamic Digital Twins capable of forecasting patient-specific benefit-risk profiles. Ultimately, validated hybrid tools embedded in clinical decision support systems could establish proactive, individualized dosing as the new standard of care in personalized pharmacotherapy.