Data-Driven Approaches Applied to Electrochemical Measurements for Therapeutic Drug Monitoring in Precision Medicine

Precision medicine increasingly relies on therapeutic drug monitoring (TDM) to individualize dosing and improve clinical outcomes. By integrating pharmacokinetic models with patient-specific observations, model-informed precision dosing (MIPD) enables quantitative dose adaptation based on measured drug concentrations. However, conventional laboratory-based TDM workflows are often slow, resource-intensive, and poorly suited for rapid clinical decision-making. In contrast, emerging point-of-care (POC) electrochemical sensing technologies offer the potential for rapid and decentralized drug monitoring. Yet, extracting reliable quantitative information from electrochemical measurements remains challenging due to signal interference, data scarcity, and measurement uncertainty, which together hinder the effective translation of electrochemical sensing into clinical dosing workflows.

This thesis develops data-driven approaches for the interpretation and utilization of electrochemical measurements in therapeutic drug monitoring and precision medicine. First, an unsupervised pipeline is proposed to automatically extract parametric representations from a limited number of electrochemical measurements and generate large-scale simulated datasets. By combining signal decomposition with data augmentation, this approach addresses the problem of data scarcity and enables systematic learning without requiring extensive experimental campaigns. Second, an interaction-aware machine learning architecture is introduced to identify and quantify drug concentrations from complex electrochemical signals. The proposed model explicitly captures nonlinear signal interactions that arise in multi-analyte environments, enabling robust concentration estimation even in the presence of overlapping electrochemical responses. Third, the thesis extends beyond signal interpretation to the downstream clinical use of these measurements by incorporating measurement uncertainty and sampling-time variability into dosage adaptation. A robustness-oriented evaluation framework is developed to assess the reliability and safety of dosing recommendations under realistic observation uncertainty. In contrast to conventional accuracy-based metrics, the framework evaluates the stability of dosing decisions with respect to perturbations in measured concentrations and sampling times.

By linking data-driven electrochemical signal interpretation with uncertainty-aware precision dosing, this work establishes a coherent pipeline from raw electrochemical measurements to clinically meaningful therapeutic decisions. The proposed approaches contribute methodological advances for point-of-care TDM and provide a foundation for robust precision medicine systems operating under real-world measurement uncertainty.