Model-Based Biomarker Selection for Dose Individualization of Tyrosine-Kinase Inhibitors

Tyrosine-kinase inhibitors (TKIs) demonstrate high inter-individual variability with respect to safety and efficacy and would therefore benefit from dose or schedule adjustments. This study investigated the efficacy, safety, and economical aspects of alternative dosing options for sunitinib in gastro-intestinal stromal tumors (GIST) and axitinib in metastatic renal cell carcinoma (mRCC). Dose individualization based on drug concentration, adverse effects, and sVEGFR-3 was explored using a modeling framework connecting pharmacokinetic and pharmacodynamic models, as well as overall survival. Model-based simulations were performed to investigate four different scenarios: (I) the predicted value of high-dose pulsatile schedules to improve clinical outcomes as compared to regular daily dosing, (II) the potential of biomarkers for dose individualizations, such as drug concentrations, toxicity measurements, and the biomarker sVEGFR-3, (III) the cost-effectiveness of biomarker-guided dose-individualizations, and (IV) model-based dosing approaches versus standard sample-based methods to guide dose adjustments in clinical practice. Simulations from the axitinib and sunitinib frameworks suggest that weekly or once every two weeks high-dosing result in lower overall survival in patients with mRCC and GIST, compared to continuous daily dosing. Moreover, sVEGFR-3 appears a safe and cost-effective biomarker to guide dose adjustments and improve overall survival (€36 784.- per QALY). Model-based estimations were for biomarkers in general found to correctly predict dose adjustments similar to or more accurately than single clinical measurements and might therefore guide dose adjustments. A simulation framework represents a rapid and resource saving method to explore various propositions for dose and schedule adjustments of TKIs, while accounting for complicating factors such as circulating biomarker dynamics and inter-or intra-individual variability.