Logo: to the web site of Uppsala University

In contemporary cancer care, several issues are garnering increasing attention. First, significant inter-individual variability among patients challenges the effectiveness of a uniform dosing approach. Second, the escalating costs of treatments necessitate careful consideration when selecting doses and other clinical modalities, including biomarkers, while balancing economic constraints. The objective of this thesis was to evaluate techniques for tailoring doses and guiding clinical decisions for cancer patients through the development and implementation of various models, with the aim of improving treatment outcomes in terms of both efficacy and safety. 

Through a model-based framework integrating sunitinib pharmacokinetics (PK), adverse events, biomarkers, tumor dynamics and their correlation with overall survival, different treatment schedules and biomarkers for dose individualization were explored. Based on the proposed threshold values, neutrophil count (ANC) and the biomarker sVEGFR-3 were identified as offering the best balance between safety and efficacy for sunitinib in gastro-intestinal stromal tumors (GIST) and could thus serve as viable guides for dose individualization in clinical practice. Given its routine measurement, dose adjustments guided by ANC may be preferable in clinical settings. The feasibility of utilizing diastolic blood pressure (dBP) for personalized dose optimization of tyrosine-kinase inhibitors in clinical settings is constrained due to its reliance on repeated measurements taken at consistent intervals. 

For axitinib and sunitinib, model-based predictions using multiple clinical measurements were more accurate than single sample measurements. For drugs with high unexplained inter-individual variability (IIV), low residual variability (RUV), and low inter-occasional variability (IOV), therapeutic drug monitoring (TDM) provided a more accurate measure of exposure. Conversely, for drugs with low IIV and high RUV and IOV, pharmacogenetic profiling was more suitable. However, the prevalence of pharmacogenetic subtypes and the challenge of measuring exposure metrics like AUC through limited sampling also influence these approaches.

This research further emphasizes how model structure affects the outcomes of cost-effectiveness analyses and consequently the potential implications for regulatory decisions. Although creating mechanistic models for these analyses demands substantial initial effort, the growing need for model-based analyses in drug approval is likely to make these models more accessible for future compounds. Moreover, such models are expected to be more biologically plausible and therefore more reflective of reality and offer flexibility for exploring alternative dosages with limited additional effort.

Using model-based assessments, the relationship between the PK and PK-pharmacodynamic (PKPD) profiles of adverse events arising from therapies for acute lymphocytic leukemia were established. For PEG-asparaginase, the PK model categorized 93% of patients who experienced inactivation against PEG-asparaginase as having an increased clearance, and 86% of patients who did not experience hypersensitivity as maintaining stable clearance throughout their asparaginase treatment. This approach marks a potential method for predicting inactivation by identifying early changes in clearance. For vincristine, model-informed precision dosing was shown to reduce the incidence of vincristine-induced peripheral neuropathy (VIPN) from 62.1% to 53.9%, though the clinical impact remains modest.