Model-Based Control Strategies for Personalized Leukemia Treatment

Abstract

Chemotherapy of acute leukemia is important in ensuring long-term positive outcomes, yet standard dosing protocols use reactive adjustments based on infrequent blood tests and do not address high inter-patient variability. This thesis develops and analyzes a series of model-based control strategies to personalize chemotherapy and maintain the Absolute Neutrophil Count (ANC) near a target concentration. The thesis begins with an analysis of the Friberg model of myelosuppression, and explicit stability regions are derived. State and output feedback controllers are designed for the Friberg model, establishing local asymptotic stability guarantees through Lyapunov and frequency-domain analysis. To address the clinical challenges of unknown parameters, infrequent measurements, and dosing constraints, the thesis culminates in an Adaptive Model Predictive Control (AMPC) framework. This practical approach integrates a joint Unscented Kalman Filter (UKF) for online state and parameter estimation with a predictive controller based on the comprehensive Jost pharmacokinetic/pharmacodynamic (PK/PD) model. Simulations using a population of 116 virtual patients are used to validate each controller, and suggest that a more conservative ANC target can enable safe regulation with lower risk of neutropenia and lower drug exposure. Overall, this work presents a progression from theoretical to practical solutions, highlighting the potential of advanced control strategies to improve the safety and effectiveness of leukemia treatment.