Modeling and Control of Pharmacological Systems

Personalized patient care has gained increasing attention in recent years. Precise drug dosing is critical for patient safety and good clinical outcomes, especially in intensive care units, where patients often are in critical conditions. Such treatments can include stabilizing blood pressure and heart rate or maintaining safe anesthesia levels. However, the inter-patient variability in the drug response makes finding a dosing regimen that works for all patients challenging. The problem with the most commonly used methods today is that they do not account (at least not sufficiently) for this variability, which can lead to under- or overdosing.This thesis aims to solve these issues by improving modeling and control strategies for individualized drug dosing. These aims are to: 1) stabilize heart donor hemodynamics to enhance organ quality for transplantation, 2) streamline the identification of covariate models that capture the inter-patient variability in the drug response, and 3) develop control strategies resilient to disturbances and poor measurement signal quality. First, we demonstrate that precise blood pressure control can delay ischemic myocardial contracture in heart donors. However, the controller performance was limited by the inter-patient variability in drug response, which motivated further research on drug modeling. Therefore, we developed a method to automate the covariate modeling process using symbolic regression networks, which enabled us to find simple and interpretable models that capture this variability well. To evaluate the covariate model’s performance, we needed to simulate a large dataset, which motivated the development of a fast simulator for pharmacokinetics. Therefore, we developed an efficient simulator that could simulate a large dataset in a fraction of the time compared to current available methods. Returning to the control problem, we proposed combining open- and closed-loop control for anesthesia using a Kalman filter. This allowed for robust control performance even when model errors, disturbances, and poor signal quality were present.In conclusion, these contributions demonstrate how pharmacological modeling and control can improve drug dosing accuracy and patient safety. Adopting the methods provided in this thesis can lead to safer and more efficient healthcare.