Model-based hemodynamic management of critically ill patients.

Hemodynamic monitoring and therapy aims to ensure adequate circulatory function, and thus organ perfusion. However, achieving this goal is challenging due to the inability to directly monitor organ perfusion, difficulty and ambiguity in ascertaining the most appropriate treatment strategy, and highly variable and complex patient response to therapy. Hence, effective measurements and protocols to clarify hemodynamic management and optimise outcomes are urgently needed to meet growing demand for intensive care arising from aging populations and rising rates of chronic disease. This thesis explores model-based solutions to provide real-time, non-additionally-invasive hemodynamic assessment for critical care. These models have the advantage of directly accounting for intra- and inter- patient variability by identifying patient-specific parameters from time-varying data, and monitoring their evolution over time and care. Model validation is performed using data from experimental pig trials, which provide high-fidelity / invasive monitoring not feasible in humans. Further validation is from two clinical databases: VitalDB, from surgical patients; and BedMasterDB, from Christchurch Hospital Intensive Care Unit. This personalised model-based approach is used to provide insight and objective assessments not possible from current measurements. Several clinically-applicable, non-additionally-invasive hemodynamic models are developed and validated in this thesis. Two beat-to-beat cardiac stroke volume estimation methods, the constant-Z windkessel and tube-load models, are shown to outperform existing methods for pig trial and clinical validation. Next, a model is presented to estimate preload changes, Frank-Starling curves, and contractility. The model performed well for the pig trial and for the VitalDB data had reasonable contractility estimation, but poor preload estimation accuracy, providing proof-of-concept. Finally, a model is developed for predicting stroke volume changes in response to fluid therapy. This model had good performance for the pig trial, and a proposed clinical trial design for further validation is presented. All these models together address key elements and insight required to simplify, personalise, and optimise hemodynamic management. This thesis also delivers the design and validation of a low-cost, open-source data acquisition system enabling direct recording of patient arterial pressure waveforms at the same time as a clinical monitor, and measurement of fluid infusion rate. The system is a useful tool for hemodynamic monitoring research, and reduces the high cost barrier of acquiring waveform data from medical monitoring systems, which stifles innovation. Overall, clinically-applicable model-based methods are developed in this thesis to deliver non-additionally-invasive stroke volume monitoring, assessment of cardiac preload responsiveness / contractility, and fluid responsiveness prediction in real-time at the patient bedside. Together, these outputs can provide patient-specific assessment of hemodynamic state and response to therapy. These outputs greatly enrich the information available to clinicians and potentially enable a smarter, more personalised approach to hemodynamic management to improve patient outcomes