Tilai Rosalina

Intravenous fluid administration is a common perioperative intervention, used to compensate for fluid loss through bleeding, evaporation and urine production. The challenge is to avoid complications related to hypovolemia, resulting in hypoxemia, and hypervolemia, causing edema, which possibly compromises organ perfusion. These complications are associated with slower patient recovery and 5% higher mortality rate.

Optimal intravenous fluid administration remains difficult due to high patient variability and a lack of sensitive patient-specific measurements of fluid status. Mathematical models may be used to gain more insight into the complex underlying physiology and eventually aid in determining patient-specific fluid administration protocols. Previously published fluid exchange models have not been introduced in the clinical work flow since they lack the opportunity to connect to clinically used indices of fluid status, such as pulse pressure variation (ppv) and stroke volume variation (svv). The aim of this thesis was to make a step towards a clinical decision support tool for fluid administration, by enhancing an existing model of fluid exchange with cardiovascular regulation, mechanical ventilation and drug administration, and formulating model output in terms of clinically measured signals.

First, we extended an existing fluid exchange model with a regulated model of cycle-averaged cardiovascular function. This combination allowed us to relate model outputs on fluid distribution in the body to standard clinical parameters, such as heart rate and arterial blood pressure. Model parameter settings were tuned to an experimental dataset, obtained in healthy volunteers subjected to intravenous infusion. The resulting model was found to correctly predict the experimentally observed response to intravenous infusion and bleeding in an independent volunteer study.

Clinically, the effect of fluid administration on hemodynamics (fluid responsiveness) can be tested by redistribution of blood through mechanical ventilation, quantified by ppv and svv. To predict these dynamic indices, we converted the model of cardiovascular hemodynamics from a cycle-averaged one into a dynamic one, and extended it with models of the pulmonary circulation and lung mechanics, to simulate the effect of spontaneous breathing and mechanical ventilation on hemodynamic function. Our results show that we can obtain realistic changes in ppv, svv, and change in cardiac index for mechanically ventilated, anesthetized patients after a fluid challenge. A preliminary sensitivity investigation showed that ppv and svv depend on intravascular volume, pericardial and vascular properties, in decreasing order. In the future, the developed mathematical model may be used for interpretation of ppv and svv on the basis of underlying patient characteristics, potentially leading to a broader use of these indices in the clinical environment.

Moving towards a clinical application, we obtained a retrospective surgical dataset in collaboration with Catharina Hospital Eindhoven. To realistically simulate the clinical patient data, we extended our model with the effect of pharmacokinetic and -dynamic models for propofol, sevoflurane, phenylephrine, ephedrine and noradrenaline. In a pilot study, we compared model predictions on blood pressure and heart rate with retrospective measured data for three surgical patients. We concluded that, to model exact intraoperative blood pressure, all anesthetics, inotropes and vasopressors must be included in the model. However, to predict an average blood pressure during surgery and fluid distribution at the end of surgery, vasopressors and inotropes can be omitted.

Finally, since most fluid related complications occur on a time scale of multiple days, we tested our fluid exchange model in an ICU setting in a prospective study with Maastricht UMC+. We tested the predictive capabilities of the model for interstitial volume, and investigated the effects of the urine production and membrane permeability on the interstitial volume model predictions. Our results show that we can simulate the trends of fluid buildup after fluid administration with our current model. Additionally, we confirmed that the interstitial volume works as a buffer, taking up most of the excess administered fluid and thus allowing vascular volume to remain fairly constant. Interestingly, on the long term, this increase could be determined directly, as the net balance of fluid administration, bleeding, urine production and perspiration, while assuming a constant vascular volume.

In conclusion, we developed and evaluated a mathematical model predicting fluid responsiveness and redistribution in critically ill patients. This may serve as a strong basis for a clinical decision support model to guide intravenous fluid administration and thereby reduce hypervolemia-related complications.