Modeling bacterial clearance from the bloodstream using computational fluid dynamics and Monte Carlo simulation

Abstract

Listen

Translate article

(HOB) elevation (between 00 and 300) and mild hypoand hyperventilation (achieved by changing the respiratory rate, RR). Each experimental session consisted of between 1 and 6 changes in HOB and/or RR over periods varying from 15 to 130 minutes. Multiple sessions were recorded for 6 patients. ICP and other signals were recorded continuously, along with clinical annotations to indicate the precise timing of the physiologic challenges. Data from early in single long session or from prior sessions were used to estimate patient-specific parameter values for a computer model of ICP dynamics that is similar to other models reported in the literature. The parameter estimates were calculated using a curvefitting optimization algorithm with the objective of minimizing the squared error between the ICP calculated by the model and the actual ICP data. The resulting patient-specific models were then used to predict the patient's ICP response to interventions at later time periods, either later in the same session or during subsequent sessions. Results: Mean absolute error (MAE) between model-calculated ICP and the actual data averaged 1.9 mm Hg for full sessions and 1.7 mm Hg for partial sessions. The average mean absolute deviation (MAD) for these segments was 3.1 and 2.4 mm Hg, respectively. The MAE for the predictions was 4.0 mm Hg within the same session and 6.7 mm Hg across sessions. Conclusions: Despite the achievement of low model errors in the training segments, the error in predicted segments was too large for the model to be useful clinically, that is to say, prediction is much more difficult than explanation. One implication for researchers is that a degree of skepticism is warranted: the fact that a model can be made to fit historical data does not mean that the model will be able to predict anything. These results could be seen as tending to support the general lack of interest from clinicians regarding computer models of ICP dynamics.