Virtual Patient Modeling and Prediction Validation for Pressure Controlled Mechanical Ventilation

Abstract Respiratory failure patients in the intensive care unit (ICU) require mechanical ventilation (MV) to support breathing and tissue oxygenation. Optimizing MV care is problematic. Significant patient variability confounds optimal MV settings and increase the risk of lung damage due to excessive pressure or volume delivery, which in turn can increase length of stay and cost, as well as mortality. Model-based care using in silico virtual patients can significantly affect ICU care, personalizing delivery and optimising care. This research presents a virtual patient model for pressure-controlled MV, an increasingly common mode of MV delivery, based on prior work applied to volume-controlled MV. This change necessitates predictions of flow and thus volume, instead of pressure, as the unspecified variable. A model is developed and validated using clinical data from five patients (N=5) during a series of PEEP (positive end expiratory pressure) changes in a recruitment maneuver (RM), creating a total of 242 predictions. Peak inspiratory volume, a measure of risk of lung damage, errors were 56 [26-95]mL (10.6 [5.3-19.1]%) for predictions of PEEP changes from 2-16cmH2O. Model fitting errors were all lower than 5%. Accurate predictions validate the model, and its potential to both personalise and optimise care.