Reconstruction of Pressure Support Ventilation Signals: A Virtual Patient Set

Abstract

Protection of the lungs and diaphragm is imperative to the safety of a patient receiving pressure support ventilation in the intensive care unit. To this end, accurate modeling of patient–ventilator interactions that occur within a breath is a vital step. Modeling of these interactions may be useful to better understand interactions that compromise patient safety, test in-silico new techniques to estimate physiological signals, and ultimately deliver safe ventilation. However, undisclosed and highly nonlinear internal ventilator dynamics hinder the derivation of such a model. Instead, in this paper, interactions were derived from clinical data, considering 400 breaths per patient to construct a set of 10 virtual patients. A simple first-order system was utilized to describe the interactions, with appropriate correlations between the system’s gain and time constant, and the magnitude of the patient’s effort to generalize the model. In parallel, generalized patient respiratory effort profiles were derived by analyzing similarities in measured efforts. Reconstruction of ventilator waveforms, utilizing the virtual patients, was achieved with a median accuracy greater than 85% in the worst case. A potential use case is also presented, further demonstrating the value of the presented virtual patients for in-silico development and validation of novel techniques. The derived virtual patients are shared via an online repository, and sufficient information is provided for readers to derive additional virtual patients.