Quantifiable identification of flow-limited ventilator dyssynchrony with the deformed lung ventilator model

Abstract

AbstractObjectiveVentilator dyssynchrony (VD) is potentially harmful to patients with or at risk for acute respiratory distress syndrome (ARDS). In addition to injury solely caused by the ventilator, ventilator-induced lung injury may be instigated and exacerbated by patient respiratory efforts. Automated detection of VD from ventilator waveforms is challenging, and efforts have been made on a human-guided ML algorithm to detect some types of VD. We currently lack a methodological ability to define sub-breath phenotypes of VD that quantify severity anchored to physiologic understanding that could be used to relate VD to damage and guide ventilator management.Materials and MethodsA mathematical model is developed that represents the pressure and volume waveform signals of a breath into several pathophysiological temporal features observed in ventilator waveforms and then deformation terms are added corresponding to hypothesized flow-limited (FL) dyssynchronous breaths. Model parameters are estimated at the resolution of a single breath using a deterministic, multivariate, constrained interior-point method to create a parametric representation of breaths. Differential estimates of different FL-VD breaths are used to create severity metrics for FL-VD breaths and their associations with the ventilator settings and healthcare interventions are analyzed.ResultsA total of 93,007 breaths were analyzed from the raw ventilator waveform dataset of 13 intensive care unit patients who met inclusion criteria. A quantitative method was developed to determine the continuously varying FL-VD severity for each breath and was successfully applied to a cohort of patient-ventilator waveform data. Additionally, cross-validation, using a previously developed ML categorical VD identification algorithm, produced an area under the receiver operator curve of 0.97.Discussion & ConclusionThe VD-deformed lung ventilator (VD-DLV) model accurately detects FL-VD breaths and is able to quantify the severity of patient effort during patient-ventilator interaction. The presence and severity of deviations from normal are modeled in a way that is based on physiological hypotheses of lung damage and ventilator interactions. Therefore, the computed phenotypes have the predictive power to determine how the healthcare variables are associated with FL-VD breaths. This work paves the way for a large-scale study of VD causes and effects by identifying and quantifying VD breaths using the VD-DLV model.