Real time noninvasive estimation of work of breathing using facemask leak-corrected tidal volume during noninvasive pressure support: validation study

Abstract

We describe a real time, noninvasive method of estimating work of breathing (esophageal balloon not required) during noninvasive pressure support (PS) that uses an artificial neural network (ANN) combined with a leak correction (LC) algorithm, programmed to ignore asynchronous breaths, that corrects for differences in inhaled and exhaled tidal volume (V_T) from facemask leaks (WOB_ANN,LC/min). Validation studies of WOB_ANN,LC/min were performed. Using a dedicated and popular noninvasive ventilation ventilator (V60, Philips), in vitro studies using PS (5 and 10 cm H₂O) at various inspiratory flow rate demands were simulated with a lung model. WOB_ANN,LC/min was compared with the actual work of breathing, determined under conditions of no facemask leaks and estimated using an ANN (WOB_ANN/min). Using the same ventilator, an in vivo study of healthy adults (n = 8) receiving combinations of PS (3–10 cm H₂O) and expiratory positive airway pressure was done. WOB_ANN,LC/min was compared with physiologic work of breathing/min (WOB_PHYS/min), determined from changes in esophageal pressure and V_T applied to a Campbell diagram. For the in vitro studies, WOB_ANN,LC/min and WOB_ANN/min ranged from 2.4 to 11.9 J/min and there was an excellent relationship between WOB_ANN,LC/breath and WOB_ANN/breath, r = 0.99, r² = 0.98 (p < 0.01). There were essentially no differences between WOB_ANN,LC/min and WOB_ANN/min. For the in vivo study, WOB_ANN,LC/min and WOB_PHYS/min ranged from 3 to 12 J/min and there was an excellent relationship between WOB_ANN,LC/breath and WOB_PHYS/breath, r = 0.93, r² = 0.86 (p < 0.01). An ANN combined with a facemask LC algorithm provides noninvasive and valid estimates of work of breathing during noninvasive PS. WOB_ANN,LC/min, automatically and continuously estimated, may be useful for assessing inspiratory muscle loads and guiding noninvasive PS settings as in a decision support system to appropriately unload inspiratory muscles.

Appendix: Artificial neural network for estimating noninvasive work of breathing (previously published10)

Five predictor variables/input elements, used in a multi-layer perceptron artificial neural network (ANN) model (figure below), found to be highly correlated with noninvasive work of breathing (WOB) per minute, were chosen because they produced the best (lowest mean squared error) possible predictive value. Incorporating inputs in addition to those five variables did not increase predictive performance, thus this set was considered sufficient for predicting/calculating noninvasive WOB per minute. (1) Minute ventilation, the spontaneous minute ventilation (not including mandatory breaths), was found to be closely tied to noninvasive WOB per minute; the larger the minute ventilation, the greater the workload and vice versa. (2) Intrinsic positive end expiratory pressure (PEEPi), is the parameter estimated from the exhalation portion of a flow-volume loop. In patients for whom PEEPi was suspected, we observed that exhaled flow did not reset/return to zero but to some discrete value at end-exhalation on a flow-volume loop. By linearly extrapolating this value to the zero flow axis on the flow-volume loop, an estimate of air trapping or excess volume remaining in the lungs at end-exhalation is made. By knowing this volume and respiratory system compliance, a pressure is calculated (pressure = volume/compliance), a reflection of PEEPi. The larger the estimated PEEPi value, the larger the workload and vice versa. Respiratory system compliance and resistance were estimated in real time using the expiratory time constant method (Al-Rawas et al., Expiratory time constant for determinations of plateau pressure, respiratory system compliance, and total resistance. Critical Care 2013; 17: R23). (3) Airway pressure trigger depth is a parameter that measured the decrease in pressure below baseline airway pressure (Paw) at the Y-piece of the breathing circuit just before the ventilator triggers “ON.” Large inspiratory efforts tend to decrease this pressure more rapidly, causing a large decrease in pressure or trigger pressure depth. The larger the decreases in trigger pressure depth, the larger the workload and vice versa. (4) Inspiratory rate of flow rise is the parameter that assessed how rapidly the inspiratory flow waveform rose during a pressure supported breath. An actively breathing patient with a strong inspiratory effort and demanding a high flow rate from the ventilator tends to display a rounded or sinusoidal-shaped inspiratory flow profile; peak flow occurs during the mid to latter portion of the breath. It takes a longer time for flow to reach maximum during inhalation. In contrast, a patient who inhales passively while receiving a pressure-supported breath typically displays a rapid rise in flow very early in the breath. Flow reaches maximum at nearly the onset of the breath in a very brief time and then decelerates for the remainder of inhalation. A coded value for inspiratory rate of flow rise was devised to range from zero to one. The higher the value, the more rounded or sinusoidal the inspiratory flow profile, and the greater the workload. The lower the value, the more a decelerating inspiratory flow profile, and the lower the workload. (5) Pmus is the respiratory muscle pressure; it is the sum of elastic and resistive pressures and was determined using the equation of motion, i.e., Pmus + Paw = (tidal volume/respiratory system compliance) + (respiratory system resistance × inspiratory flow rate). This is a reflection of the pressure generated by the inspiratory muscles during spontaneous inhalation; the larger the value, the larger the estimated effort to inhale and vice versa.