Prediction and estimation of pulmonary response and elastance evolution for volume-controlled and pressure-controlled ventilation
Introduction
Acute lung injury (ALI) and acute respiratory distress syndrome (ARDS) have detrimental impacts on lung stiffness and oxygenation resulting in high morbidity and mortality [1], [2], [3], [4]. Mechanical ventilation (MV) is the core treatment for patients suffering from life-threatening respiratory failure in the intensive care unit (ICU). The primary goal of MV is to minimise the work of breathing, ensure adequate gas exchange, and recruit and hold open lung volume to enable recovery [5], [6], [7], [8], [9]. However, suboptimal MV settings can lead to over-distension and ventilator induced lung injury (VILI), both of which increase morbidity and mortality [1], [6], [8], [9], [10], [11], [12]. To avoid these harmful effects, protective MV settings have been proposed [1], [4], [13], [14], [15].
Low tidal volume is clinically well-accepted in MV to mitigate VILI, but can lead to alveolar de-recruitment [16], [17], [18]. Thus, a protective ‘open lung’ approach uses positive end-expiratory pressure (PEEP) during breathing to prevent alveolar collapse combined with low tidal volume ventilation, thus maintaining an open lung at the end of expiration to ensure sufficient oxygenation and pressure support [6], [8], [9], [18], [19], [20], [21]. Staircase recruitment maneuvres (RMs), comprising a series of incremental and decremental PEEP steps, have been used as one important component in clinical care for recruiting lung volume for lung protective strategies and assessing the choice of PEEP [6], [15], [22], [23], [24].
RMs with subsequent PEEP changes can be effective in improving oxygenation, while minimising harm [20], [25], [26], [27]. However, optimal RM and PEEP settings remain patient-specific, time-varying, and thus not standardized [11], [12], [18], [28], [29], [30], [31]. The ‘best’ setting can be different between patients, as well as varying over different conditions and time [5], [14], [32], [33], [34], [35]. An suboptimal PEEP setting can lead to excessive or insufficient support for patients, inducing VILI and leading to higher morbidity and mortality [36], [37]. Thus, it is critical to to provide clinicians with better information to monitor patient-specific pulmonary state and forecast the influence of new PEEP settings on pulmonary response for each patient, to improve and personalize care, minimize risk, and maximize care and safety [8], [19], [34], [35]. Therefore, accurate, predictive and patient-specific MV strategies are a major need in advancing care and minimising MV-associated injury [5], [19], [34].
Furthermore, two different MV modes are both widely used, volume controlled ventilation (VCV) and pressure controlled ventilation (PCV) [38], [39], [40]. VCV allows clinicians to control tidal volume directly, eliminating volutrauma, but they need to be alert to the resulting peak inspiratory pressure (PIP) and barotrauma. Conversely, PCV controls pressure, but risks volutrauma from too large a peak inspiratory volume (PIV) [38]. Both limitations may lead to unexpected VILI [39], [41]. To date, no noticeable clinical outcome differences have been seen comparing VCV and PCV [42], [43]. Thus, the decision on MV strategies relies on clinician preference, patient characteristics, or patient comfort [38], [39]. Therefore, accurate, model-based, and patient-specific pulmonary response prediction is necessary.
In the last two decades, several complex models have been proposed and can effectively capture a large range of nonlinear pulmonary dynamics [34], [44], [45], [46], [47], [48], [49], [50], [51]. However, their complexity means they suffer poor or non-identifiability [34], [52], or are too complex to identify or apply at the bedside [53], [54], [55], [56], [57], [58], [59], thus limiting or eliminating their potential for clinical application. Far simpler black box models can be created, but require large amounts of data to train and may lack the ability to capture or describe all physiological features in various situations [58], [60], [61]. In addition, physiological relevance is important because it supports clinical confidence and use and provides further insight to clinical end-users [52], [62], but such black-box models cannot offer physiological relevance. Finally, some models of all types capture lung mechanics well with good personalization of parameters, but are poor in predicting the response to changes in care [63], [64], [65], lacking the means to offer guidance to clinicians, and suggesting the identified parameters may not be correct. However, accurate prediction is a major need in guiding and improving the safety and efficacy of clinical MV treatment [66], [67], [68]. Thus, there is a need for simpler, identifiable, physiologically relevant models which can offer accurate prediction to changes in care by capturing the evolution of lung mechanics as MV parameters change [34], [52].
Currently, while several models can identify data [9], [11], [44], [69], [70], [71], the authors are aware of only one approach able to accurately predict outcomes from changes in MV care for resulted airway pressure for VCV and tidal volume for PCV[8], [19], [72]. These approaches use physiologically relevant basis functions to define respiratory mechanics over all possible pressures and volumes seen in MV, which is uncommon [35], [58]. However, while they predict well, the basis functions proposed are independent of known PEEP levels and changes, creating complexity in understanding and implementation, especially for PCV predictions. Thus, if accurate prediction can be obtained with a simpler model where elastance evolution as PEEP changes is captured as a function of PEEP, it could provide an easier, more intuitive, and clinically applicable approach in clinical use.
This research presents physiologically relevant, simpler, basis functions to estimate elastance and resistance evolution as MV parameters change using the same clinically validated single compartment lung mechanics model as previous studies [59], [73]. It is validated by assessing prediction error when made patient specific using data from one single PEEP level to predict pressure and flow at higher PEEP levels. Such a predictive model would offer the ability to quantify the trade-off or compromise between increasing basis function simplicity and improving clinical utility.
Section snippets
Patient data
Pressure and flow data from 36 mechanically ventilated ICU patients (4 from the CURE pilot trial [74], 18 from the McREM pilot trial [19], [75], and 14 from the Maastricht pilot trial) were used to validate the method developed in this study.
Elastance evolution and prediction
Fig. 3 (a) shows an example of elastance evolution for the CURE trial Patient 4, Set 1 across 6 PEEP levels, identified at PEEP = 11 cmH2O and predicting response at higher PEEP levels. Fig. 3 (b) shows an example for Patient 15 across 6 PEEP levels in the McREM trial, identified at PEEP = 10 cmH2O and then predicting response, where T0 is the time when inspiration ends and reaching maximum tidal volume. Typical prediction cases are shown in Fig. 4, with low absolute median pressure predicted
Discussion
The personalized, predictive virtual patient model presented uses only data from the first clinically relevant PEEP level to predict the respiratory mechanics and response at higher PEEP, where ΔPEEP can be up to 20 cmH2O, a clinically unrealistically change used only to validate the model. Changing PEEP is a key setting to optimise MV care and outcomes [6], [12], [31]. This overall outcome is achieved using a relatively simple first order single compartment lung mechanics model and
Conclusion
This paper presents a predictive model with a novel, simpler model to capture nonlinear lung elastance and its evolution. In particular, this newly proposed compensatory function is validated in use to predict the pulmonary response in pressure during VCV and volume during PCV, where prediction accuracy is the key element in creating model-based control and validating these virtual patient models before clinical testing. All fixed parameter choices were checked for robustness using a
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
This work was supported by the NZ Tertiary Education Commission (TEC) fund MedTech CoRE (Centre of Research Excellence; #3705718) and the NZ National Science Challenge 7, Science for Technology and Innovation (2019-S3-CRS). The authors also acknowledge support from the EU H2020 R&I programme (MSCA-RISE-2019 call) under grant agreement #872488 — DCPM.
References (85)
- et al.
Optimising mechanical ventilation through model-based methods and automation
Annual Reviews in Control
(2019) - et al.
Positive end expiratory pressure in patients with acute respiratory distress syndrome – The past, present and future
Biomed Signal Process Control
(2012) - et al.
Ventilator-induced Lung Injury
Clin Chest Med
(2016) - et al.
A model of ventilation of the healthy human lung
Comput. Methods Programs Biomed
(2011) - et al.
A virtual patient model for mechanical ventilation
Comput. Methods Programs Biomed
(2018) - et al.
Clinical management of pressure control ventilation: An algorithmic method of patient ventilatory management to address “forgotten but important variables”
J Crit Care
(2018) - et al.
Pressure versus volume controlled modes in invasive mechanical ventilation
Medicina Intensiva (English Edition)
(2013) - et al.
Pressure-Controlled vs Volume-Controlled Ventilation in Acute Respiratory Failure: A Physiology-Based Narrative and Systematic Review
Chest
(2015) - et al.
Assessing respiratory mechanics using pressure reconstruction method in mechanically ventilated spontaneous breathing patient
Comput Methods Programs Biomed
(2016) - et al.
Extrapolation of a non-linear autoregressive model of pulmonary mechanics
Math Biosci.
(2017)