ORIGINAL RESEARCH published: 07 July 2021 doi: 10.3389/fmedt.2021.690442 Patient-Ventilator Synchronization During Non-invasive Ventilation: A Pilot Study of an Automated Analysis System Christophe Letellier 1*, Manel Lujan 2,3 , Jean-Michel Arnal 4 , Annalisa Carlucci 5 , Michelle Chatwin 6 , Begum Ergan 7 , Mike Kampelmacher 8 , Jan Hendrik Storre 9,10 , Nicholas Hart 11 , Jesus Gonzalez-Bermejo 12,13 and Stefano Nava 14 1 Normandie Université — CORIA, Avenue de l’Université, Saint-Etienne du Rouvray, France, 2 Servei de Pneumologia, Corporació Parc Taulí, Sabadell, Spain, 3 Departament de Medicina, Universitat Autònoma de Bellaterra, Barcelona, Spain, 4 Service de Réanimation Polyvalente, Unité de Ventilation à domicile, Hôpital Sainte Musse, Toulon, France, 5 Pulmonary Rehabilitation, Istituti Clinici Scientifici Maugeri, Istituto di Ricovero e Cura a Carattere Scientifico, Pavia and Department of Medicine and Surgery, Respiratory Diseases, University of Insubria, Varese-Como, Italy, 6 Clinical and Academic Department of Sleep and Breathing, Royal Brompton & Harefield, National Health Service Foundation Trust, London, United Kingdom, 7 Division of Intensive Care, Department of Pulmonary and Critical Care, Faculty of Medicine, Dokuz Eylul University, Izmir, Turkey, 8 Department of Pulmonology, Antwerp University Hospital and Antwerp University, Antwerp, Belgium, 9 Department of Pneumology, University Medical Hospital, Freiburg, Germany, 10 Pneumologie Solln, Munich, Germany, 11 Lane Fox Clinical Respiratory Physiology Research Centre, Guy’s and St Thomas’ NHS Foundation Trust, London, United Kingdom, 12 Sorbonne Université, INSERM, UMRS1158 Neurophysiologie Respiratoire Expérimentale et Clinique, Paris, France, 13 AP-HP, Groupe Hospitalier Pitié-Salpêtrière Charles Foix, Service de Soins de Suites et réhabilitation Edited by: respiratoire-Département R3S, Paris, France, 14 Respiratory and Critical Care, Sant’Orsola Malpighi Hospital, Alma Mater Dinesh Kumar, Studiorum, University of Bologna, Department of Specialistic, Diagnostic and Experimental Medicine (DIMES), Bologna, Italy RMIT University, Australia Reviewed by: Behzad Aliahmad, Background: Patient-ventilator synchronization during non-invasive ventilation (NIV) can RMIT University, Australia be assessed by visual inspection of flow and pressure waveforms but it remains time Mohammod Abdul Motin, Rajshahi University of Engineering and consuming and there is a large inter-rater variability, even among expert physicians. TM Technology, Bangladesh SYNCSMART software developed by Breas Medical (Mölnycke, Sweden) provides an *Correspondence: automatic detection and scoring of patient-ventilator asynchrony to help physicians Christophe Letellier in their daily clinical practice. This study was designed to assess performance of the christophe.letellier@univ-rouen.fr automatic scoring by the SYNCSMART software using expert clinicians as a reference in Specialty section: patient with chronic respiratory failure receiving NIV. This article was submitted to Environmental Psychology, Methods: From nine patients, 20 min data sets were analyzed automatically by a section of the journal SYNCSMART software and reviewed by nine expert physicians who were asked to Frontiers in Medical Technology score auto-triggering (AT), double-triggering (DT), and ineffective efforts (IE). The study Received: 02 April 2021 procedure was similar to the one commonly used for validating the automatic sleep Accepted: 28 May 2021 Published: 07 July 2021 scoring technique. For each patient, the asynchrony index was computed by automatic Citation: scoring and each expert, respectively. Considering successively each expert scoring Letellier C, Lujan M, Arnal J-M, as a reference, sensitivity, specificity, positive predictive value (PPV), κ-coefficients, and Carlucci A, Chatwin M, Ergan B, Kampelmacher M, Storre JH, Hart N, agreement were calculated. Gonzalez-Bermejo J and Nava S Results: The asynchrony index assessed by SYNSMART was not significantly different (2021) Patient-Ventilator Synchronization During Non-invasive from the one assessed by the experts (18.9 ± 17.7 vs. 12.8 ± 9.4, p = 0.19). When Ventilation: A Pilot Study of an compared to an expert, the sensitivity and specificity provided by SYNCSMART for DT, AT, Automated Analysis System. Front. Med. Technol. 3:690442. and IE were significantly greater than those provided by an expert when compared to doi: 10.3389/fmedt.2021.690442 another expert. Frontiers in Medical Technology | www.frontiersin.org 1 July 2021 | Volume 3 | Article 690442 Letellier et al. Patient-Ventilator Synchronization During Non-invasive Ventilation Conclusions: SYNCSMART software is able to score asynchrony events within the inter-rater variability. When the breathing frequency is not too high (< 24), it therefore provides a reliable assessment of patient-ventilator asynchrony; AT is over detected otherwise. Keywords: non-invasive ventilation, patient ventilator asynchrony, chronic obstructive pulmonary disease, ineffective triggering, monitoring, automatic scoring 1. INTRODUCTION works from the pressure and airflow waveforms sampled at the device rate (here 64 Hz). This study aimed to is to assess the Nocturnal non-invasive ventilation (NIV) is recognized as an ability of this algorithm to detect the three main PVA events effective treatment for chronic hypercapnic respiratory failure. during pressure support ventilation (PSV) by only processing Monitoring NIV during sleep is a necessary adjunct to daytime flow and pressure waveforms. assessment of the treatment, which may affect prognosis, quality of sleep, or morning dyspnea (1–4). A systematic approach for determining undesired events such as leaks, upper airway 2. MATERIALS AND METHODS obstruction (UAO) with or without a decrease in respiratory drive, and patient-ventilator asynchrony (PVA) from polygraphy Nine stable patients monitored under NIV (PSV mode with performed under NIV was provided by the SomnoNIV group a backup frequency) in the Pneumology Unit Care of the (5–7). An asynchrony index (AI) greater than 10% is quite often Corporació Parc Tauli (Sabadell, Spain) were included in observed (8–12) and may be sleep-dependent (13). PVA is often this study. The data were randomly selected from a study the source of discomfort (11, 14). During long-term home NIV, focused on the prevalence of AEs. This study was approved a high prevalence of ineffective efforts (IEs) in patients with by the local ethics committee of Corporació Parc Tauli obstructive and restrictive diseases using polygraphic assessment (CIR2010/015). Written informed consent was obtained from was reported by Fanfulla et al. (13) and Guo et al. (10) and was patients. Nine expert physicians from France, Germany, Italy, later confirmed by Ramsay et al. with a parasternal EMG (12). The Netherlands, Spain, Turkey, and the United Kingdom In general, waveforms of pressure and flow contain all reviewed the automatic scorings. the necessary information required for identifying the type of Two restrictive patients, one obstructive patient, three patients asynchrony events (AEs) (6, 7, 9, 11, 15–19). Pressure waveforms with chronic obstructive pulmonary disease (COPD), and three provide a direct access to ventilator cycles. The flow results from patients with amyotrophic lateral sclerosis (ALS) were selected. a combination between ventilator cycles and patient breathing They were ventilated with different devices: three with a Vivo cycles. Interpreting these waveforms is not always simple, 40 (Beas Medical, Mölnlycke, Sweden), four with a Lumis 150, primarily because the interplay between these two cycles is not one with an Astral 150 (ResMed, North Ryde, Australia), and trivial. Indeed, the ability of intensive care unit (ICU) physicians one with a Trilogy (Philips Respironics, Murrysville, USA). Each to score AEs is quite low (20, 21) and scarcely implemented due patient used a full-face mask. The waveforms were recorded to a lack of skills (22). Nevertheless, IEs can be reliably detected by in napping patients in their most comfortable position during visual inspection (19). Specific training in mechanical ventilation the initiation of mechanical ventilation or during routine increases the ability to identify AEs from waveforms, but the controls in a quiet room. Flow Q, airway pressure Paw , and years of experience are not necessarily associated with a better belt waveforms Bthorax and Babdom were measured. The data ability to recognize the three main AEs [IE, double-triggering were acquired using an external polygraph (Powerlab 16Sp, (DT), and auto-triggering (AT), the latter one being less often ADInstruments, Sydney, Australia), equipped with a pressure well-recognized] (21). transducer (model 1050, ADInstruments, Sydney, Australia) and An automatic analysis of the pressure and flow waveforms a pneumotachograph (model S300, instrumental dead space is required to substantially shorten the long duration spent 70 ml, resistance rp = 0.0018 cmH2 O/l/s, ADInstruments, performing visual scoring (23). It was shown that detection of IEs Sydney, Australia), both inserted in the ventilation circuit close by the means of an algorithm applied to the flow and pressure to the mask and with respiratory inductance plethysmography in the ventilation circuit was possible (24–29). All studies, but belts (Pro-Tech, Canada). The polygraph was connected to a two, were performed in invasive ventilation where the leak is computer equipped with data acquisition software (Chart 7.0, not relevant. Moreover, either only IEs were often considered ADInstruments, Sydney, Australia). The sampling frequency of (24, 25, 27) or only AEs without any further specifications (26, measurements was fs = 1, 000 Hz, but the data were then 28, 30). DT has only been considered in two studies (29, 31). resampled at the frequency fV = 64 Hz, as used by SyncSmart ATs were never investigated with an automatic detection. It is software. The acquired data were read by the SyncSmart software therefore important to develop an algorithm to automatically in a text format. detect the three main AEs (IE, DT, and AT) for NIV. The For each patient, we recorded a long session of NIV. One 20- TM SyncSmart software was developed for such purpose, which min window of data was extracted from each of the recordings: Frontiers in Medical Technology | www.frontiersin.org 2 July 2021 | Volume 3 | Article 690442 Letellier et al. Patient-Ventilator Synchronization During Non-invasive Ventilation FIGURE 1 | Excerpt from the 20-min times series recorded in patient 1. The leakage 8 is averaged over each ventilatory cycle. The belt signals Bthorax and Babdom were provided to the physicians for reviewing the automatic scoring. The ventilatory cycles at t = 713 s and t = 728 s (marked with red cross) were automatically discarded by SyncSmart due to its excessively turbulent aspect and because the leakage 8 exceeds the threshold 8t , here equal to 18 l min−1 . The selected data window starts at least 5 min after the beginning at which the next inspiratory effort is initiated). According to this of the session to avoid transient patient-ventilator interactions as equation, the leakage 8 has a constant value over each cycle. commonly observed. No other criterion was used to select these This leakage is used to discard the excessively turbulent parts windows. The belt signals, only used by the expert physicians of the waveforms. As exemplified in Figure 1, it is impossible to to review the automatic scoring produced by the SyncSmart reliably score these cycles, neither by visual inspection by expert software, were filtered to remove the long-term drift and to physicians nor by an automatic algorithm. The corresponding improve their readability. A typical excerpt of the four measured ventilatory cycles are thus marked with a red cross and are waveforms with the total (intentional and non-intentional) discarded from the statistical analysis. The events scored by leakage computed by SyncSmart is shown in Figure 1. SyncSmart software are shown in Figure 2 and are defined The SyncSmart software is analyzes the pressure Paw and the as follows. flow Q, either measured by the ventilator or by the external sensors as in the present protocol. The SyncSmart software DT Double-triggered ventilatory cycles for which there are two considers three ventilator settings: expiratory positive airway pressure rises during one inspiratory effort; pressure (EPAP), inspiratory positive airway pressure (IPAP), and AT Auto-triggered ventilatory cycles when the pressure rise is the backup frequency (fbck ) at which the ventilator delivers the not triggered by an inspiratory effort and thus occurs during pressure cycles. The software computes a leakage the expiratory phase; Z tend IE Ineffective efforts when there is a patient inspiratory effort, 1 8= Q(t) dt (1) which is not followed by a pressure cycle delivered by the tend − tinit tinit ventilator; where tinit is the time at which the inspiratory, effort is initiated Bck Backup cycles when the pressure rise is triggered by the and tend is the time at the end of expiration (tend is also the time ventilator according to the backup frequency fbck ; Frontiers in Medical Technology | www.frontiersin.org 3 July 2021 | Volume 3 | Article 690442 Letellier et al. Patient-Ventilator Synchronization During Non-invasive Ventilation FIGURE 2 | (A) Double-triggering. (B) Auto-triggering. (C) Ineffective effort. Examples of double-triggering (DT), auto-triggering (AT), and ineffective efforts (IEs) in a patient ventilated with a pressure support ventilation (PSV) mode. Waveforms of pressure P, flow Q, thoracic belt Bthorax , and total leakage 8. Frontiers in Medical Technology | www.frontiersin.org 4 July 2021 | Volume 3 | Article 690442 Letellier et al. Patient-Ventilator Synchronization During Non-invasive Ventilation TABLE 1 | Patient characteristics. # G BMI Disorders TT IPAP EPAP fbck Device 1 F 26 Restrictive: > 48 16 4 12 Breas Vivo 40 Kyphoscoliosis + hiatal hernia 2 F 34 OHS <2 13 3 14 ResMed Lumis 150 3 F 30 Restrictive: <2 20 5 18 Breas Vivo 40 kyphoscoliosis + Obesity 4 F 37 COPD > 120 21 10 12 ResMed Lumis 150 5 M 28 COPD 24 21 5 10 ResMed Lumis 150 6 M 23 COPD 12 22 6 14 ResMed Lumis 150 7 F 22 ALS 1 13 4 16 Respironics Trilogy 8 F 78 ALS 3 20 6 12 ResMed Astral 150 9 F 22 ALS 9 17 4 14 Breas Vivo 40 Experience of a patient in mechanical ventilation is reported in months. The ventilator type is also reported. G, Gender; BMI, body mass index (kg.m−2 ); TT , training period in months; IPAP, inspiratory positive airway pressure (cmH2 O); EPAP, expiratory positive airway pressure (cmH2 O); fbck , backup frequency (cycle per min). N Ventilatory cycles are considered normal when there is 1 min of tracing. Each physician reviewed at least 110 min of absence of one of the events described above. tracing; consequently, each patient recording was reviewed by at least five expert physicians. The automatic scoring was then The asynchrony index (AI) is defined as compared to the reviewed scoring of each expert, which was NDT + NAT + NIE considered as a reference. Inter-rater comparisons were also AI = 100 ; (2) computed by selecting successively each scorer as a reference Ntot and then by computing the mean values. The number of each where Ntot is the number of ventilatory cycles (N, Bck, DT, AT, type of AE and the was computed. We computed the sensitivity IE, and discarded cycles). AI is expressed as a percentage. (Se), specificity (Sp) PPV, κ-coefficients, and agreement when All the physicians have been working with NIV for 21.5 ± 7 SyncSmart scorings were compared with the scorings of the years and, consequently, are considered as experts with NIV. experts. These quantities were also computed to compare experts All of them were asked to independently review the automatic with each other as commonly carried out for assessing inter-rater scoring of PVA produced by SyncSmart according to their agreement in sleep scoring (33–35). Student’s t-test is used with a skills and knowledge. Expert physicians were reviewing the significance level at 0.05. scoring by visual inspection of the pressure and flow as used by the SyncSmart software in addition to the belt signals as recommended by Gonzales-Bermejo et al. (6) and Longhini et al. 3. RESULTS (32), providing a reliable gold standard scoring. Since there is a long experience in sleep scoring for reviewing automatic Characteristics of the patient at enrollment and ventilator settings scoring, we chose to reproduce the corresponding methodology: are reported in Table 1. Five patients were well-established on starting from the automatic scoring, we asked each expert mechanical ventilation, and four were recently initiated. A total physicians to individually correct it, and then, we compared of 4,201 ventilatory cycles were analyzed. the corrected scoring among each other and with the automatic The nine physicians performed the analysis of the flow, scoring (33–35). The software contains a functionality that allows pressure, and belt waveforms using the SyncSmart graphical the physician to review the automatic scoring and to correct interface. The number of AEs as scored by SyncSmart and as the type of ventilatory cycle. The physician may also add or scored by the experts is reported for each patient in Table 2. remove a ventilatory cycle. Since the expert physicians have the On average, the SyncSmart software reported the same smaller belt signals, they can identify UAO which may be defined by numbers of DT (4.1 ± 3.6 vs. 6.6 ± 13.0, p = 0.31) and IE the decrease in patient flow during a pressure cycle delivered (24.1±21.7 vs. 22.0±21.5, p = 0.39) as the by experts. Conversely, by the ventilator (6). By definition, there is no effective IE the automatic scoring reported significantly more AT than the and the pressure rises are triggered by the backup frequency experts (71.7 ± 88.0 vs. 25.5 ± 35.3, p = 0.0036). The Se, Sp, PPV, fbck : The SyncSmart software thus scored these ventilatory κ-coefficient, and agreement are reported in Table 3 in which cycles as Bck cycles. Consequently, backup cycles scored as the results provided by SyncSmart software are successively obstruction by physicians are retained as Bck cycles for the compared with each expert, and where each expert is compared statistical analysis. to one another. The AI assessed by SyncSmart is not significantly Physicians were thus asked to focus on the main PVA events, different from the at assessed by the experts (18.9 ± 17.7% vs. that is, on IE, AT, and DT. Physicians were moderately under 12.8 ± 9.4%, p = 0.19) as shown in Figure 3. Se and Sp provided “time pressure” since they had approximately 2 min to review by SyncSmart for N, Bck, DT, and IE are significantly greater than Frontiers in Medical Technology | www.frontiersin.org 5 July 2021 | Volume 3 | Article 690442 Letellier et al. Patient-Ventilator Synchronization During Non-invasive Ventilation TABLE 2 | Distribution of asynchrony events (AEs) according to the automatic scoring by SyncSmart. Ntot NIE NAT NDT AI NBck NX Nmod fbreath 1 S 324 5 7 3 4.6% 237 36 – 13 E 307 9.4 14.1 1.6 8.4% 186 54 (4.0) (7.0) (1.2) (2.2) (27.8) (12.6) 2 S 375 16 6 4 6.9% 105 1 – 15 E 360 22.0 3.5 2.7 7.8% 84.7 41.0 (4.9) (2.2) (1.5) (1.5) (9.3) (15.8) 3 S 565 41 0 5 8.2% 39 14 – 22 E 541 46.5 5.2 4.0 10.0% 37.7 26.2 (4.5) (6.7) (2.2) (1.3) (5.0) (15.8) 4 S 459 18 15 3 8.1% 65 30 – 15 E 449 16.8 18.2 3.6 8.7% 53.2 – 39 (14.3) (3.1) (2.7) (3.9) (11.5) (34.0) 5 S 498 69 147 6 44.6% 0 72 – 32 E 498 71.8 44.5 1.3 23.1% 1.8 – 96 (7.4) (68.4) (1.0) (13.2) (0.6) (69.6) 6 S 430 4 63 3 16.3% 254 12 – 17 E 430 1.6 64.4 1.6 16.5% 253 12.2 (0.5) (2.2) (1.4) (0.4) (2.0) (0.5) 7 S 743 29 221 3 34.1% 0 9 – 36 E 743 12.8 24.8 1.3 7.3% 1.8 151 (12.3) (17.0) (0.5) (2.2) (1.7) (95) 8 S 497 33 183 13 46.1% 251 17 – 24 E 492 24.6 88.4 1.6 31.8% 24.4 206 (25.0) (47.7) (7.2) (12.2) (25.3) (54) 9 S 310 3 0 0 1.6% 2 0 – 12 E 312 1.6 1.8 1.6 1.7% 9.4 12.2 (1.1) (1.3) (3.9) (1.3) (11.1) (13.2) The mean numbers of AE according to the expert physicians are also reported with the SD between parenthesis. The breathing frequency fbreath (in breath per min) as evaluated by the SyncSmart software is also reported. S, SyncSmart; E, experts; Ntot , total number of detected ventilatory cycles in the automatic scoring (s) and the mean number of ventilatory cycles once the scoring is reviewed by the experts (E); IE, ineffective effort; AT, auto-triggering; DT, double-triggering; AI, asynchrony index; Bck, backup cycles; X, discarded from the analysis due to too high leakage; mod, modified by the expert; fbreath , breathing frequency. those provided by experts when compared to one another. The results show that the events detected by automatic scoring PPV for IE is significantly greater for the automatic scoring than were well recognized by expert physicians but scored as a for experts. The inter-scorer variability for DT and AT is nearly different AE. This explains why the inter-rater variability is twice that of IE. AT is clearly the event that leads to the most greater than the disagreement with the automatic scoring. This important discrepancies between the automatic scoring and the partly explains the large inter-rater variability for determining experts (Table 2), particularly for patients 5, 7, and 8 (Table 4). the type of PVA although the inter-rater variability in the AI is rather low (the variance is equal to 9.9%). For six of the nine patients, there is a slight trend in SyncSmart to miss some 4. DISCUSSION AEs (about 10% as detailed in Table 2) as exemplified with 4.1. Comparison With Other Studies the AT shown in Figure 1 (t = 738 s) that is considered This study confirms that the inter-rater variability between as a normal cycle. It is also known that IE may be missed physicians to detect PVA events during NIV by visual inspection from a simple waveform analysis (36). Conversely, in three is pretty large for the three main AEs (DT, AT, and IE). patients (5, 7, and 8), there was an over detection of AT. Over- This was already shown by Longhini et al. (32). Most often detection of AT occurred in two patients with ALS and one physicians recognize correctly the presence of an AE but they patient with COPD whose breathing frequencies were greater had difficulties to discriminate them; for instance, it is rather than 24 breath per min (see Table 2). Indeed, the other six difficult to distinguish DT from a combination of AT-N. Our patients had a breathing frequency that is lower and AT as Frontiers in Medical Technology | www.frontiersin.org 6 July 2021 | Volume 3 | Article 690442 Letellier et al. Patient-Ventilator Synchronization During Non-invasive Ventilation TABLE 3 | Sensitivity (Se), specificity (Sp), positive predictive value (PPV), κ-coefficient, and agreement for the automatic scoring by SyncSmart successively compared to each expert and for each expert successively compared to one another. AE Sensitivity Specificity PPV κ-coeff. Agreement Sync. Inter Sync. Inter Sync. Inter Sync. Inter Sync. Inter N 0.92 0.90 0.84 0.78 0.91 0.90 0.37 0.33 0.87 0.83 (0.10) (0.10) (0.21) (0.25) (0.10) (0.10) (0.13) (0.13) (0.14) (0.13) p = 0.084 p = 0.23 p = 0.23 p = 0.13 p = 0.17 Bck 0.63 0.60 0.98 0.96 0.64 0.62 0.30 0.28 0.74 0.72 (0.45) (0.40) (0.03) (0.05) (0.45) (0.41) (0.22) (0.19) (0.34) (0.26) p = 0.20 p = 0.0013 p = 0.15 p = 0.36 p = 0.10 DT 0.65 0.55 0.99 0.99 0.62 0.56 0.29 0.26 0.75 0.72 (0.40) (0.41) (0.02) (0.03) (0.37) (0.41) (0.18) (0.19) (0.27) (0.26) p = 0.082 p = 0.0029 p = 0.26 p = 0.18 p = 0.47 AT 0.71 0.60 0.95 0.97 0.62 0.60 0.29 0.26 0.77 0.75 (0.38) (0.36) (0.10) (0.08) (0.40) (0.36) (0.19) (0.16) (0.24) (0.18) p = 0.031 p < 10−5 p = 0.42 p = 0.0044 p = 0.35 IE 0.90 0.81 1.00 1.00 0.92 0.81 0.45 0.40 0.94 0.88 (0.21) (0.16) (0.01) (0.01) (0.20) (0.17) (0.10) (0.18) (0.15) (0.21) p = 0.031 p = 0.046 p = 0.058 p = 0.0056 p = 0.030 SDs are reported in parenthesis. PPV, positive predicitive value; Sync., SyncSmart vs. experts; Inter, an expert vs. another one; AE, asynchrony event; N, normal; Bck, backup cycle; DT, double-triggering; AT, auto-triggering; IE, ineffective effort. Values in bold fonts are those for which SyncSmart has better results than the inter-rater variability. FIGURE 3 | Asynchrony index (AI) computed from the SyncSmart scoring and the mean AI (with the standard deviation) computed from the expert scorings for each patient and for all patients, respectively. correctly detected. Even with this overdetection, the main result To our knowledge, this pilot study is the first attempt to the of this study is that automatic scoring is within the inter- assess performance of software to automatically detect DT, AT, rater variability and provides, in general, a lower bound for and IE from the pressure and flow waveforms. In most of the the AI. previous studies, only IEs were investigated (24, 25, 27, 31). These Frontiers in Medical Technology | www.frontiersin.org 7 July 2021 | Volume 3 | Article 690442 Letellier et al. Patient-Ventilator Synchronization During Non-invasive Ventilation TABLE 4 | Sensitivity (Se), specitify (Sp), PPV, κ-coefficient, and agreement between (i) the experts and the automatic scoring by SyncSmart and (ii) between experts, successively computed for each patient. Sensitivity Specificity PPV κ-coeff. Agreement Sync. Inter Sync. Inter Sync. Inter Sync. Inter Sync. Inter P1 0.81 0.72 0.98 0.97 0.84 0.72 0.38 0.33 0.88 0.82 (0.28) (0.31) (0.03) (0.05) (0.25) (0.32) (0.12) (0.14) (0.12) (0.14) P2 0.91 0.78 0.98 0.97 0.85 0.79 0.42 0.36 0.91 0.84 (0.17) (0.28) (0.04) (0.05) (0.25) (0.28) (0.10) (0.13) (0.17) (0.19) P3 0.73 0.74 0.98 0.97 0.74 0.74 0.36 0.35 0.83 0.84 (0.41) (0.38) (0.05) (0.32) (0.41) (0.38) (0.20) (0.18) (0.27) (0.20) P4 0.82 0.79 0.96 0.95 0.90 0.82 0.40 0.36 0.90 0.86 (0.17) (0.18) (0.07) (0.09) (0.12) (0.17) (0.07) (0.07) (0.07) (0.07) P5 0.66 0.55 0.94 0.93 0.55 0.53 0.25 0.23 0.69 0.70 (0.42) (0.40) (0.10) (0.14) (0.42) (0.39) (0.19) (0.18) (0.30) (0.22) P6 0.94 0.71 0.99 0.92 0.97 0.70 0.47 0.30 0.97 0.77 (0.10) (0.38) (0.02) (0.18) (0.05) (0.38) (0.04) (0.19) (0.04) (0.25) P7 0.69 0.57 0.90 0.88 0.53 0.57 0.23 0.21 0.67 0.67 (0.43) (0.42) (0.17) (0.22) (0.43) (0.41) (0.20) (0.31) (0.30) (0.25) P8 0.59 0.55 0.90 0.91 0.56 0.55 0.22 0.23 0.67 0.69 (0.43) (0.35) (0.12) (0.09) (0.35) (0.35) (0.18) (0.16) (0.27) (0.24) P9 0.60 0.46 0.90 0.85 0.62 0.46 0.25 0.15 0.67 0.56 (0.49) (0.48) (0.28) (0.31) (0.48) (0.48) (0.24) (0.21) (0.37) (0.10) P 0.76 0.69 0.95 0.94 0.74 0.69 0.34 0.30 0.81 0.78 (0.35) (0.36) (0.12) (0.14) (0.36) 0.36) (0.18) (0.17) (0.25) (0.22) SDs are reported in parenthesis. Sync., SyncSmart; Inter, inter-scorer; PPV, positive predictive value. different algorithms were detecting IEs with Se and Sp greater AT-N (37). It seems that when the breathing frequency is too than 0.90 as observed with SyncSmart. One study investigated high, such feature is no longer reliable: this could explain the over DT in 67 patients but there is no information on the validation detection of AT observed in our study. of the algorithm (29). Two studies investigated AE without Some studies show that more than 30% of patients in NIV distinguishing them. One reported the AI with a Se and a Sp have more than 10% of AEs (11, 38, 39). In our cohort, 44% of at about 0.90 (28). The second one investigated the asynchrony patients had AI > 30%. Omitting the over detected AT, 349 AEs between patient and ventilator (30). The latter study was, in fact, were actually detected. This is not sufficient for robust results and devoted to neurally adjusted non-invasive ventilator (NAVA), investigating a larger cohort is required. Nevertheless, previous and the algorithm uses the diaphragm electrical activity, pressure, study showed that the detection of AEs is inversely related to their and flow waveforms. So, its purpose was different from our pilot prevalence, indicating that the ability to recognize PVA is reduced study. None of these studies, whose main characteristics are when their occurrence increases (20, 32). For these reasons, reported in Table 5, investigated AT, which is, as revealed in this asynchrony events were not too numerous during our ventilation study, the most difficult AE to determine, even when the belt sessions for optimizing the ability of physicians to correctly signal is provided. recognize the asynchrony events. The Se, PPV, κ-coefficients, and In the study of Mulqueeny et al. (31), DTs were defined as agreement are the lowest for the three patients, for which AT was two pressure cycles separated by less than 500 ms. This definition over detected and, consequently, with AI> 34%. Clearly, this does not allow to distinguish a DT from a combination of AT- drawback of SyncSmart should be corrected. As physicians are N. Distinguishing these two types of AEs is indeed sometimes commonly doing for sleep scoring (40), in our study, experts start very difficult, thus explaining the low Se found for these two also from the automatic scoring. types of AE, from the SyncSmart software as well as from the experts. In DT, the first oscillation of the flow has an amplitude 4.2. Limitations of the Study that is larger than the second one (7, 37). During AT, the patient There are possible limitations in this study. The first being that is still expiring: the inspired volume should be therefore smaller the experts were not blind to automatic scores and had, in fact, to than for that normal cycle, thus inducing lower amplitude flow reviewed them. This may create a bias by reducing the inter-rater waveforms (7, 37). This is a possible way to discriminate DT from variability compared to scoring from raw tracings. Nevertheless, Frontiers in Medical Technology | www.frontiersin.org 8 July 2021 | Volume 3 | Article 690442 Letellier et al. Patient-Ventilator Synchronization During Non-invasive Ventilation TABLE 5 | Brief overview of the studies devoted to automatic scoring of AE in mechanical ventilation. Reference Events Sensitivity Specificity Npatient Nbreath Mulqueenyl et al. (31) IE 0.59 0.99 ICU-NIV 23 5,624 Chen et al. (24) IE 0.93 0.93 ICU 14 5,899 Cuvelier et al. (25) IE 0.95 1.00 NIV 9 2,127 Blanch et al. (27) IE 0.92 0.80 ICU 8 1,024 Chiew et al. (28) AI 0.90 0.88 ICU 11 5,701 de Haro et al. (29) DT — — ICU 67 9,694,573 Dorduin et al. (30) AE — — NIV 11 — The type of AE, sensitivity (Se), and specificity (Sp), when provided, the ventilation mode [intensive care unit (ICU) or non-invasive ventilation (NIV)], the number of patients analyzed, and the number of breaths are reported for each of them. IE, ineffective effort; AI, asynchrony index; DT, double-triggering; AE, asynchrony event. even with this bias, the statistics (κ-coefficients and agreement as of all AEs (32). As pointed out by Longhini et al., “because of reported in Table 3) are still in favor of automatic scoring since the poor performance of visual inspection of ventilator waveforms, the expert scoring better matches with the automatic scoring than algorithms able to recognize patient-ventilator asynchrony might with each other. Indeed, the modified scores very often differ indeed represent an important advance for the management of from one expert to the other. With blind scoring, the results patient underlying NIV” (32). SyncSmart overcomes the lack of becomes intractable as we observed in a preliminary study (non- a tool for helping physicians to identify the source for poor published), leading us to adopt the procedure commonly used to mechanical coupling between patient and ventilator. assess the performance of automatic sleep scoring. As the aim of the present pilot study is to evaluate the SyncSmart software as 5. CONCLUSION a proof of concept, further study is warranted to evaluate it for longer NIV sessions and for a larger cohort of patients. To our knowledge, this is the first study to show that it is There is not yet a standardization for coding AEs for patients possible to automatically detect AEs from solely pressure and using NIV. There are few contributions in that direction which flow waveforms and that the results are within the inter-rater are based on visual inspection (6, 7, 41) and, consequently, which variability. This pilot study shows that such a procedure can be use qualitative arguments that may sometimes lead to subjective used to validate automatic scoring of AEs. Most of the events interpretation. Another problem is related to obstructions that from more than 4,200 ventilatory cycles were well-detected by the are not considered in the present study. Two of the patients SyncSmart software. When the breathing frequency is lower than have a noticeable number of obstructions. These events were 24 breath per min, SyncSmart returns an AI slightly less than the merged with backup cycles. In this study, as it is considered by the at assessed by expert physicians. A validation with a larger cohort SyncSmart software, the ventilatory cycles are discarded (< 5% of is required to evaluate whether the AI provided by automatic the number of breaths) from the analysis when leaks are too large: scoring could be considered as a lower bound. This avoids inappropriate tracings that could prevent physicians from correctly detecting PVA. The presence of high leakage is to be considered as a primary event during NIV, which needs to be DATA AVAILABILITY STATEMENT corrected before any other action, adjustment of treatment, or The datasets analyzed for this study can be found in the web page settings that can be considered or recommended. http://www.atomosyd.net/spip.php?article222 It should be noted that, in the daily clinical practice, physicians use “visual” signs as chest movement or signs of discomfort and possibly feedback provided by the patient. In addition to ETHICS STATEMENT the waveforms (pressure, flow, and belt signal), the SyncSmart software provides the patient and ventilator frequencies, which The studies involving human participants were reviewed are useful to identify PVA. These last two limitations are assumed and approved by Corporacio Parc Tauli (CIR2010/015). The to be a source for increasing the inter-rater variability (32). Since patients/participants provided their written informed consent to “minor” AEs as advanced and delayed cycling (pressure release) participate in this study. (42, 43) are not detected by the SyncSmart software, they were not considered in this study. As they are suspected to be important AUTHOR CONTRIBUTIONS for inducing major AEs, detecting them would lead to a better understanding of the causes of the main events. All the authors contributed to the literature search and to the When the AI is typically greater than 10%, an adjustment study design. ML collected the data measured in the patients of the ventilator settings is required (9, 11, 20). Since in the Pneumology Unit Care of the Corporacio Parc Tauli these adjustments are event type dependent, it is relevant to (Sabadell, Spain). All the authors, but CL, reviewed the automatic discriminate DT, AT, and IE. DT occurs to restrictive patient scoring provided by the software. CL processed the data provided with a low breathing frequency (37). AT may contribute for 40% by the software and by the manual review of the physician. Frontiers in Medical Technology | www.frontiersin.org 9 July 2021 | Volume 3 | Article 690442 Letellier et al. Patient-Ventilator Synchronization During Non-invasive Ventilation The analysis of the data was performed by all the authors. of data, the writing of this article, or the decision to submit it The manuscript was prepared by CL and reviewed by all the for publication. other authors. ACKNOWLEDGMENTS FUNDING This study would not have been possible without the great This study received funding from Breas Medical. The funder was assistance of Sally Cozens, Carl van Loey, Didier Menguy, and not involved in study design, collection, analysis, interpretation Sebastian Mommers. REFERENCES different levels of pressure support ventilation. Chest. (2004) 126:851–9. doi: 10.1378/chest.126.3.851 1. Adler D, Perrig S, Takahashi H, Espa F, Rodenstein D, Pépin JL, et al. 15. Dhand R. Ventilator graphics and respiratory mechanics in the patient with Polysomnography in stable COPD under non-invasive ventilation to reduce obstructive lung disease. Respir Care. (2005) 50:246–61. patient-ventilator asynchrony and morning breathlessness. Sleep Breath. 16. Georgopoulos D, Prinianakis G, Kondili E. Bedside waveforms interpretation (2012) 16:1081–90. doi: 10.1007/s11325-011-0605-y as a tool to identify patient-ventilator asynchronies. Intens Care Med. (2006) 2. Gonzalez-Bermejo J, Morelot-Panzini C, Arnol N, Meininger V, Kraoua 32:34–47. doi: 10.1007/s00134-005-2828-5 S, Salachas F, et al. Prognostic value of efficiently correcting nocturnal 17. Camporota L. Bedside detection of patient-ventilation asynchrony. Signa desaturations after one month of non-invasive ventilation in amyotrophic Vitae. (2016) 11(Suppl. 2):31–4. doi: 10.22514/SV112.062016.5 lateral sclerosis: a retrospective monocentre observational cohort study. 18. Blanch L, Arnal JM, Mojoli F. Optimising patient-ventilator synchronisation. Amyotroph Lateral Scler Frontotemporal Degener. (2013) 14:373–9. ICU Manage Pract. (2018) 18:supplements. doi: 10.3109/21678421.2013.776086 19. Ciorba C, Gonzalez-Bermejo J, Salva MAQ, Annane D, Orlikowski D, Lofaso 3. Vrijsen B, Testelmans D, Belge C, Robberecht W, Damme PV, F, et al. Flow and airway pressure analysis for detecting ineffective effort during Buyse B. Non-invasive ventilation in amyotrophic lateral sclerosis. mechanical ventilation in neuromuscular patients. Chronic Respir Dis. (2018) Amyotroph Lateral Scler Frontotemporal Degener. (2013) 14:85–95. 16:1–9. doi: 10.1177/1479972318790267 doi: 10.3109/21678421.2012.745568 20. Colombo D, Cammarota G, Alemani M, Carenzo L, Barra FL, Vaschetto 4. Georges M, Attali V, Golmard JL, Morélot-Panzini C, Crevier-Buchman L, R, et al. Efficacy of ventilator waveforms observation in detecting Collet JM, et al. Reduced survival in patients with ALS with upper airway patient-ventilator asynchrony. Crit Care Med. (2011) 39:2452–7. obstructive events on non-invasive ventilation. J Neurol Neurosurg Psychiatry. doi: 10.1097/CCM.0b013e318225753c (2016) 87:1045–50. doi: 10.1136/jnnp-2015-312606 21. Ramirez II, Arellano DH, Adasme RS, Landeros JM, Salinas FA, Vargas 5. Rabec C, Rodenstein D, Léger P, Rouault S, Perrin C, Gonzalez-Bermejo J, AG, et al. Ability of ICU health-care professionals to identify patient- et al. Ventilator modes and settings during non-invasive ventilation: effects ventilator asynchrony using waveform analysis. Respir Care. (2017) 62:144–9. on respiratory events and implications for their identification. Thorax. (2011) doi: 10.4187/respcare.04750 66:170–8. doi: 10.1136/thx.2010.142661 22. Enrico B, Cristian F, Stefano B, Luigi P. Patient-ventilator asynchronies: types, 6. Gonzalez-Bermejo J, Perrin C, Janssens JP, Pepin JL, Mroue G, Léger P, outcomes and nursing detection skills. Acta Bio Med Atenei Parmensis. (2018) et al. Proposal for a systematic analysis of polygraphy or polysomnography 89:6–18. doi: 10.23750/abm.v89i7-S.7737 for identifying and scoring abnormal events occurring during non-invasive 23. Luján M, Sogo A, Monsó E. Home mechanical ventilation monitoring ventilation. Thorax. (2012) 67:546–52. doi: 10.1136/thx.2010.142653 software: measure more or measure better? Arch Bronconeumol. (2012) 7. Gonzalez-Bermejo J, Janssens JP, Rabec C, Perrin C, Lofaso F, Langevin B, et al. 48:141–86. doi: 10.1016/j.arbr.2011.10.008 Framework for patient-ventilator asynchrony during long-term non-invasive 24. Chen CW, Lin WC, Hsu CH, Cheng KS, Lo CS. Detecting ineffective ventilation. Thorax. (2019) 74:715–7. doi: 10.1136/thoraxjnl-2018-213022 triggering in the expiratory phase in mechanically ventilated patients based on 8. Chao DC, Scheinhorn DJ, Stearn-Hassenpflug M. Patient-ventilator trigger airway flow and pressure deflection: feasibility of using a computer algorithm. asynchrony in prolonged mechanical ventilation. Chest. (1997) 112:1592–9. Crit Care Med. (2008) 36:455–61. doi: 10.1097/01.CCM.0000299734.34469.D9 doi: 10.1378/chest.112.6.1592 25. Cuvelier A, Achour L, Rabarimanantsoa H, Letellier C, Muir JF, Fauroux 9. Thille AW, Rodriguez P, Cabello B, Lellouche F, Brochard L. Patient-ventilator B. A noninvasive method to identify ineffective triggering in patients with asynchrony during assisted mechanical ventilation. Intens Care Med. (2006) noninvasive pressure support ventilation. Respiration. (2010) 80:198–206. 32:1515–22. doi: 10.1007/s00134-006-0301-8 doi: 10.1159/000264606 10. Guo YF, Sforza E, Janssens JP. Respiratory patterns during sleep in 26. Gutierrez G, Ballarino GJ, Turkan H, Abril J, Cruz LDL, Edsall C, obesity-hypoventilation patients treated with nocturnal pressure support: a et al. Automatic detection of patient-ventilator asynchrony by spectral preliminary report. Chest. (2007) 131:1090–9. doi: 10.1378/chest.06-1705 analysis of airway flow. Crit Care. (2011) 15:R167. doi: 10.1186/cc 11. Vignaux L, Vargas F, Roeseler J, Tassaux D, Thille AW, Kossowsky MP, 10309 et al. Patient-ventilator asynchrony during non-invasive ventilation for acute 27. Blanch L, Sales B, Montanya J, Lucangelo U, Garcia-Esquirol O, Villagra A, respiratory failure: a multicenter study. Intens Care Med. (2009) 35:840–6. et al. Validation of the Better Care R system to detect ineffective efforts during doi: 10.1007/s00134-009-1416-5 expiration in mechanically ventilated patients: a pilot study. Intens Care Med. 12. Ramsay M, Mandal S, Suh ES, Steier J, Douiri A, Polkey MI, et al. Parasternal (2012) 38:772–80. doi: 10.1007/s00134-012-2493-4 electromyography to determine the relationship between patient-ventilator 28. Chiew YS, Pretty CG, Beatson A, Glassenbury D, Major V, Corbett S, et al. asynchrony and nocturnal gas exchange during home mechanical ventilation Automated logging of inspiratory and expiratory non-synchronized breathing set up. Thorax. (2015) 70:946–52. doi: 10.1136/thoraxjnl-2015-206944 (ALIEN) for mechanical ventilation. In: 2015 37th Annual International 13. Fanfulla F, Taurino AE, Lupo ND, Trentin R, D’Ambrosio C, Nava S. Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). Effect of sleep on patient/ventilator asynchrony in patients undergoing Milan (2015). p. 5315–8. chronic non-invasive mechanical ventilation. Respir Med. (2007) 101:1702–7. 29. de Haro C, López-Aguilar J, Magrans R, Montanya J, Fernández-Gonzalo S, doi: 10.1016/j.rmed.2007.02.026 Turon M, et al. Double cycling during mechanical ventilation: frequency, 14. Vitacca M, Bianchi L, Zanotti E, Vianello A, Barbano L, Porta R, mechanisms, and physiologic implications. Crit Care Med. (2018) 46:1385–92. et al. Assessment of physiologic variables and subjective comfort under doi: 10.1097/CCM.0000000000003256 Frontiers in Medical Technology | www.frontiersin.org 10 July 2021 | Volume 3 | Article 690442 Letellier et al. Patient-Ventilator Synchronization During Non-invasive Ventilation 30. Doorduin J, Sinderby CA, Beck J, van der Hoeven JG, Heunks LM. Automated capnography, built-in ventilator software and autonomic markers of sleep patient-ventilator interaction analysis during neurally adjusted non-invasive fragmentation. Thorax. (2011) 66:438–45. doi: 10.1016/j.rmr.2013.08.003 ventilation and pressure support ventilation in chronic obstructive pulmonary 42. Calderini E, Confalonieri M, Puccio PG, Francavilla N, Stella L, Gregoretti disease. Crit Care. (2014) 18:550. doi: 10.1186/cc13489 C. Patient-ventilator asynchrony during noninvasive ventilation: the 31. Mulqueeny Q, Ceriana P, Carlucci A, Fanfulla F, Delmastro M, Nava role of expiratory trigger. Intensive Care Med. (1999) 25:662–7. S. Automatic detection of ineffective triggering and double triggering doi: 10.1007/s001340050927 during mechanical ventilation. Intens Care Med. (2007) 33:2014–8. 43. Gentile MA. Cycling of the mechanical ventilator breath. Respir Care. (2011) doi: 10.1007/s00134-007-0767-z 56:52–60. doi: 10.4187/respcare.01088 32. Longhini F, Colombo D, Pisani L, Idone F, Chun P, Doorduin J, et al. Efficacy of ventilator waveform observation for detection of patient—ventilator Conflict of Interest: ML reports personal fees for lectures from Philips and asynchrony during NIV: a multicentre study. ERJ Open Res. (2017) 3:00075. Resmed. JA is part time employee at Hamilton Medical. He has received advisory doi: 10.1183/23120541.00075-2017 honorarium from Breas Medical and ResMed. He has received honorarium for 33. Norman RG, Pal I, Stewart C, Walsleben JA, Rapoport DM. Interobserver lecturing from Philips. AC has received fees for lecturing and roundtable meetings agreement among sleep scorers from different centers in a large dataset. Sleep. from Philips, Breas and Resmed. MC has received advisory honorarium from Breas (2000) 23:901–8. doi: 10.1093/sleep/23.7.1e Medical and ResMed. She has received honorarium for lecturing from, Philips, 34. Berthomier C, Drouot X, Herman-Stoïca M, Berthomier P, Prado J, Bokar- ResMed, Breas and MPR. CL has received fees for consulting from Breas Medical. Thire D, et al. Automatic analysis of single-channel sleep EEG: validation JS reports grants and personal fees for lectures from Heinen und Löwenstein in healthy individuals. Sleep. (2007) 30:1587–95. doi: 10.1093/sleep/30.11. and VitalAire, grants, personal fees for lectures and non-financial support for 1587 meeting attendance from Vivisol GmbH, grants from Weinmann Deutschland, 35. Stepnowsky C, Levendowski D, Popovic D, Ayappad I, Rapoport DM. Scoring personal fees for consultancy/advisory board work from Breas Medical, during accuracy of automated sleep staging from a bipolar electroocular recording the conduct of the study; personal fees for consultancy and lectures, and non- compared to manual scoring by multiple raters. Sleep Med. (2013) 14:1199– financial support for meeting attendance from Boehringer Ingelheim Pharma, 207. doi: 10.1016/j.sleep.2013.04.022 personal fees for consultancy and lectures from SenTec AG, Keller Medical 36. Younes M, Brochard L, Grasso S, Kun J, Mancebo J, Ranieri M, et al. A GmbH, Linde Deutschland and Santis GmbH, outside the submitted work. NH method for monitoring and improving patient: ventilator interaction. Intens has received unrestricted research grants from Philips, Philips-Respironics, B&D Care Med. (2007) 33:1337–46. doi: 10.1007/s00134-007-0681-4 Electromedical, Breas and Fisher-Paykel. He has has received fees for lecturing 37. Fresnel E. Étude comparative des performances des ventilateurs de domicile and roundtable meetings from Philips, Philips-Respironics, B&D Electromedical, et analyse des interactions patient-ventilateur en ventilation non invasive. Breas and Fisher-Paykel. SN has received fees for lecturing and roundtable Normandie University. Rouen, France (2016). meetings from Philips, Breas and Resmed. He is on the advisory board for Breas 38. Vignaux L, Tassaux D, Carteaux G, Roeseler J, Piquilloud L, Brochard L, and Philips. et al. Performance of noninvasive ventilation algorithms on ICU ventilators during pressure support: a clinical study. Intens Care Med. (2010) 36:2053–9. The remaining authors declare that the research was conducted in the absence of doi: 10.1007/s00134-010-1994-2 any commercial or financial relationships that could be construed as a potential 39. Carlucci A, Pisani L, Ceriana P, Malovini A, Nava S. Patient-ventilator conflict of interest. asynchronies: may the respiratory mechanics play a role? Crit Care. (2013) 17:R54. doi: 10.1186/cc12580 Copyright © 2021 Letellier, Lujan, Arnal, Carlucci, Chatwin, Ergan, Kampelmacher, 40. Rotem AY, Sperber AD, Krugliak P, Freidman B, Tal A, Tarasiuk Storre, Hart, Gonzalez-Bermejo and Nava. This is an open-access article distributed A. Polysomnographic and actigraphic evidence of sleep fragmentation under the terms of the Creative Commons Attribution License (CC BY). The use, in patients with irritable bowel syndrome. Sleep. (2003) 26:747–52. distribution or reproduction in other forums is permitted, provided the original doi: 10.1093/sleep/26.6.747 author(s) and the copyright owner(s) are credited and that the original publication 41. Janssens JP, Borel JC, Pépin JL, Group S. Nocturnal monitoring of home non- in this journal is cited, in accordance with accepted academic practice. No use, invasive ventilation: the contribution of simple tools such as pulse oximetry, distribution or reproduction is permitted which does not comply with these terms. Frontiers in Medical Technology | www.frontiersin.org 11 July 2021 | Volume 3 | Article 690442