Machine Learning Algorithms to Detect Patient-Ventilator Asynchrony. A Feasibility Study.

doi:10.21203/rs.3.rs-4907650/v1

Download PDF

Research Article

Machine Learning Algorithms to Detect Patient-Ventilator Asynchrony. A Feasibility Study.

https://doi.org/10.21203/rs.3.rs-4907650/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background: Adequate ventilatory support requires frequent assessment of patient-ventilator interactions. It is desirable, therefore, to develop a reliable, automated method for this task. This study evaluates the feasibility of developing machine-learning algorithms to emulate how experienced clinicians evaluate normal and abnormal breathing patterns, including patient-ventilator asynchrony.

Methods: We enrolled 44 adult patients within 24 hours of initiating invasive mechanical ventilation. Airway flow and pressure signals were acquired directly from the ventilator and stored as sequential 2.2-minute epochs for waveform classification. Experienced clinicians visually classified 50,712 epochs, encompassing approximately 2.6 million breathing cycles. Nineteen clinical variables were used to train four Random Forest algorithms to: 1) detect asynchronous breathing, 2) identify asynchrony type, 3) grade signal disruption, and 4) identify dynamic hyperinflation. Algorithm accuracy was assessed by the percentage of correctly identified epochs, while clinical reliability was evaluated by comparing the algorithms’ predictions to those of clinicians with varying experience in asynchrony classification.

Results: The algorithm detected asynchronous breathing with 91% accuracy. Accuracies for asynchrony classification, severity grading, and dynamic hyperinflation were 82%, 87%, and 93%, respectively. Algorithm classifications aligned more closely with expert clinicians (kappa = 0.46, and 0.59) than non-experts (kappa = 0.25, and 0.38; p < 0.05). Greater time asynchronous was associated with increased 28-day mortality (p = 0.015).

Conclusions: Machine-learning algorithms may be trained to emulate experienced clinicians in evaluating breathing during mechanical ventilation. Larger databases and advancements in artificial intelligence may lead to powerful algorithms capable of establishing associations between airway signals and successful ventilatory support.

Patient ventilator asynchrony

ICU monitoring

artificial intelligence

respiratory rate variability

Artificial intelligence machine learning algorithms can be effectively trained to replicate the expertise of experienced clinicians in evaluating breathing patterns during mechanical ventilation, including the detection of patient-ventilator asynchrony.

140-character Tweet: AI algorithms can emulate expert clinicians in assessing breathing patterns during mechanical ventilation, including detecting asynchrony.

Mechanically ventilated patients often develop asynchronous breathing, defined as a mismatch between the patient’s neural drive and the set rhythm of the ventilator [1]. The incidence of patient-ventilator asynchrony (PVA) ranges from 25–100%, and it has been associated with prolonged ventilatory support [2] and increased mortality [3]. Despite its clinical significance, PVA identification can be challenging [4], as detection relies on the clinician's expertise and the intermittent nature of ventilator waveform monitoring. Therefore, it is desirable to develop a method to detect and classify PVA in ventilated patients. Ideally, this method should be autonomous and non-invasive.

Advances in computer software and data storage capacity has made it possible to apply the principles of artificial intelligence (AI), in particular the subset known as machine learning, to the task of formulating mathematical algorithms capable of analyzing airway waveforms to detect PVA on a continuous basis.

We present a study on the feasibility of developing machine learning algorithms to detect asynchronous breathing patterns emulating the approach of an experienced clinician reviewing ventilator airway signals. These algorithms were developed by applying machine learning techniques to a large database of airway signals compiled from adult mechanically ventilated patients.

This was a prospective, observational study performed at The George Washington University Hospital and approved by The George Washington University IRB (# 101228). Informed consent to record and analyze deidentified data was obtained from all patients or surrogates prior to data collection. We sequentially enrolled adult patients within 24 hours from the initiation of invasive mechanical ventilation. All patients were intubated via the nasotracheal or orotracheal route and received ventilatory support using Servo_i or Servo_s ventilators (Getinge, Solna, Sweden) with various modes of ventilation. Treatment decisions were independent of the study. Patients were monitored for a maximum of four days or until weaned from the ventilator. Patients treated with serial or continuous use of paralytic agents were excluded from enrollment.

We recorded days on mechanical ventilation, ICU and hospital length of stay, and all-cause mortality at 28 days. Demographic data, SAPS II scores [5] and hourly use of intravenous propofol were extracted from the patient’s chart and recorded in physical notebooks that were destroyed after transferring nonidentifiable details to the database using a study number.

We continuously sampled airway pressure and flow signals at 31.25 Hz using a Raspberry Pi 3B microprocessor connected to the ventilator's RS232 data port via a null DB9 cable. The data were stored as sequential 2.2-minute epochs, each containing 4,096 samples per signal. Upon completing each epoch’s signal sampling, the microprocessor collected other ventilator-related information, including the set mode of ventilation and fraction of inspired oxygen (F_IO₂), and calculated breathing data such as peak pressure, respiratory rate, and expired tidal volume.

Data were transferred from the microprocessor to a digital computer for analysis once the patient completed the study. Software was written to display data pertaining to any given epoch, showing the airway signals, patient demographics, ventilator settings, and breathing data (Fig. 1E of Supplementary Material Section). Selectable graphic radio buttons allowed for epoch classification by clinicians with expertise in mechanical ventilation who visually evaluated each epoch’s breathing pattern.

Epochs exhibiting fewer than four asynchronies were classified as either "synchronous" or "exhibiting timing variability." Synchronous epochs were characterized by near-perfect regularity in breath timing, while timing-variability epochs displayed observable differences in time between breaths. Epochs with four or more identifiable asynchronies were designated as asynchronous and further subdivided into specific categories: double triggering, auto triggering, ineffective effort, and cycling asynchrony (either delayed or premature cycling). Epochs demonstrating two types of asynchronies were classified according to the predominant type, whereas epochs with three or more asynchrony types were categorized as complex. Figure 2E (Supplementary Material Section) illustrates examples of breathing patterns with corresponding asynchrony classifications.

Epochs were also evaluated for severity based on the extent of airway signal disruption, categorized as mild, moderate, or severe. This assessment was designed to mimic clinical decision-making and was thus subjective, relying on the clinical judgment of the classifier. Epochs where the end-expiratory flow did not return to the zero baseline in four or more breaths were classified as potentially exhibiting dynamic hyperinflation [6].

Given the large number of epochs generated during the study, each epoch was classified only once by one assessor. All information related to an epoch was recorded as a row, or instance, in the database.

Machine learning algorithms.

Following best practices for machine learning algorithm development [7], we randomly selected 70% of the database instances to create the training dataset for algorithm development. The remaining 30% of the database served as the testing dataset to evaluate the algorithms' performance.

Four Random Forest algorithms were developed (Fig. 1). A binary classifier trained to detect the presence or absence of asynchronous breathing (Model 1). A multiclass classifier trained to recognize signal morphology, including synchrony, variable breathing and the four types of asynchronies previously described (Model 2). A multiclass classifier trained to predict epoch severity (Model 3). Lastly, a binary classifier trained to detect dynamic hyperinflation (Model 4).

Algorithm assessment metrics included accuracy, defined as the fraction of correct predictions made by the algorithms using the testing dataset; precision (positive predictive value, PPV); sensitivity; specificity; and the F1-score, defined as the harmonic mean of precision and sensitivity.

We also evaluated the clinical reliability of Models 2 and 3 by comparing the algorithms predictions to those made by two groups of clinicians with different levels of expertise in identifying asynchronous breathing. The "Non-expert" group comprised junior clinicians with limited experience in identifying asynchrony (n = 4), while the "Expert" group included attending physicians with extensive proficiency in asynchrony identification (n = 3). Members of both groups were tasked with classifying identical datasets consisting of 1,000 epochs randomly selected from the database. We hypothesized that the classifications made by the Expert group would more closely align with the predictions of the Random Forest algorithms.

Cohen’s kappa statistic [8] was used to compare the agreement between the predictions made by each reader and those of the models. The magnitude of agreement was categorized a priori [9] as poor (0 to 0.19), fair (0.20 to 0.39), moderate (0.40 to 0.59), and substantial (0.60 to 0.79).

The Mann-Whitney and Kruskal-Wallis tests were used to assess differences in distributions between groups without assuming a normal distribution. Unless specified otherwise, data are shown as median [IQR]. Additional information on epoch scoring and machine learning algorithm development may be found in the online data supplement.

We enrolled 44 patients aged 62.5 [57.0, 70.8] years. All participants were medical patients, the majority were African-American, and half were women (Table 1E of Supplementary Material Section). Most patients were intubated due to pulmonary conditions (52.3%), including COPD, pneumonia, and ARDS. Patients remained on ventilatory support for a median of 3 [2, 5] days, with ICU and hospital stays of 6 [3, 11] days and 15 [8, 26] days, respectively. The modes of ventilation used were pressure-regulated volume control (PRVC; 61% of epochs), pressure control (PC; 31%), pressure support (PS; 6%), and volume control (VC; 2%).

Patients were enrolled within 13 [6, 17] hours from the time they were placed on invasive ventilatory support. Patient were monitored for 46 [24, 70] hours for a total cohort monitoring time of 2,112 hours. We captured and classified 50,712 epochs, encompassing approximately 2.6 million breathing cycles.

As shown in Table 1, variable breathing was the most frequently observed pattern, occurring in 36.5% of epochs, while fully synchronous breathing was present in 21.0%. Asynchronous breathing accounted for 42.5% of epochs. Patients experienced asynchrony for 39 [20,59] % of the monitored time, with all patients displaying asynchronies at some point during the time monitored. Among asynchronous epochs, 57.2% were cycling asynchronies, 20.2% were ineffective efforts, 12.9% were double triggering, and 1.2% were auto-triggering. 33.8% of epochs classified as asynchronous displayed more than one morphological type of asynchrony, with 8.5% exhibiting three or more types.

Table 1

Classification of epoch’s patterns and severity made by three expert readers.
		Severity % within each category
Total epochs (n = 50,712)	% of all epochs	Mild (%)	Moderate (%)	Severe (%)
Normal breathing
Fully synchronous pattern (n = 10,631)	21.0
Variable-timing pattern (n = 18,528)	36.5	69.8	22.8	7.4
Asynchronous breathing (n = 21,553)
Auto Triggering	0.5	2.4	7.5	90.1
Cycling	24.3	33.7	28.4	37.9
Complex	3.6	2.7	9.1	88.2
Double Triggering	5.5	6.5	15.4	78.1
Ineffective Effort	8.6	20.9	37.2	41.9
Asynchronies total:	42.5	24.6	26.6	48.8

The distribution of epoch severity varied. Only 7.4% of epochs classified as variable breathing were rated as severe. In contrast, nearly half of the asynchronous epochs were deemed severe, with auto-triggering and complex asynchronies having the highest percentages of severe classifications.

Figure 2 illustrates the percentage of epochs classified as synchronous, timing-variable, and asynchronous that were treated with intravenous propofol. A greater percentage of synchronous epochs occurred in patients treated with intravenous propofol (p < 0.01), and these patients also received a higher infusion rate (p < 0.01).

The overall 28-day mortality rate for the cohort was 31.8%; lower than that predicted by their combined SAPS II score of 54.5 [47.0, 63.3]. Although no significant differences were noted in SAPS II scores between patients who died and those who survived (59.5 vs. 51.5), those who died spent a greater percentage of their monitored time asynchronous (52.4% [41.0, 62.5] vs. 33.7% [13.5, 50.4]; p = 0.015).

Algorithm development.

Nineteen variables having ≥ 5% correlation with the predicted outcomes were used to develop the four models. Ten of these variables were related to the frequency spectra of airway pressure and flow (online data supplement). The others were: F_IO₂, respiratory rate, tidal volume, minute ventilation, peak airway pressure, PEEP, the patient’s age and the covariances of total breath cycle timing and of peak pressure. Neither ventilator mode nor the primary diagnosis met the correlation requirement. Except for age, no other demographic variable appeared to improve model prediction.

Algorithm accuracy.

Table 2 shows evaluation metrics for the four trained Random Forest models. Model 1 detected whether the signals in a given epoch were asynchronous with 91% accuracy and excellent sensitivity and specificity. Model 2 excelled at identifying synchronous breathing but performed worst at identifying complex asynchronies. Asynchrony discrimination was good to excellent, with PPVs in the range of 70–100%. Model 3 excelled at identifying mild or severe epochs, with precisions of 94% and 82%, respectively. Model 4 showed excellent accuracy (93%) in detecting the possible presence of dynamic hyperinflation.

Table 2

Evaluation metrics for the Random Forest Classifier models (%)
	Instances in dataset (%)	PPV (%)	Sensitivity (%)	Specificity (%)	F1-score
Model 1 (Accuracy 91%)	Presence of asynchrony
No	58.2	92	92	89	0.92
Yes	41.8	86	89	92	0.89
Model 2 (Accuracy 82%)	Epoch identification
Synchronous breathing	21.1	92	91	97	0.91
Variable breathing	36.6	80	87	87	0.83
Asynchronies:
Auto Triggering	0.5	100	73	100	0.84
Cycling	24.3	77	81	92	0.79
Complex	3.6	70	52	100	0.59
Double Triggering	5.6	84	64	99	0.73
Ineffective Effort	8.6	83	67	99	0.74
Model 3 (Accuracy 87%)	Epoch Severity
Mild	60	94	94	89	0.94
Moderate	19	72	69	94	0.70
Severe	21	82	85	95	0.84
Model 4 (Accuracy 93%)	autoPEEP detection
No	74.6	95	97	83	0.96
Yes	25.4	90	83	97	0.86
PPV = Positive predictive value; f1 score = harmonic mean of precision and sensitivity

Algorithm reliability.

Table 3 presents the Cohen's kappa statistics for agreement between each individual reader and the predictions of Models 2 and 3 in classifying identical 1,000-epochs datasets. The agreement between the non-experts and the models was judged as fair, while those of the expert group ranged from moderate to substantial. Kappa statistics were significantly greater for expert readers compared to non-experts (p = 0.02 and 0.04 for Models 2 and 3, respectively).

Table 3

Cohen’s kappa statistic for agreement between expert and non-expert classification for Models 2 (Asynchrony Type) and Model 3 (Assess Severity)
	Model 2 Asynchrony Type	Model 3 Assessed Severity
Non-expert readers (n = 4)	0.25 ± 0.09	0.38 ± 0.15
Expert readers (n = 3)	0.46 ± 0.10 *	0.59 ± 0.11*
* p < 0.05 comparing Non-experts to Expert readers

The current standard for detecting asynchronous breathing consists of visually inspecting 30-minute-long recordings of airway flow and pressure signals [10, 11]. Quantification is based on the asynchrony index (AI), defined as the ratio of identified asynchronies to the total respiratory rate [1]. This post hoc, subjective, and labor-intensive method is low in sensitivity and positive predictive value [4, 12].

Traditional computer methods for identifying asynchrony have focused on comparing mathematical descriptions of normal breath morphology with those of asynchronous breaths. [13–16]. Alternatively, machine learning algorithms, which also rely on signal morphology comparison, have been trained to identify specific types of asynchronies. [17–21]. More recently, the timing of respiratory patterns during pressure support ventilation has been used to train neural networks. However, the method's requirement for esophageal measurements constrains its practical application. [22].

Our approach to identifying asynchronous breathing differs fundamentally from these prior efforts. Instead of analyzing individual breaths for morphological markers of asynchrony, we aimed to replicate the decision-making process of an experienced clinician when assessing a time window of airway signal data.

Epoch classification findings.

To achieve this goal, three expert clinicians visually assessed over 50,000 epochs of airway flow and pressure data, classifying each according to a predetermined set of options that included two non-asynchronous patterns (fully synchronous and variable breath-to-breath timing) and five asynchronous patterns. Each epoch was also evaluated based on the clinicians' subjective assessment of airway signal disruption and for the possible presence of dynamic hyperinflation, determined by the difference between end-expiratory flow and the zero-flow baseline.

Nearly one-fifth of all epochs were classified as fully synchronous, with a significant proportion linked to higher rates of intravenous propofol infusion, suggesting that fully synchronous epochs could be an indication of over-sedation. In contrast, variable-timing breathing was the most common classification, occurring in 36.5% of epochs, and is likely indicative of normal respiratory variability.

Epochs were classified as asynchronous when exhibiting four or more asynchronies. This threshold was chosen because the presence of four asynchronies within an epoch at a respiratory rate of 16 bpm corresponds to an Asynchrony Index (AI) greater than 10%, the accepted threshold for asynchronous breathing [1]. Consistent with the findings of Blanch et al [23], we observed that all patients experienced asynchronous breathing at some point during the monitoring period.

Asynchronous breathing accounted for 42% of all epochs, with cycling asynchrony being the most frequent classification. Nearly half of the asynchronous epochs were graded as severe, suggesting inadequate sedation or insufficient ventilatory support during these periods. Although less common, the vast majority of double-triggering and complex epochs were also graded as severe, indicating the need to reassess the adequacy of ventilation in patients experiencing these asynchronies.

We observed dynamic hyperinflation in association with other breathing patterns in 25% of all epochs. This finding aligns with the high prevalence of patients with pulmonary disease in our cohort (52%) who are prone to developing this condition [24, 25].

Model performance.

We used relatively few variables to train the different types of machine learning models. Except for age, these training variables were derived directly from ventilator-acquired data. The study extensively utilized spectral analysis by applying the Fast Fourier Transform (FFT) to airway signals, providing a comprehensive assessment of their behavior over each epoch's duration. The frequency spectra are relatively insensitive to artifacts that can distort these signals, such as strong heart palpitations or water in the ventilator tubing. Furthermore, the frequency spectrum is largely independent of ventilatory mode, which is a crucial advantage, given that most patients were ventilated with two or more modes during the monitoring period.

The Random Forest algorithm was selected due to its superior performance with our dataset relative to other machine learning models, as detailed in the Supplementary Material. Nevertheless, it is important to note that alternative machine learning algorithms may yield comparable or superior results, particularly when applied to larger datasets.

The excellent performances of Model 1 in detecting asynchronous breathing and that of Model 4 in identifying dynamic hyperinflation are likely due to their training with dichotomous data. In contrast, the performance of Models 2 and 3, which are multiclass classifiers, is inherently lower due to the complexity of handling multiple classes.

Model 2 effectively differentiated fully synchronous breathing from variable-timing synchrony with high sensitivity and specificity, an important feature given that highly synchronous patients are known to experience worse outcomes [26, 27]. The model also demonstrated good predictive value and excellent specificity in identifying asynchronous epochs. However, sensitivity was fair to poor for detecting double triggering and complex asynchronies, perhaps due to their relatively low prevalence in the database.

Model 3 is arguably the weakest of the four models, as it depends on epoch classifications that are purely subjective and heavily reliant on the classifier’s clinical experience. Conversely, we reasoned that establishing arbitrary guidelines to assess signal disruption would have introduced bias, undermining the study's aim to replicate clinician judgment. While Model 3 accurately identified mild and severe breathing patterns, it was less effective in predicting moderate severity, possibly reflecting uncertainty in scorer’s assessments of this severity level.

Algorithm reliability vs. accuracy.

Algorithm performance is typically evaluated by its accuracy, defined as the number of correct predictions made on an unseen test set. While models trained with reliable data tend to be both accurate and reliable, those trained with unreliable data may achieve high accuracy but be unreliable. Although artificial intelligence can reduce human bias by identifying relational patterns within large datasets, it cannot guarantee the model’s reliability.

We addressed the issue of algorithm’s reliability by using identical datasets to compare the classifications made by Models 2 and 3 to those made by clinicians with varying experience in asynchrony classification. As hypothesized, the expert group exhibited moderate to substantial agreement with the model predictions, whereas the non-expert group showed only fair agreement. Furthermore, the significantly higher kappa values for the expert group compared to the non-expert group, reinforces the clinical reliability of our algorithms.

Model limitations and further improvements.

The dataset, originating from a single institution and primarily employing PRVC and PC ventilation, may not generalize well to other settings due to its specific clinical practices. The limited ethnic diversity, with only 18% Caucasian participants, could also introduce bias.

During the study design, we recognized that visual inspection for asynchrony identification might introduce subjectivity and variability into the classification process. This issue, however, is inherent to machine learning model development, emphasizing the need for a large, well-classified dataset.

Assessors often encountered difficulties when evaluating epochs that displayed two types of asynchronies, necessitating classification based on the predominant asynchrony type. Further, some asynchrony types, such as flow asynchronies and reverse triggering, were excluded from the classification scheme given their low prevalence in the database.

Model accuracy and reliability could be enhanced by expanding the database incorporating diverse clinical experiences and data sources from other centers. Additionally, the models would benefit from including a broader range of breathing patterns, such as reverse triggering, flow asynchronies, and accounting for a duality of different asynchronies within a single epoch.

In summary, this study demonstrates the feasibility of developing machine learning algorithms to emulate experienced clinicians in evaluating breathing patterns during mechanical ventilation. The application of larger databases, along with advancements in artificial intelligence, may lead to powerful algorithms capable of establishing associations between airway signals and successful ventilatory support.

Ethical Approval and Consent to participate: Thedatabase used in the present study was collected with approval from The George Washington University Institutional Review Board (IRB No. 101228) in compliance with the 1964 Helsinki Declaration. Informed consent was obtained from the patients or their designated surrogates for participation in these studies.

Availability of supporting data: The dataset analyzed during the current study can be found in the Supplementary Material Section. Access to the database storing the raw data may be granted to qualified researchers upon reasonable request and adequate vetting.

Competing interests: GG is the owner of a patent related to the topic of the study: US 8,573,207,B2 Method and System to Detect Respiratory Asynchrony. The other authors declare no competing interests.

Authors contributions:

JW– Conception and design of the work. Acquisition of data and drafting the manuscript.

JA - Acquisition and interpretation of data; drafting the manuscript.

KW - Acquisition, analysis, and interpretation of data; drafting the manuscript.

AJ - Acquisition and interpretation of data; drafting the manuscript.

GG - Conception and design of the work; development of the machine-learning algorithms, acquisition, analysis, and interpretation of data; drafting the manuscript.

All authors agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Source of funding: Self-funded study.

Acknowledgments: The authors wish to thank the nurses of The George Washington University Intensive Care Unit for their cooperation in the study. We also extend our thanks to Drs. Ana McLean, Lisa Glass, Ali Khalofa, and Nawaf Almeshal for their help in data collection.

Disclaimer: The content is solely the authors’ responsibility and does not necessarily represent the official views of The George Washington University.

Thille A, Rodriguez P, Cabello B, Lellouche F, Brochard L (2006). Patient-ventilator asynchrony during assisted mechanical ventilation. Intensive Care Med 32:1515–22. doi: 10.1007/s00134-006-0301-8. Epub 2006 Aug 1. PMID: 16896854.
Epstein SK (2011). How often does patient-ventilator asynchrony occur and what are the consequences? Respir Care 56:25-38. doi: 10.4187/respcare.01009. PMID: 21235836.
Blanch L, Villagra A, Sales B, Montanya J, Lucangelo U, Luján M, et al (2015). Asynchronies during mechanical ventilation are associated with mortality. Intensive Care Med. 41:633-41. doi: 10.1007/s00134-015-3692-6. Epub 2015 Feb 19. PMID: 25693449.
Colombo D, Cammarota G, Alemani M, Carenzo L, Barra FL, Vaschetto R, et al (2011). Efficacy of ventilator waveforms observation in detecting patient-ventilator asynchrony. Crit Care Med. 39:2452-7. doi: 10.1097/CCM.0b013e318225753c. PMID: 21705886.
Le Gall JR, Neumann A, Hemery F, Bleriot JP, Fulgencio JP, Garrigues B, et al (2005). Mortality prediction using SAPS II: an update for French intensive care units. Crit Care 9:R645-52. doi: 10.1186/cc3821. Epub 2005 Oct 6. PMID: 16280063; PMCID: PMC1414016.
Chanques G, Kress JP, Pohlman A, Patel S, Poston J, Jaber S, et al (2013). Impact of ventilator adjustment and sedation-analgesia practices on severe asynchrony in patients ventilated in assist-control mode. Crit Care Med. 41:2177-87. doi: 10.1097/CCM.0b013e31828c2d7a. PMID: 23782972.
Gutierrez G (2020). Artificial Intelligence in the Intensive Care Unit. Crit Care 24:101. doi: 10.1186/s13054-020-2785-y. Erratum in: Crit Care 2024 Mar 21;28(1):94. doi: 10.1186/s13054-024-04856-9. PMID: 32204716; PMCID: PMC7092485.
Cohen J. A (1960). A coefficient of agreement for nominal scales. Educational and Psychological Measurement 20: 37–46.
Landis JR, Koch GG (1977). The measurement of observer agreement for categorical data. Biometrics 33: 159–174.
Georgopoulos D, Prinianakis G, Kondili E (2006). Bedside waveforms interpretation as a tool to identify patient-ventilator asynchronies. Intensive Care Med 32:34-47. doi: 10.1007/s00134-005-2828-5. Epub 2005 Nov 9. PMID: 16283171.
Doerschug KC (2022). Patient-Ventilator Synchrony. Clin Chest Med. 43:511-18. doi: 10.1016/j.ccm.2022.05.005. PMID: 36116818.
Silva DO, de Souza PN, de Araujo Sousa ML, Morais CCA, Ferreira JC, Holanda MA, et al (2023). Impact on the ability of healthcare professionals to correctly identify patient-ventilator asynchronies of the simultaneous visualization of estimated muscle pressure curves on the ventilator display: a randomized study (P_mus study). Crit Care 27:128. doi: 10.1186/s13054-023-04414-9. PMID: 36998022; PMCID: PMC10064577.
Younes M, Brochard L, Grasso S, Kun J, Mancebo J, Ranieri M, et al (2007). A method for monitoring and improving patient: ventilator interaction. Intensive Care Med. 33:1337-46. doi: 10.1007/s00134-007-0681-4. Epub 2007 May 31. PMID: 17541554.
Mulqueeny Q, Ceriana P, Carlucci A, Fanfulla F, Delmastro M, Nava S (2007). Automatic detection of ineffective triggering and double triggering during mechanical ventilation. Intensive Care Med. 33:2014-18. doi: 10.1007/s00134-007-0767-z. Epub 2007 Jul 5. PMID: 17611736.
Chen CW, Lin WC, Hsu CH, Cheng KS, Lo CS (2008). Detecting ineffective triggering in the expiratory phase in mechanically ventilated patients based on airway flow and pressure deflection: feasibility of using a computer algorithm. Crit Care Med 36:455-61. doi: 10.1097/01.CCM.0000299734.34469.D9. PMID: 18091543.
Blanch L, Sales B, Montanya J, Lucangelo U, Garcia-Esquirol O, Villagra A, et al (2012). Validation of the Better Care® system to detect ineffective efforts during expiration in mechanically ventilated patients: a pilot study. Intensive Care Med 38:772-80. doi: 10.1007/s00134-012-2493-4. Erratum in: Intensive Care Med. 2013 Feb;39(2):341. PMID: 22297667.
Adams JY, Lieng MK, Kuhn BT, Rehm GB, Guo EC, Taylor SL, et al (2017). Development and Validation of a Multi-Algorithm Analytic Platform to Detect Off-Target Mechanical Ventilation. Sci Rep. 7:14980. doi: 10.1038/s41598-017-15052-x. PMID: 29101346; PMCID: PMC5670237.
Mulqueeny Q, Redmond SJ, Tassaux D, Vignaux L, Jolliet P, Ceriana P, et al (2009). Automated detection of asynchrony in patient-ventilator interaction. Conf Proc IEEE Eng Med Biol Soc. 5324-27. doi: 10.1109/IEMBS.2009.5332684. PMID: 19963896.
Sottile PD, Albers D, Higgins C, Mckeehan J, Moss MM (2018). The Association Between Ventilator Dyssynchrony, Delivered Tidal Volume, and Sedation Using a Novel Automated Ventilator Dyssynchrony Detection Algorithm. Crit Care Med 46:e151-e157. doi: 10.1097/CCM.0000000000002849. PMID: 29337804; PMCID: PMC5772880.
Gholami B, Phan TS, Haddad WM, Cason A, Mullis J, Price L, et al (2018). Replicating human expertise of mechanical ventilation waveform analysis in detecting patient-ventilator cycling asynchrony using machine learning. Comput Biol Med. 97:137-44. doi: 10.1016/j.compbiomed.2018.04.016. Epub 2018 Apr 23. PMID: 29729488.
Loo NL, Chiew YS, Tan CP, Arunachalam G, Ralib AM, Mat-Nor MB (2018). A machine learning model for real-time asynchronous breathing monitoring. IFAC-PapersOnLine 51; 378-83.
Bakkes T, van Diepen A, De Bie A, Montenij L, Mojoli F, Bouwman A, et al (2023). Automated detection and classification of patient-ventilator asynchrony by means of machine learning and simulated data. Comput Methods Programs Biomed. 230:107333. doi: 10.1016/j.cmpb.2022.107333. Epub 2023 Jan 2.
Blanch L, Villagra A, Sales B, Montanya J, Lucangelo U, Luján M, et al (2015). Asynchronies during mechanical ventilation are associated with mortality. Intensive Care Med. 41:633-41. doi: 10.1007/s00134-015-3692-6. Epub 2015 Feb 19. PMID: 25693449.
Beitler JR, Sands SA, Loring SH, Owens RL, Malhotra A, Spragg RG, et al (2016). Quantifying unintended exposure to high tidal volumes from breath stacking dyssynchrony in ARDS: the BREATHE criteria. Intensive Care Med. 42:1427-36. doi: 10.1007/s00134-016-4423-3. Epub 2016 Jun 24. PMID: 27342819; PMCID: PMC4992404.
Pohlman MC, McCallister KE, Schweickert WD, Pohlman AS, Nigos CP, Krishnan JA, et al (2008). Excessive tidal volume from breath stacking during lung-protective ventilation for acute lung injury. Crit Care Med 36:3019-23. doi: 10.1097/CCM.0b013e31818b308b. PMID: 18824913.
Gutierrez G, Das A, Ballarino G, Beyzaei-Arani A, Türkan H, Wulf-Gutierrez M, et al (2013). Decreased respiratory rate variability during mechanical ventilation is associated with increased mortality. Intensive Care Med. 39:1359-67. doi: 10.1007/s00134-013-2937-5. Epub 2013 Jun 7. PMID: 23743521.
Rolland-Debord C, Poitou T, Bureau C, Rivals I, Similowski T, Demoule A (2023). Decreased breathing variability is associated with poorer outcome in mechanically ventilated patients. ERJ Open Res. 9:00544-2022. doi: 10.1183/23120541.00544-2022. PMID: 37143829; PMCID: PMC10152249.

Download PDF

Version 1

posted

You are reading this latest preprint version

Machine Learning Algorithms to Detect Patient-Ventilator Asynchrony. A Feasibility Study.

Status:

Version 1

Abstract

Figures

Take-home message

Introduction

MATERIALS AND METHODS

RESULTS

DISCUSSION

Statements and Declarations

References

Supplementary Files

Status:

Version 1