Development and Validation of a Deep Learning Classifier Using Chest Radiographs to Predict Extubation Success in Patients Undergoing Invasive Mechanical Ventilation.

The decision to extubate patients on invasive mechanical ventilation is critical; however, clinician performance in identifying patients to liberate from the ventilator is poor. Machine Learning-based predictors using tabular data have been developed; however, these fail to capture the wide spectrum of data available. Here, we develop and validate a deep learning-based model using routinely collected chest X-rays to predict the outcome of attempted extubation. We included 2288 serial patients admitted to the Medical ICU at an urban academic medical center, who underwent invasive mechanical ventilation, with at least one intubated CXR, and a documented extubation attempt. The last CXR before extubation for each patient was taken and split 79/21 for training/testing sets, then transfer learning with k-fold cross-validation was used on a pre-trained ResNet50 deep learning architecture. The top three models were ensembled to form a final classifier. The Grad-CAM technique was used to visualize image regions driving predictions. The model achieved an AUC of 0.66, AUPRC of 0.94, sensitivity of 0.62, and specificity of 0.60. The model performance was improved compared to the Rapid Shallow Breathing Index (AUC 0.61) and the only identified previous study in this domain (AUC 0.55), but significant room for improvement and experimentation remains.

2022 Year in Review: Ventilator Liberation.

Mechanical ventilation is ubiquitous in critical care, and duration of ventilator liberation is variable and multifactorial. While ICU survival has increased over the last two decades, positive-pressure ventilation can cause harm to patients. Weaning and discontinuation of ventilatory support is the first step in ventilator liberation. Clinicians have a wealth of evidence-based literature at their disposal; however, more high-quality research is needed to describe outcomes. Additionally, this knowledge must be distilled into evidence-based practice and applied at the bedside. A proliferation of research on the subject of ventilator liberation has been published in the last 12 months. Whereas some authors have reconsidered the value of applying the rapid shallow breathing index in weaning protocols, others have begun to investigate new indices to predict liberation outcomes. New tools such as diaphragmatic ultrasonography have begun to appear in the literature as a tool for outcome prediction. A number of systematic reviews with both meta-analysis and network meta-analysis that synthesize the literature on ventilator liberation have also been published in the last year. This review describes changes in performance, monitoring of spontaneous breathing trials, and evaluations of successful ventilator liberation.

Association Between Dead Space to Tidal Volume Ratio and Duration of Respiratory Support After Extubation in Critically Ill Children.

BACKGROUND: The dead-space-to-tidal-volume ratio (VD/VT) has been used to successfully predict extubation failure in children who are critically ill. However, a singular reliable measure to predict the level and duration of respiratory support after liberation from invasive mechanical ventilation has remained elusive. The objective of this study was to evaluate the association between VD/VT and the duration of postextubation respiratory support. METHODS: This was a retrospective cohort study of subjects who were mechanically ventilated and admitted to a single-center pediatric ICU between March 2019 and July 2021, and who had been extubated with a recorded VD/VT. A cutoff of 0.30 was chosen a priori, with subjects divided into 2 groups, VD/VT < 0.30 and VD/VT ≥ 0.30, and postextubation respiratory support was recorded at specified time intervals (24 h, 48 h, 72 h, 7 d, and 14 d). RESULTS: We studied 54 subjects. Those with VD/VT ≥ 0.30 had a significantly longer median (interquartile range) duration of respiratory support after extubation (6 [3-14] d vs 2 [0-4] d; P = .001) and longer median (interquartile range) ICU stay (14 [12-19] d vs 8 [5-22] d; P = .046) versus the subjects with VD/VT < 0.30. The distribution of respiratory support did not differ significantly between VD/VT at the time of extubation (P = .13) or at 14 d after extubation (P = .21) but was significantly different during the intervening time points after extubation (24 h [P = .01], 48 h [P < .001], 72 h [P < .001], and 7 d [P = .02]). CONCLUSIONS: VD/VT was associated with the duration and level of respiratory support needed after extubation. Prospective studies are needed to establish if VD/VT can successfully predict the level of respiratory support after extubation.

Association of Noninvasive Respiratory Support with Extubation Outcomes in Brain-injured Patients Receiving Mechanical Ventilation: A Secondary Analysis of the ENIO Prospective Observational Study.

Rationale: Noninvasive respiratory support using a high-flow nasal cannula (HFNC) or noninvasive positive pressure ventilation (NIPPV) can decrease the risk of reintubation in patients being liberated from mechanical ventilation, but effects in patients with acute brain injury (ABI) are unknown. Objectives: To evaluate the association between postextubation noninvasive respiratory support and reintubation in patients with ABI being liberated from mechanical ventilation. Methods: This was a secondary analysis of a prospective, observational study of mechanically ventilated patients with ABI (clinicaltrials.gov identifier NCT03400904). The primary endpoint was reintubation during ICU admission. We used mixed-effects logistic regression models with patient-level covariates and random intercepts for hospital and country to evaluate the association between prophylactic (i.e., planned) HFNC or NIPPV and reintubation. Measurements and Main Results: 1,115 patients were included from 62 hospitals and 19 countries, of whom 267 received HFNC or NIPPV following extubation (23.9%). Compared with conventional oxygen therapy, neither prophylactic HFNC nor NIPPV was associated with decreased odds of reintubation (respectively, odds ratios of 0.97 [95% confidence interval, 0.54-1.73] and 0.63 [0.30-1.32]). Findings remained consistent in sensitivity analyses accounting for alternate adjustment procedures, missing data, shorter time frames of the primary endpoint, and competing risks precluding reintubation. In a Bayesian analysis using skeptical and data-driven priors, the probabilities of reduced reintubation ranged from 17% to 34% for HFNC and from 46% to 74% for NIPPV. Conclusions: In a large cohort of brain-injured patients undergoing liberation from mechanical ventilation, prophylactic use of HFNC and NIPPV were not associated with reintubation. Prospective trials are needed to confirm treatment effects in this population. Primary study registered with www.clinicaltrials.gov (NCT03400904).

Prediction Model of Extubation Outcomes in Critically Ill Patients: A Multicenter Prospective Cohort Study.

Liberation from mechanical ventilation is of great importance owing to related complications from extended ventilation time. In this prospective multicenter study, we aimed to construct a versatile model for predicting extubation outcomes in critical care settings using obtainable physiological predictors. The study included patients who had been extubated after a successful 30 min spontaneous breathing trial (SBT). A multivariable logistic regression model was constructed to predict extubation outcomes (successful extubation without reintubation and uneventful extubation without reintubation or noninvasive respiratory support) using eight parameters: age, heart failure, respiratory disease, rapid shallow breathing index (RSBI), PaO2/FIO2, Glasgow Coma Scale score, fluid balance, and endotracheal suctioning episodes. Of 499 patients, 453 (90.8%) and 328 (65.7%) achieved successful and uneventful extubation, respectively. The areas under the curve for successful and uneventful extubation in the novel prediction model were 0.69 (95% confidence interval (CI), 0.62−0.77) and 0.70 (95% CI, 0.65−0.74), respectively, which were significantly higher than those in the conventional model solely using RSBI (0.58 (95% CI, 0.50−0.66) and 0.54 (95% CI, 0.49−0.60), p = 0.004 and <0.001, respectively). The model was validated using a bootstrap method, and an online application was developed for automatic calculation. Our model, which is based on a combination of generally obtainable parameters, established an accessible method for predicting extubation outcomes after a successful SBT.

Respiratory Therapist-Driven Extubation Readiness Testing in a Single Pediatric ICU.

BACKGROUND: There is currently no standardized way to determine suitability for extubation of pediatric ICU (PICU) patients, potentially resulting in prolonged duration of mechanical ventilation. We aimed to design and implement a protocol for screening all intubated PICU patients for extubation readiness. METHODS: We adopted the quality improvement (QI) Model for Improvement with Plan-Do-Study-Act (PDSA) cycles to achieve this aim. This QI project was conducted over 11 months in a multidisciplinary PICU. Outcome measures included the (1) development of a standardized extubation readiness test (ERT) that was acceptable and safe; (2) performance of ERT on > 80% of all mechanically ventilated subjects; and (3) maintenance or reduction in mechanical ventilation duration, extubation failure (non-elective re-intubation within 48 h of extubation), and need for rescue noninvasive ventilation (NIV). Balancing measures were to ensure (1) no compromise of the subject's clinical status; and (2) acceptability of the ERT workflow by medical, nursing, and respiratory therapist (RT) teams. RESULTS: Four PDSA cycles were necessary to achieve the aims of this study. During the QI period, 438 subjects were admitted to the PICU. The ERT was championed by the RTs who conducted the test during office hours. ERT performance increased from 0% (baseline) to 90% (fourth PDSA cycle). Extubation failure rate after implementing ERT was reduced compared to baseline (4/31 [12.9%] vs 3/127 [2.4%], P = .01), whereas need for rescue NIV (3/31 [9.7%] vs 10/127 [7.9%], P = .74) and duration of mechanical ventilation (2 [1-7] d vs 1 [1-3] d, P = .09) were unchanged. PICU length of stay was reduced after implementing ERT (5 [3-10] d vs 3 [1-6] d, P = .01). No subject was destabilized as a result of ERT, and PICU staff found the workflow acceptable. CONCLUSIONS: An acceptable and safe ERT protocol was implemented and found to improve outcomes in PICU subjects on mechanical ventilation.

Mechanical Ventilator Liberation of Patients With COVID-19 in Long-term Acute Care Hospital.

BACKGROUND: Mechanical ventilation (MV) via tracheostomy is performed commonly for patients who are in long-term acute care hospitals (LTACHs) after respiratory failure. However, the outcome of MV in COVID-19-associated respiratory failure in LTACHs is not known. RESEARCH QUESTION: What is the ventilator liberation rate of patients who have received tracheostomy with COVID-19-associated respiratory failure compared with those with respiratory failure unrelated to COVID-19 in LTACHs? STUDY DESIGN AND METHODS: In this retrospective cohort study, we examined mechanically ventilated patients discharged between June 2020 and March 2021. Of 242 discharges, 165 patients who had undergone tracheostomy arrived and were considered for ventilator liberation. One hundred twenty-eight patients did not have COVID-19 and 37 patients were admitted for COVID-19. RESULTS: The primary outcome of the study was ventilator liberation; secondary outcomes were functional recovery, length of stay (LOS) at the LTACH, and discharge disposition. After controlling for demographics, the number of comorbidities, hemodialysis, vasopressor need, thrombocytopenia, and the LOS at the short-term acute care hospital, our results indicated that patients with COVID-19 showed a higher adjusted ventilator liberation rate of 91.4% vs 56.0% in those without COVID-19. Functional ability was assessed with the change of Functional Status Score for the Intensive Care Unit (FSS-ICU) between admission and discharge. The adjusted mean change in FSS-ICU was significantly higher in the COVID-19 group than in the non-COVID-19 group: 9.49 (95% CI, 7.38-11.6) vs 2.08 (95% CI, 1.05-3.11), respectively (P < .001). Patients with COVID-19 experienced a shorter adjusted LOS at the LTACH with an adjusted hazard ratio of 1.57 (95% CI, 1.0-2.46; P = .05) compared with patients without COVID-19. We did not observe significant differences between the two groups regarding discharge location, but a trend toward need for lower level of care was found in patients with COVID-19. INTERPRETATION: Our study suggests that patients with COVID-19 requiring MV and tracheostomy have a higher chance for recovery than those without COVID-19.

Establishing Medicaid incentives for liberating nursing home patients from ventilators.

BACKGROUND: Chronic ventilator use in Tennessee nursing homes surged following 2010 increases in respiratory care payment rates. Tennessee's Medicaid program implemented multiple policies between 2014 and 2017 to promote ventilator liberation in 11 nursing homes, including quality reporting, on-site monitoring, and pay-for-performance incentives. METHODS: Using repeated cross-sectional analysis of Medicare and Medicaid nursing home claims (2011-2017), hospital discharge records (2010-2017), and nursing home quality reports (2015-2017), we examined how service use changed as Tennessee implemented policies designed to promote ventilator liberation in nursing homes. We measured the annual number of nursing home patients with ventilator-related service use; discharge destination of ventilated inpatients and percent of nursing home patients liberated from ventilators. RESULTS: Between 2011 and 2014, the number of Medicare SNF and Medicaid nursing home patients with ventilator use increased more than sixfold. Among inpatients with prolonged mechanical ventilation, discharges to home decreased as discharges to nursing homes increased. As Tennessee implemented policy changes, ventilator-related service use moderately declined in nursing homes from a peak of 198 ventilated Medicare SNF patients in 2014 to 125 in 2017 and from 182 Medicaid patients with chronic ventilator use in 2014 to 145 patients in 2017. Nursing home weaning rates peaked at 49%-52% in 2015 and 2016, but declined to 26% by late 2017. Median number of days from admission to wean declined from 81 to 37 days. CONCLUSIONS: This value-based approach demonstrates the importance of designing payment models that target key patient outcomes like ventilator liberation.

Role of a successful spontaneous breathing trial in ventilator liberation in brain-injured patients.

BACKGROUND: Spontaneous breathing trials (SBTs) have been shown to improve outcomes in critically ill patients. However, in patients with brain injury, indications for intubation and mechanical ventilation are different from those of non-neurological patients, and the role of an SBT in patients with brain injury is less established. The aim of the present study was to compare key respiratory variables acquired during a successful SBT between patients with successful ventilator liberation versus failed ventilator liberation. METHODS: In this prospective study, patients with brain injury (≥18 years of age), who completed a 30-min SBT, were enrolled. Airway pressure, flow, esophageal pressure, and diaphragm electrical activity (ΔEAdi) were recorded before (baseline) and during the SBT. Respiratory rate (RR), tidal volume, inspiratory muscle pressure (ΔPmus), ΔEAdi, and neuromechanical efficiency (ΔPmus/ΔEAdi) of the diaphragm were calculated breath by breath and compared between the liberation success and failure groups. Failed liberation was defined as the need for invasive ventilator assistance within 48 h after the SBT. RESULTS: In total, 46 patients (51.9±13.2 years, 67.4% male) completed the SBT. Seventeen (37%) patients failed ventilator liberation within 48 h. Another 11 patients required invasive ventilation within 7 days after completing the SBT. There were no differences in baseline characteristics between the success and failed groups. In-depth analysis showed similar changes in patterns and values of respiratory physiological parameters between the groups. CONCLUSIONS: In patients with brain injury, ventilator liberation failure was common after successful SBT. In-depth physiological analysis during the SBT did not provide data to predict successful liberation in these patients. TRIAL REGISTRATION: The trial was registered at ClinicalTrials.gov (No. NCT02863237).

Extubation strategies in neuro-intensive care unit patients and associations with outcomes: the ENIO multicentre international observational study.

BACKGROUND: Prolonged invasive ventilation is common in patients with severe brain injury. Information on optimal management of extubation and on the use of tracheostomy in these patients is scarce. International guidelines regarding the ventilator liberation and tracheostomy are currently lacking. METHODS: The aim of 'Extubation strategies in Neuro-Intensive care unit patients and associations with Outcomes' (ENIO) study is to describe current management of weaning from invasive ventilation, focusing on decisions on timing of tracheal extubation and tracheostomy in intensive care unit (ICU) patients with brain injury. We conducted a prospective, international, multi-centre observational study enrolling patients with various types of brain injury, including trauma, stroke, and subarachnoid haemorrhage, with an initial Glasgow Coma Score equal or less than 12, and a duration of invasive ventilation longer than 24 hours from ICU admission. ENIO is expected to include at least 1,500 patients worldwide. The primary endpoint of the ENIO study is extubation success in the 48 hours following endotracheal tube removal. The primary objective is to validate a score predictive of extubation success. To accomplish this, the study population will be randomly divided to a development cohort (2/3 of the included patients) and a validation cohort (the remaining 1/3). Secondary objectives are: to determine the incidence of extubation success rate according to various time-frames (within 96 hours, >96 hours after extubation); to validate (existing) prediction scores for successful extubation according to various time-frames and definitions (i.e., tracheostomy as extubation failure); and to describe the current practices of extubation and tracheostomy, and their associations. DISCUSSION: ENIO will be the largest prospective observational study of ventilator liberation and tracheostomy practices in patients with severe brain injury undergoing invasive mechanical ventilation, providing a validated predictive score of successful extubation. TRIAL REGISTRATION: The ENIO study is registered in the Clinical Trials database: NCT03400904.

2018 Year in Review: Adult Invasive Mechanical Ventilation.

Research in the area of adult invasive mechanical ventilation is rich and diverse. With more than 3,200 articles on mechanical ventilation published in 2018, isolating the most relevant literature is a challenge. Separated into 5 themes (ie, ventilatory support, hyperoxia, ventilator-associated events, prevention of events, and ventilator liberation), this article will describe the most important papers published on adult invasive mechanical ventilation in 2018.