Patient–ventilator asynchrony in acute brain-injured patients: a prospective observational study

This prospective observational study was conducted in the intensive care unit (ICU) of Beijing Tiantan Hospital, Capital Medical University. We included mechanically ventilated patients with acute brain injury, which was defined a priori as traumatic brain injury, stroke (ischemic stroke, spontaneous intracerebral hemorrhage and subarachnoid hemorrhage) and post-craniotomy for brain tumor [13, 14]. Exclusion criteria were as follows: (1) age younger than 18 years old; (2) anticipated duration of mechanical ventilation less than 24 h; (3) presented with epilepsy; (4) presented with agitation; (5) contraindications for esophageal balloon catheter insertion which included evidence of severe coagulopathy, diagnosed or suspected esophageal varices, and history of esophageal, gastric or lung surgery; (6) evidence of active air leak from the lung which included bronchopleural fistula, pneumothorax, pneumomediastinum, and an existing chest tube; (7) moribund conditions with a low likelihood of survival for more than 24 h; and (8) refusal to participate.

The study protocol was approved by the institutional review boards of Beijing Tiantan Hospital, Capital Medical University (KY 2017-028-02), and was registered at ClinicalTrials.gov on July 4, 2017 (NCT03212482). Written informed consent was obtained from the patient or appropriate substitute decision-makers.

After the patient enrollment, an AVEA ventilator (CareFusion Co., USA) was applied. The nasogastric tube was replaced by an esophageal balloon catheter (SmartCath-G catheter: LOT 7003300, CareFusion Co., San Diego, CA, USA) which can provide esophageal pressure monitoring and tube feeding simultaneously.

Before the catheter insertion, a balloon leak test was performed using the Esophageal Maneuver function on the ventilator. In order to reflect the pleural pressure accurately, the esophageal balloon was placed in the lower two-thirds of the intrathoracic esophagus. The balloon volume was determined by the ventilator automatically. Then the optimal balloon position was adjusted by Baydur’s occlusion test in patients with spontaneous breathing [15] or positive pressure occlusion test in patients during passive ventilation [16].

The investigators did not involve in the treatment of the patients. The selection of ventilator modes and settings was decided by the responsible ICU physicians and remained unchanged during waveform data collection.

In our unit, assist/control modes, either pressure assist/control ventilation (PACV) or volume assist/control ventilation (VACV), are usually initiated in patients with acute hypoxemic respiratory failure. The ventilator mode is switched to pressure support ventilation (PSV) as long as the patient triggers all ventilator breaths during assist/control ventilation. If the backup ventilation is induced during PSV, pressure-preset synchronized intermittent mandatory ventilation plus pressure support (SIMV + PS) is used. The tidal volume is set as 6–8 ml/kg predicted body weight, which is obtained by adjusting inspiratory pressure during pressure-preset ventilation. The respiratory rate is usually set to maintain an arterial partial pressure of carbon dioxide of 35–40 mmHg as long as possible. The inspiratory trigger sensitivity is usually set as 1–2 L/min for flow-trigger and 1.5–3 cmH₂O for pressure-trigger. During VACV, a decelerating flow with a peak flow between 45 and 60 L/min is set to obtain an inspiratory-to-expiratory ratio of 1:1.5–2. During PSV, a flow cycling is usually set as 25% of peak inspiratory flow. Inspired oxygen fraction and PEEP are set according to the patient’s oxygenation status.

Clinical data were collected at the study entry, including demographic data, Acute Physiology and Chronic Health Evaluation (APACHE) II score at the ICU admission, classification of brain injury, the ratio of the arterial partial pressure of oxygen to inspired oxygen fraction and the Sequential Organ Failure Assessment (SOFA) score at the inclusion, and duration of mechanical ventilation before the inclusion.

Flow–time, airway pressure–time and esophageal pressure–time waveforms were collected with a laptop via a dedicated acquisition system (VOXP Research Data Collector 3.2, Applied Biosignals GmbH, Weener, Germany) at 100 Hz for offline analysis. The signals were recorded four times daily (at 03:00 AM, 09:00 AM, 15:00 PM, and 21:00 PM), each for 15 min and saved as one waveform dataset. We collected ventilatory modes and settings, physiological variables, the use of sedatives, analgesics and neuromuscular blocking agents at the beginning of waveform recording. After the dataset collection, the Glasgow Coma Scale (GCS) and the Sedation-Agitation Scale (SAS) were evaluated. Arterial blood gas analysis was also collected at least once per day for each patient after the recording. The investigators were not involved in the treatment of the patients. Ventilatory modes and settings, as well as sedation and analgesia, were decided by the responsible ICU physicians and remained unchanged during the waveform data collection. At 30 min prior to waveform recording, the airway was suctioned. During the recording, unnecessary suctioning and physical therapy were avoided. The waveform dataset collection was conducted for 3 days as long as the patient was not weaned from the ventilator.

At the end of each waveform dataset recording, end-expiratory occlusion was performed. For patients with spontaneous breathing, the airway occlusion pressure (P_0.1) was measured and averaged for five occlusions to represent the respiratory drive [17].

Patients were followed up until 60 days after the enrollment, hospital discharge or death, whichever occurred first. Duration of mechanical ventilation after the inclusion, length of stay in the ICU and hospital after the inclusion, the Glasgow Outcome Scale (GOS), and all-cause mortality were documented.

Seven asynchrony patterns were determined through a priori definition, including ineffective triggering, double-triggering, auto-triggering, flow insufficiency, premature cycling, delayed cycling, and reverse triggering [1, 2, 18‐20]. The detailed description and example waveforms are shown in Additional file 1.

The collected waveform datasets were offline analyzed independently by the two investigators (XYL and XH). All breaths were inspected and labeled as no patient–ventilator asynchrony or one of the seven types of asynchrony defined above. The diagnosis of the patient–ventilator asynchrony was confirmed as the same decisions were made by the two investigators. When the two readings were discrepant, the final decision was made based on a group discussion (HLL, YLY and JXZ). Investigators taking part in the diagnosis of asynchrony were blinded to the patient’s clinical data and outcome data.

Overall and each type of asynchronous breath were counted. The prevalence of patient–ventilator asynchrony was calculated on the basis of all collected breaths and enrolled patients. The severity of the patient–ventilator asynchrony was represented by the asynchrony index, which is equal to the number of asynchronous breaths divided by the total number of breaths collected and multiplied by 100% [21]. The asynchrony index was calculated in each 15-min waveform dataset and in each patient after merging the collected datasets into a single recording session. Severe patient–ventilator asynchrony was defined as an asynchrony index  ≥ 10% in either dataset-based or patient-based analysis [22, 23].

Inter-observer reliability for detecting asynchrony was evaluated using a weighted kappa statistic and 95% confidence interval (CI). The prevalence and 95% CI of patient–ventilator asynchrony were calculated in all collected breaths and patient-based analysis. The present study was a descriptive exploratory study; therefore, no formal sample size calculation was performed. We planned to enroll 100 patients with anticipating 1000 15-min waveform datasets collection.

Continuous variables were presented as the median and interquartile range (IQR) and categorical variables as numbers and percentages. The prevalence of asynchrony was compared among different classifications of brain injury by the Chi square test. The asynchrony index was compared among different ventilatory modes and different analgesia and sedation strategies by the Kruskal–Wallis test. P_0.1 was also compared among different types of asynchrony by the Kruskal–Wallis test. All pairwise comparisons were performed with Bonferroni correction. The asynchrony index between dataset with SAS ≤ 2 and ≥ 3, as well as between datasets with GCS ≤ 8 and ≥ 9, was compared by Mann–Whitney U-test.

Associated factors with the severe ineffective triggering were analyzed by the Mann–Whitney U-test for continuous variables and the Chi square test or Fisher’s exact test for categorical variables.

Among the included patients, a total of 1076 15-min waveform datasets covering 330,292 breaths were collected and analyzed, in which 70,156 (38%, 95% CI 35%–40%) asynchronous breaths were detected. Ineffective triggering (50%) and premature cycling (28%) comprised the majority of asynchrony types (Fig. 2).

After merging datasets into one single session in each patient, asynchrony occurred in 96 (96%, 95%CI 92%–100%) patients, with the most prevalent type of ineffective triggering followed by double-triggering, premature cycling, delayed cycling, and flow insufficiency (Table 2). The median (IQR) asynchrony index was 12.4% (4.3%–26.4%) (Additional file 1: Table S2). No significant difference was found in either prevalence or asynchrony index among the three classifications of brain injury (Table 2 and Fig. 3).

During waveform datasets recording, four ventilatory modes were used, including PSV (65%), PACV (19%), SIMV + PS (10%), and VACV (6%). The proportion of ventilatory mode in each type of asynchrony is shown in (Additional file 1: Figure S1). For 503 datasets with a single asynchrony type, triggering (ineffective, double and auto-triggering) and cycling asynchrony were predominantly detected during PSV (70% to 100%). All single occurred flow insufficiency was detected during VACV, while all single occurred reverse triggering was detected during assist/control modes (PACV 53% and VACV 47%). The proportion of modes in datasets with multiple asynchrony types (n = 403) was 62%, 17%, 16% and 5% during PSV, PACV, SIMV + PS, and VACV, respectively.

Patient–ventilator asynchrony detected in datasets with PACV (65%) was significantly less than those with PSV (89%), VACV (86%) and SIMV + PS (89%) (p values ranged from < 0.001 to 0.003), but no significant difference was found among the latter three modes (p values ranged from 0.538 to > 0.999) (Fig. 4a).

A significant difference in the asynchrony index was also found among the four ventilatory modes (p < 0.001), with significantly lower indexes during PACV and PSV compared to those during VACV and SIMV + PS (p values ranged from < 0.001 to 0.011) (Fig. 4b).

Spontaneous breathing was preserved in 1037 (96%) datasets during recording. P_0.1 values in datasets with a single type of asynchrony are shown in Fig. 5. Compared to the datasets without asynchrony (n = 141, 1.7 [1.2–3.1] cmH₂O), P_0.1 was significantly lower in datasets with ineffective triggering (n = 430, 1.2 [0.9–2.0] cmH₂O, p < 0.001). Although there was a tendency in elevated P_0.1 values in datasets with premature cycling (n = 31, 3.0 [1.5–3.3] cmH₂O) and flow insufficiency (n = 9, 4.5 [1.4–4.9] cmH₂O), no statistical significance was found (p > 0.999).

Asynchrony index in datasets with the combined use of opioids and sedatives was significantly higher than those with single use and no use of these two kinds of drugs (Fig. 6). For analysis of specific drugs, the asynchrony index was significantly higher during the combined use of fentanyl and midazolam, compared to that during single administration of midazolam (p < 0.001) or remifentanil (p = 0.025) as well as that without opioids and sedatives administration (p = 0.016) (Additional file 1: Fig. S2). No significant difference was found in asynchrony index either between datasets with SAS ≤ 2 and ≥ 3 (median [IQR] of 5.2% [0.4%–36.8%] vs. 4.8% [0.7%–26.0%], p = 0.827) or between datasets with GCS ≤ 8 and ≥ 9 (median [IQR] of 4.0% [0.4%–31.1%] vs. 5.9% [1.0%–27.0%], p = 0.124).

In the dataset-based analysis, 275 (26%) datasets were defined as severe ineffective triggering (asynchrony index ≥ 10%), and associated factors included SIMV + PS mode compared to PSV mode, a lower respiratory rate, a higher tidal volume, lower minute ventilation, the use of flow-trigger with a more sensitivity setting, a lower P_0.1, a higher GCS, and higher arterial partial pressure of carbon dioxide and concentration of bicarbonate (Additional file 1: Table S3).

In datasets collected during pressure-preset ventilatory modes (PACV, PSV and SMIV + PS), set inspiratory pressure was significantly higher in the severe ineffective triggering group than that in the non-severe group (median [IQR] of 10 [8‐12] vs. 10 [10‐12] cmH₂O, p < 0.001). In datasets undergoing VCAV, tidal volume was likely to preset higher in the severe ineffective triggering group (median [IQR] of 7.3 [6.8–8.0] vs. 7.5 [7.3–8.2] ml/kg IPW), but with no statistical significance (p = 0.095).

Severe ineffective triggering was diagnosed in 31 (31%) patients. Only a longer duration of ventilation before the inclusion was found in the severe ineffective triggering group (median [IQR]: 20 [16‐23] vs. 19 [15‐21] hours, p = 0.050). Duration of ventilation after the inclusion and length of stay in the hospital was significantly longer in patients with severe ineffective triggering (Additional file 1: Table S4).