Early prediction of extubation failure in patients with severe pneumonia: a retrospective cohort study

Pneumonia is a common respiratory infectious disease and is one of the top six causes of death [1]. Approximately 10% of hospitalized patients with pneumonia require intubation for invasive mechanical ventilation (IMV) due to respiratory failure, which is termed as severe pneumonia [2,3]. IMV provides essential life support and a treatment window by increasing arterial oxygenation, rebalancing ventilation/perfusion ratio, and decreasing respiratory workload, and eventually improves prognosis and reduces mortality. However, the long-term use of IMV may result in hazardous complications, such as ventilator-associated pneumonia, barotrauma, oxygen toxicity, and increased use of sedatives and cost [4,5]. Furthermore, it has been reported that prolonged weaning was associated with higher hospital mortality [6]. Therefore, early identification of pneumonia patients with risk of extubation failure is of great clinical significance because interventions might be planned and implemented when risk could be predicted and monitored.

Previous studies have found a variety of factors are associated with extubation failure, including: (1) demographics and clinical status, such as age, fluid balance, comorbidity burden, cardiac function, mental status, and severity of illness; (2) laboratory abnormalities, such as hypercapnia, serum albumin, and blood urea nitrogen (BUN); (3) respiratory mechanics, such as maximal inspiratory pressure and diaphragmatic function; and (4) treatment interventions, such as sedatives [5,7–9]. However, these predictors are identified from a cluster of diseases including pneumonia, chronic obstructive pulmonary disease, heart failure, and trauma. Moreover, pneumonia and acute respiratory distress syndrome (ARDS) have been elucidated to be independent predictors for extubation failure [9].

Based on the lack of specific investigations of predictors and risks for extubation failure predictors in patients with severe pneumonia, we retrospectively analyzed the data from intubated patients with severe pneumonia and aimed to identify the independent predictors for extubation failure.

From April 2014 to December 2015, medical records of the patients, who were admitted to Respiratory Intensive Care Unit (RICU) of West China Hospital, Sichuan University, were retrieved from the hospital inpatient database system. Study protocol was approved by the Institutional Ethical Committee for Clinical and Biomedical Research of West China Hospital, and all patients or their legal substitute decision makers were consented for publication of in-hospital data. All methods were performed in accordance with the relevant guidelines and regulations released by the Chinese National Institutes of Health and the Clinical Trial Center of West China Hospital.

Patients were eligibly included if they were diagnosed of severe pneumonia upon admission and needed IMV via oral intubation or tracheostomy for more than 24 h due to acute respiratory failure. Severe pneumonia was defined if patients met either one of the major criteria (acute respiratory failure requiring IMV and septic shock with need for vasopressors) or at least three minor criteria (respiratory rate ≥ 30 bpm, ratio of partial pressure of arterial oxygen to fraction of inspired oxygen (PaO₂/FiO₂) ≤ 250, BUN ≥ 20 mg/dl, white blood cell count < 4 × 10⁹/l, platelet count < 100 × 10⁹/l, body temperature < 36°C, multilobar infiltrates, confusion/disorientation, and hypotension requiring aggressive fluid resuscitation) [2]. The indication for IMV was evaluated based on the arterial blood gas (ABG) analysis indicating refractory hypoxemia with PaO₂ less than 50 mmHg although on oxygen supplement.

We excluded patients if their pneumonia was secondary to known pulmonary diseases such as acute exacerbation of chronic obstructive pulmonary disease (COPD), asthma, and pulmonary thromboembolism, or if they were in fatal and life-threatening comorbidities including, but not limited to, aggressive carcinoma with unstable clinical conditions and cerebral injury. Patients with immunosuppressive therapy were also excluded in our study.

The present study was designed as a retrospective cohort study, in which eligible patients were classified into extubation success and failure groups. Patients’ baseline demographics and clinical characteristics, laboratory measurements, imaging information, and treatment details at the time of ICU admission were compared between these two groups to identify potential independent risk factors for the prediction of extubation failure in patients with severe pneumonia.

Successful extubation was defined as removal of endotracheal tube for more than 72 h regardless of the need for sequential non-invasive positive pressure ventilation (NPPV); while extubation failure was defined as unavoidable reintubation within 72 h of extubation [10]. Failure of IMV discontinuation was not treated as extubation failure only if the patients failed the extubation afterward. All patients who fell into the extubation failure criteria were mechanically ventilated after the trial of extubation or ventilator discontinuation.

In general, mechanical ventilation was initiated with a tidal volume (V_T) of 8–10 ml/kg of predicted body weight and a minimal positive end-expiratory pressure (PEEP), which was titrated with FiO₂ to target an oxygen saturation of pulse oximetry (SpO₂) ≥ 90% or PaO₂ ≥ 60 mmHg using pressure-controlled or volume-controlled assist/control mode [11,12]. However, “lung protective ventilation” and “open-lung approach” were given if patients developed ARDS, in which the patients’ lungs were protected by a lower V_T of 4–8 ml/kg of predicted body weight with a plateau airway pressure ≤ 30 cm H₂O and the collapsed alveoli were opened by intermittent recruitment maneuvers and maintained open by a PEEP of 2 cm H₂O higher than pressures at which an abrupt increment was detected in the upward slope of the pressure–volume curve [11,12]. Respiratory therapists up- or down-regulated the support of ventilation based on the ABG analysis every day.

Empirical antibiotics treatment was administered in accordance with the consensus by the Infectious Diseases Society of America and American Thoracic Society, and reasonable changes were made once a specific pathogen was identified and considered as clinical importance [2]. All patients were sedated with midazolam and (or) propofol on the basis of analgesia with fentanyl to meet a goal of Richmond Agitation-Sedation Scale between -2 and 0 and a target of critical care pain observation tool < 3. Daily interruption by temporarily ceasing the sedative infusions was also implemented to avoid potential delirium, which was defined if Richmond Agitation-Sedation Scale was between −3 and +4 and Confusion Assessment Method for Intensive Care Unit was positive [13,14]. Blood glucose level was measured every 4 h routinely and insulin dosing was commenced when two consecutive blood glucose levels are greater than 10 mmol/l with a target upper blood glucose ≤ 10 mmol/l [15]. Other treatments including but not limited to organ function supports, airway clearance by bedside bronchoscopy, circulation and nutrition maintenance, as well as internal environment homeostasis stabilization were also performed when necessary according to patients’ clinical conditions.

Spontaneous breathing trial (SBT) was conducted by respiratory therapists if the following criteria were met: (1) the causes of mechanical ventilation were relieved or alleviated; (2) PaO₂/FiO₂ > 150–200 at FiO₂ ≤ 0.4–0.5 and PEEP ≤ 5–8 cm H₂O; (3) stable hemodynamics without need for vasopressors or dopamine/dobutamine <5–10 μg/kg/min; and (4) the capability of spontaneous breathing. [16]. During SBT, a low level of pressure support between 5 and 7 cm H₂O was used for 30 min, and patients were considered as readiness for extubation if they have no following conditions: (1) agitation, anxiety, depressed mental status, and diaphoresis; (2) evidence of increased accessory muscle activity; (3) PaO₂ ≤ 50–60 mmHg at FiO₂ ≥ 0.5 or arterial oxygen saturation (SaO₂) < 90%; (4) partial pressure of arterial carbon dioxide (PaCO₂) > 50 mmHg or an increase of PaCO₂ > 8 mmHg; (5) pH < 7.32 or a decrease of pH ≥ 0.07 pH units; (6) respiratory rate (RR) > 35 breaths per minute (bpm) or increased by ≥ 50%; (7) rapid shallow breathing index (RSBI = ratio of RR to V_T) > 105 bpm; (8) heart rate > 140 beats per minute or increased by ≥20%; (9) systolic blood pressure < 90 mmHg or > 180 mmHg or increased by ≥ 20%; and (10) cardiac arrhythmias [17]. Otherwise, the previous ventilation mode or a non-fatiguing mode would be switched back and any possible reversible etiologies for failure would also be reviewed before the next SBT every day.

Patients’ demographics including age, gender, body mass index (BMI), and smoking and disease histories were recorded. The disease severities were assessed upon the admission to ICU by different scoring systems including acute physiology and chronic health evaluation II (APACHE II) scores, multiple organ dysfunction (Marshall) scores, sequential organ failure assessment (SOFA) scores, lung injury scores (LIS), and gas exchange, organ failure, cause, and associated disease (GOCA) scores. APACHE II is a comprehensive scoring system to identify acutely ill patients with high risk of death based on 14 parameters such as body temperature, mean arterial pressure, RR, PaO₂, ABG, and mental status [18]. Marshall and SOFA scores are used to evaluate the organ function or rate of failure, which consists of 4 or 6 items involving cardiovascular, hepatic, coagulation, renal, respiratory, and neurological systems [19]. LIS scoring system specifically stratifies the extent of lung injury into three categories from no lung injury to severe lung injury by the infiltration area on chest X-ray, PaO₂/FiO₂, PEEP, and static compliance of lung [20]. Similar to LIS, GOCA score was developed for patients with ARDS to define the disease but not to predict prognosis [21].

Clinical conditions as well as the laboratory measurements at the time of ICU admission rather than extubation were recorded, including vital signs, ABG, blood cell count, blood electrolytes, myocardial, and inflammatory biomarkers. The worst (highest) level of blood glucose measured within 24 h of ICU admission was recorded. Pathogenic microorganisms were generally screened in sputum (including bronchoalveolar lavage fluid, BALF), blood, and urine by culture or chemiluminescence. We also recorded details of treatment during ICU such as mechanical ventilation settings, types and dose of analgesics and sedatives, usage of vasopressors and muscle relaxants, as well as component and amount of blood transfusions.

Statistical analysis was performed by SPSS 21.0 [Copyright (c) SPSS Inc. 1989-2007]. Continuous data were presented as mean ± standard deviation (SD), while dichotomous data were reported as frequency and proportion. Independent Student’s t-test and Chi-square test or Fisher’s exact test were performed between extubation success and failure groups. Variables with two-sided P value of < 0.05 from the independent Student’s t-test and Chi-square test were individually analyzed by univariate logistic regression model, and were assessed for potential correlations and multicollinearity among individual variables by correlation matrixs and variance inflation factors (VIF). Thereafter, significant predictors with P value of < 0.10 were further analyzed in the multivariate logistic regression to identify the independent risk factors associated with extubation failure, and the adjusted odds ratio (OR) and 95% confidence interval (CI) were calculated. Receiver operating characteristic (ROC) curves were conducted to depict the area under the curve (AUC) for the accuracy of each independent risk factor in recognizing extubation failure, and cutoff points together with sensitivity, specificity, positive and negative predictive value (PPV and NPV), and likelihood ratio (LR) were also calculated by maximal Youden index (= sensitivity + specificity − 1).

A total of 666 patients with RICU admission were found, of which 509 patients were excluded according to the pre-defined exclusion criteria and 32 patients were excluded for not using IMV. Finally, 125 patients with severe pneumonia were included and divided into extubation success (n = 82) and failure (n = 43) groups (Figure 1). The rate of extubation failure was as high as 34.4% in patients with severe pneumonia, and the total mortality reached up to 58.4% (n = 73). The average days to extubation were 13.4 days.

Tables 1 and 2 summarized patients’ baseline characteristics and clinical and laboratory measurements, respectively. Compared with patients with extubation success, patients with extubation failure were associated with significantly higher mortality (97.7% vs. 37.8%, P < 0.001), but no significant difference was found in ICU stay (14.42 ± 12.87 vs. 18.29 ± 15.67 days, P = 0.166).

In the group of extubation failure, we found significantly higher APACHE II score (18.30 ± 4.23 vs. 16.30 ± 4.87, P = 0.025) and blood glucose (10.47 ± 5.24 vs. 8.93 ± 3.96 mmol/l, P = 0.038), but significantly lower pH (7.350 ± 0.105 vs. 7.383 ± 0.069, P = 0.039) and plasma complement 3 level (0.7309 ± 0.2959 vs. 0.8550 ± 0.2253 g/l, P = 0.035) compared with the group of extubation success. Significant increase was also found in LIS score, GOCA score, RR, incidence of new-onset septic shock, total bilirubin (TB), activated partial thromboplastin time (APTT), and brain natriuretic peptide (BNP).

Acinetobacter baumannii was the most commonly detected bacterium in sputum (including BALF), blood and urine, while Candida albicans were the most common fungi identified in sputum (including BALF) and urine (Table 3). Pseudomonas aeruginosa and Candida tropicalis could also be easily found in sputum (including BALF) but not in blood or urine. Herpes simplex virus predominated the causes for severe viral pneumonia, followed by Epstein–Barr virus. However, we did not find any significant differences in each pathogenic microorganism from sputum (including BALF), blood and urine between the extubation success and failure.

Compared with patients with extubation success, significantly higher PEEP (9 ± 4 vs. 7 ± 3 cmH₂O, P = 0.021) and FiO₂ (80 ± 22 vs. 72 ± 22%, P = 0.037) during IMV were needed and more midazolam (73.08 ± 55.17 vs. 48.98 ± 48.08 mg/d, P = 0.013) and fentanyl (0.93 ± 0.48 vs. 0.75 ± 0.45 mg/d, P = 0.043) were used in patients with extubation failure (Table 4). Additionally, significantly more patients used vasopressors (86.0% vs. 62.2%, P = 0.006) and needed transfusion of red blood cells (RBC) (44.2% vs. 28.0%, P = 0.040) in patients with extubation failure. The use of muscle relaxants, corticosteroids, diuretics, and insulin were similar in the two groups.

Correlation matrix analysis showed significant correlations among the potential risk factors for extubation failure identified previously except for APACHE II, EF, blood glucose, dose of fentanyl, and percentage of patients using RBC transfusion, and variance inflation factors for these parameters showed no multicollinearity between each other (VIF ranged from 1.011 to 1.037) (data not shown). Multivariate logistic regression found APACHE II score (OR 1.141, 95% CI 1.022–1.273, P = 0.019), blood glucose (OR 1.122, 95% CI 1.008–1.249, P = 0.035), cumulative dose of fentanyl (OR 3.010, 95% CI 1.100-8.237, P = 0.032), and percentage of patients using RBC transfusion (OR 2.774, 95% CI 1.062–7.252, P = 0.037) were independent risk factors for predicting extubation failure but not EF (OR 0.965, 95%CI 0.925–1.007, P = 0.097) (Table 5).

Figure 2 depicted the ROC curves for APACHE II score, blood glucose, and dose of fentanyl, in which all these factors showed acceptable capacity in the identification of extubation failure in patients with severe pneumonia (APACHE II score: AUC 0.644, 95% CI 0.543–0.745, P = 0.008; blood glucose: AUC 0.585, 95% CI 0.479–0.690, P = 0.020; dose of fentanyl: AUC 0.611, 95% CI 0.504–0.717, P = 0.043). A cutoff point of 17.5 for APACHE II score, 9.87 mmol/l for blood glucose, and 1.135 mg/d for fentanyl usage resulted in a sensitivity of 62.79%, 41.86% and 37.21%, and a specificity of 69.51%, 71.95% and 85.37%, respectively, in identifying severe pneumonia patients with high risk of extubation failure, as well as a good negative predictive value (NPV: 78.08%, 70.24%, and 72.16%) (Table 6).

It has been recognized that extubation failure is associated with increased mortality, longer ICU and hospital stays, and more nosocomial pneumonia [22,23]. In our study, we also demonstrated that patients with extubation failure had higher risk of death but not longer ICU stays, which mainly resulted from the death of these patients before they met the readiness of extubation. Due to the high incidence of extubation failure as well as the subsequent association of mortality, we are intrigued to explore potential factors associated with extubation failure to improve patients’ outcomes by optimizing treatment strategies and ventilator settings.

APACHE II score is the most commonly used tool to evaluate disease severity in ICU [18,24]. It has also been reported to have potential predictive values in extubation [25]. In our study, we also found that APACHE II score is an independent factor for predicting extubation failure from IMV although with a relatively low sensitivity (62.79%) and specificity (69.51%) at a cut-off point of greater than 17.5. However, we did not find such power of SOFA or Marshall score, which was not consistent with the previous finding that a high SOFA score on day 21 of IMV was associated with extubation and weaning failure in patients requiring prolonged IMV [26]. These differences might be because: (1) the different diseases for mechanical ventilation in different studies. Our study specifically focused on patients with severe pneumonia; (2) the offset effect of APACHE II score on the significance of SOFA as the items of SOFA are also included in APACHE II score. In terms of the LIS and GOCA scores, both of them were shown to be higher in patients with extubation failure, but they were not considered for multivariate logistic regression due to the significant correlations with other variables. Furthermore, LIS and GOCA are designed particularly for ARDS development, and our study did not find significant difference in ARDS development between patients with extubation failure and success.

Hyperglycemia is a common phenomenon in critically ill patients resulted from the development of hyperglycemia or insulin resistance [27,28]. It is found to be associated with increased mortality and adverse events such as postoperative infection and critical-illness polyneuropathy [29]. Studies have demonstrated that lowered blood glucose level rather than the insulin dose could bring beneficial effects [30,31], and the blood glucose control was also recognized as an independent predictor for prolonged mechanical ventilation [32,33] In our study, we also found that blood glucose of greater than 9.87 mmol/l was associated with higher risk of extubation failure, which might be due to the normal immune function maintenance, systematic inflammation reduction, endothelium and mitochondrial ultrastructure production, and skeletal muscle strength improvement after the recovery of normal glucose uptake [34,35].

Overuse and accumulation of opioid agents have been elucidated to be associated with increased adverse events, such as respiratory depression, iatrogenic withdrawal syndrome, and prolongation of time to extubation and weaning [36,37]. In our study, we also found that the use of both fentanyl and midazolam were significantly increased in patients with extubation failure, and the mean dose of fentanyl >1.135 mg/d was an independent predictor in predicting extubation failure. However, midazolam was not further assessed for potential risk factors for extubation failure due to the significant correlation in the correlation matrix although it has been found that unnecessary provision of sedatives is also associated with increased ventilation days [38]. Therefore, further studies are warranted to adjust the interdependence and interruption of sedatives such as midazolam before the conclusive consensus being made.

Previous studies found that increase of blood hemoglobin level rather than RBC transfusion treatment could improve extubation success [39–41]. Moreover, evidence also showed that hemoglobin breakdown products (heme, iron) could induce pro-inflammatory responses through various cellular signals even in the process of acute lung injury and ARDS [42], which suggests that decrease of hemoglobin level might indicate severe lung inflammation. In our study, we found similar baseline hemoglobin levels in the two patient groups, but more patients received RBC transfusion in the group with extubation failure; this implicates that the patients in extubation failure group had severer lung inflammation due to the more need for RBC transfusion, which further explained our finding that RBC transfusion was associated with extubation failure and served as an independent predictor.

Potential limitations in our study included: (1) sample size was relatively low, which might result in inaccuracy or underestimation of the power of the four independent risk factors and missing of other potential risk factors, especially the previously reported risk factor of PaO₂/FiO₂ for respiratory failure and outcomes, which was calculated to be at the cutoff of a statistical significance; (2) significant correlations were found among the factors identified in univariate logistic regression, which might also miss some factors of potential predictive values; and (3) the study was designed as a retrospective cohort, which disenabled us to (a) evaluate some other critical factors such as RSBI, static lung compliance, and maximal inspiratory pressure, (b) further perform subgroup analysis of different categories of extubation failure such as simple extubation, difficult extubation, and prolonged extubation, and (c) analyze the correlation between the trend of these clinical and treatment outcomes and extubation failure, which might be of greater clinical importance.

In patients with severe pneumonia, APACHE II score of greater than 17.5, blood glucose of greater than 9.87 mmol/l, fentanyl usage of greater than 1.135 mg/day, and the need for RBC transfusion at the time of weaning might potentially discriminate the patients with high risk of extubation failure. Further prospective studies with large patient populations are necessitated to validate more factors, especially the ones without potential correlations, to confirm and determine the exact risk factors, as well as to establish a comprehensive scoring system or algorism for better predicting extubation failure.

The authors declare that there are no competing interests associated with the manuscript.

This work was supported by the Sichuan Science and Technology Agency Grant [grant numbers 2017JY0271 and 2015JY0176]; and the Ministry of Science and Technology of the People’s Republic of China [grant number 2015BAI12B10].

He Yu substantially contributed to the conception and design of the present study, analysis and interpretation of data, and drafting and revising for important intellectual content. Jian Luo substantially contributed to the conception and design of the present study, acquisition, analysis and interpretation of data, and critically revising for important intellectual content. Yuenan Ni acquired and analyzed data in the present study. Yuehong Hu acquired and analyzed data in the present study. Dan Liu acquired and analyzed data in the present study. Maoyun Wang interpreted data in the present study and revised important intellectual content. Binmiao Liang substantially contributed to the conception and design of the present study, revising critically for important intellectual content, and final approval of the version to be published. Zongan Liang interpreted data in the present study and revised important intellectual content.

ABG

arterial blood gas

APACHE II

acute physiology and chronic health evaluation II

APTT

activated partial thromboplastin time

ARDS

acute respiratory distress syndrome

AUC

area under the curve

BALF

bronchoalveolar lavage fluid

BMI

body mass index

BNP

brain natriuretic peptide

BUN

blood urea nitrogen

CI

confidential interval

COPD

chronic obstructive pulmonary disease

EF

ejection fraction

GOCA

gas exchange, organ failure, cause, associated disease

HCO₃⁻

bicarbonate ion

IMV

invasive mechanical ventilation

LIS

lung injury score

LR

likelihood ratio

NPPV

noninvasive positive pressure ventilation

NPV

negative predictive value

OR

odds ratio

PaCO₂

partial pressure of arterial carbon dioxide

PaO₂/FiO₂

ratio of partial pressure of arterial oxygen to fraction of inspired oxygen

PEEP

positive end-expiratory pressure

PPV

positive predictive value

RBC

red blood cell

RICU

respiratory and infectious intensive care unit

ROC

receiver operating characteristic

RR

respiratory rate

RSBI

rapid shallow breathing index

SaO₂

arterial oxygen saturation

SBT

spontaneous breathing trial

SD

standard deviation

SOFA

sequential organ failure assessment

SpO₂

oxygen saturation of pulse oximetry

TB

total bilirubin

VIF

variance inflation factors

V_T

tidal volume