Successful, timely weaning and tracheal extubation of critically ill patients has a considerable bearing on ultimate outcome. The tolerance of a spontaneous breathing trial (SBT) is an evidence-based strategy to predict successful liberation from mechanical ventilation.1 However, even after a successful SBT, the possibility of an extubation failure is still 20% or more.2,3 Accordingly, extensive efforts have been made to identify predictors of successful extubation or weaning. Of the many parameters suggested for the prediction, only the rapid shallow breathing index (RSBI) and airway occlusion pressure (P0.1) were used widely and proposed by the Collective Task Force of the American College of Chest Physicians as promising weaning predictors.4 However, none of the predictors has demonstrated more than modest accuracy in predicting extubation outcome, and results from various studies are conflicting.3–6
There are multiple explanations for the low accuracy. First, the study population with respect to age differs from study to study. Previous studies have shown that the predictive power of weaning indices should be investigated separately for different patient populations, such as the elderly, to improve accuracy.7,8 Second, previous studies of weaning prediction have been mostly confined to static indices,4,9 which are measured on a one-time basis. Therefore, the dynamic weaning process, during which the physiologic variables (i.e., RSBI) are changing continuously, cannot be captured by a single static measure.10,11 Segal et al.12 investigated the evolution of respiratory pattern, showing that change in RSBI at 30 minutes of the SBT may improve the ability to predict extubation outcome. However, the low negative predictive value (NPV) reported in the study does not make it an excellent predictor for physicians making extubation decisions. In fact, weaning failure is due to several factors,1,9,13 which cannot be reflected by the results of a single test.
We hypothesized that a clinical decision rule, which integrates a number of physiological functions by including multiple variables, may reliably identify a patient's risk of successful extubation (ES) or failed extubation (EF). In the present study, a decision-tree model was constructed by the classification and regressive tree (CART) algorithm to develop a simple prediction rule for physicians as guidance in making extubation decisions.
Patients and Setting
A prospective observational cohort study was conducted in a 16-bed medical intensive care unit (ICU) of a university-affiliated teaching hospital between October 2007 and October 2008. The local IRB approved the study, and informed patient consent was waived because of the observational nature of the study.
Patients were enrolled if they met the following criteria1: improvement in the underlying conditions; adequate oxygenation, indicated by PaO2 >60 mm Hg at FIO2 of ≤0.4 with an extrinsic positive end-expiratory pressure (PEEPe) <7 cm H2O; cardiovascular stability (heart rate [HR] <130 beats per minute [bpm] and absence of vasopressor use); body temperature <38°C; adequate hemoglobin >8 g/dL; fully awake; effective cough strength on command; and normal acid base and electrolyte condition. Moreover, the patients enrolled were ≥65 years of age and their lungs mechanically ventilated for >48 hours. We excluded patients with a history of neurologic diseases, those who failed SBT, and those tracheostomized. Lack of available ventilators equipped with P0.1 measurement was also an exclusion criterion. At the time of enrollment, all patients' lungs were ventilated with an Evita-4 or XL ventilator (Draeger, Lubeck, Germany), with a pressure support below 15 cm H2O (10.3 ± 2.1 cm H2O), a PEEPe of 5 cm H2O and FiO2 below 0.5 (0.39 ± 0.03).
Patients eligible for enrollment, in a semirecumbent position, were submitted to 60-minute SBT immediately. Endotracheal suctioning was performed before SBT. During the SBT, the patients were allowed to breathe through the ventilatory circuit by using flow triggering (2 L/min) with automatic tube compensation (ATC) of 100% and 5 cm H2O PEEPe. FiO2 was set to the same value used during mechanical ventilation. The tolerance to SBT was considered poor in the presence of at least one of the following criteria: a decrease in oxygen saturation to <90% while requiring an FiO2 >0.5; evidence of respiratory distress; sustained increase in HR (>140 bpm or >20% baseline); systolic blood pressure >200 mm Hg or <90 mm Hg; uncoordinated thoraco-abdominal movement; and agitation or depressed mental status. If the 60-minute SBT was clinically well tolerated, patients were extubated immediately. Patients not tolerating the SBT were reconnected to the ventilator. The decisions to reconnect the patients were made by the primary physicians who were blind to the results of the measurements performed during the SBT.
Patients were classified into ES and EF groups. ES was defined as the ability to maintain spontaneous unassisted breathing for >48 hours after extubation. Cases of EF included patients who died within the 48 hours after extubation and those requiring reintubation within 48 hours after extubation. Noninvasive mechanical ventilation was not used in the 48-hour postextubation period to avoid confounding factors.
Electrocardiogram, arterial blood pressure, HR, and SpO2 were continuously monitored. The following variables were recorded before the SBT: demographic data, APACHE (acute physiology and chronic health evaluation) II score, reasons for mechanical ventilation (MV), days receiving MV before SBT, ventilatory variables, arterial PaCO2 and PaO2/FiO2. The respiratory variables—including minute ventilation (VE), respiratory rate (f), tidal volume (VT), and P0.1—were recorded at the first minute, 30th minute, and 60th minute of the SBT. Values were displayed on the ventilator. Measurements were performed by 1 investigator and were repeated 3 times separated by an interval of not <15 s; mean values were used for data analysis. The RSBI (f/VT, breaths/min/L) and P0.1 × RSBI (cm H2O*breaths/min/L) were calculated at the first minute (RSBI1, P0.1 × RSBI1), 30th minute (RSBI30, P0.1 × RSBI30), and the 60th minute of SBT (RSBI60, P0.1 × RSBI60), respectively. Endotracheal tube suction was performed 5 minutes before each measurement.
The change in RSBI was expressed by the ratio of RSBI at a given time point to RSBI1 as a percentage. Two predefined variables were evaluated: (1) change of RSBI at the 30th minute of SBT (▵RSBI30) as assessed by ratio of the RSBI30 to the RSBI1 as a percentage and (2) change of RSBI at the 60th minute of SBT (▵RSBI60) as assessed by ratio of the RSBI60 to the RSBI1 as a percentage.
The statistical analysis was performed using the statistical software package SPSS 16.0 (SPSS, Chicago, Illinois) unless otherwise indicated. For data normally distributed, the results were expressed as the mean and SD, whereas for data nonnormally distributed, the results were presented as median and 25th and 75th percentiles. We used the t test or the Mann–Whitney U test to compare continuous variables, as appropriate. Two-tailed P values of <0.05 were used to indicate statistical significance. The comparisons of variables at different time points during the SBT were made by analysis of variance (ANOVA) for repeated measures.
Standard formulas were used to calculate the sensitivity, specificity, positive predictive value (PPV), NPV, and diagnostic accuracy (DA) of each index in predicting ES.10True positive was defined as when a test predicted ES (i.e., RSBI < threshold value) and extubation succeeded; true negative was defined as when a test predicted EF (i.e., RSBI > threshold value) and extubation failed. The threshold values were determined as the value that provided the best sensitivity and specificity by using the receiver operating characteristic (ROC) curve, which was performed with MedCalc software version 22.214.171.124 (2010 MedCalc Software bvba, Mariakerke, Belgium). The predictive performance of each index was further evaluated by the area under the ROC curve (AUC) estimated using the nonparametric approach described by DeLong et al.14
CART analysis was performed with the statistical environment S-PLUS 8.0 (Insightful Corporation, Seattle, Washington) and Rpart package. Variables found to differ significantly between the 2 groups (ES and EF) were entered in a CART model. In the tree-growing process, every value of each of the variables entered in the model is considered a potential “splitter,” and the one that leads to the greatest reduction in impurity is chosen as the best splitter. The Gini Index was used to quantify the impurity in each node. For a binary (0/1) target, the Gini index of a node t is G(t) = 1 − p(t)2 − [1 − p(t)]2, where p(t) is the relative frequency of class 1 (1 = successful extubation) in the node. The loss function used in our classification tree construction was misclassification rate. The tree-building process continued until it was impossible to continue. The process was stopped when (1) the number of cases reaching each leaf was <10; or (2) the number of observations in a terminal node was <2; or (3) the leaf was homogeneous enough. The tree constructed by this process was usually overgrown and overfitted. Therefore, the original overgrown tree was pruned to the optimal one having the minimum 5-fold cross-validation error. The resulting pruned tree was then validated by the 5-fold cross-validation, which was introduced as a resampling method having small bias for classification trees in a small sample.15,16
Of the 113 patients enrolled in the study, 22 (19.5%) failed the SBT and were not included in the final analysis. Among the 22 patients who failed the SBT, 20 did so during the first 30 minutes of the SBT, and only 2 showed clinical intolerance to the SBT during the last 30 minutes. Of 91 patients who tolerated the SBT, 73 (80.2%) were successfully extubated, and 18 (19.8%) required reintubation within 48 hours after extubation. No patients died within 48 hours after the extubation. The causes for reintubation were hypoxemia in 5 cases, hypercapnia in 6, congestive heart failure in 5, and excessive respiratory work in 2. No subject was reintubated because of upper airway obstruction or insufficient cough. Ten patients were reintubated within 24 hours, and 8 between 24 and 48 hours. Demographic characteristics of the 91 patients are shown in Table 1.
Respiratory data obtained at the 3 time points (at first, 30th, and 60th minutes of the SBT) are presented in Table 2. At all 3 time points, the ES group demonstrated a lower RSBI, P0.1, P0.1 × RSBI, in comparison with the EF group. Figure 1 illustrates the time course of RSBI during the SBT. A significant increase in RSBI was observed during the SBT in the EF group (F = 9.36, P = 0.007), whereas RSBI decreased significantly in the ES group (F = 18.2, P < 0.001). At the 30th minute, the ▵RSBI30 was significantly higher in the EF group (118% ± 34%) than in the ES group (97% ± 35%) (P = 0.01).
The ROC curves were plotted for RSBI, P0.1 × RSBI, and ▵RSBI30 to illustrate their ability to predict extubation outcome (Fig. 2). Neither RSBI nor ▵RSBI30 could reliably predict extubation outcome, as is suggested by the low values of AUC and NPV. With a threshold of <98%, ▵RSBI30 performed poorly in predicting ES (sensitivity = 68.5%, specificity = 83.3%, PPV = 94.3%, NPV = 39.5%). P0.1 × RSBI30 ≤328 presented moderate accuracy in predicting ES (sensitivity = 89.0%, specificity = 88.9%, PPV = 97.0%, NPV = 66.7%, DA = 89.1%; Table 3).
The results of the CART analysis are shown in Figure 3. Of the variables found to be statistically different between the 2 groups (Tables 2), the CART analysis selected 3 indices that allowed the splitting of the 2 initial groups (success and failure). The principal discriminator was P0.1 × RSBI30. Further partitioning was based either on RSBI1 in patients with P0.1 × RSBI30 ranging from 328 to 474 or on ▵RSBI30 in patients with P0.1 × RSBI30 >474. Indeed, for patients with P0.1 × RSBI30 >474, ▵RSBI30 >98% defined a group including all EF patients but no ES patients, and ▵RSBI30 ≤98% included all ES patients with no EF patients. For patients with P0.1 × RSBI30 ≤474, the combination of both a P0.1 × RSBI30 >328 and RSBI1 >112 defined a group including all ES patients but no EF patients. Furthermore, the DA of the tree model, which was 89.1% with only the P0.1 × RSBI30 included, increased to 94.5% when both P0.1 × RSBI30 and ▵RSBI30 were included. The final tree model with the inclusion of all 3 discriminators can capture ES with a DA of 96.7%, and AUC of 0.94 (95% confidence interval [CI]: 0.87 to 0.98).
In this prospective observational cohort study conducted in medical ICU patients, 3 indices were identified by CART to be responsible for ES in elderly patients after a successful SBT. Using CART analysis, we have constructed an adequately explicit bedside clinical decision rule that can predict ES with an accuracy of 96.7% and AUC of 0.94.
Generally, weaning failure was defined as either the failure of a SBT or the failure of extubation after a successful SBT. However, from our point of view, it is less important to predict a successful SBT, because SBT is safe, well-tolerated, and easy to perform. Thus, the aim of our study was to identify the indices predictive of ES after successful SBT.
It is important to emphasize that RSBI was not used as an inclusion criterion in the present study; thus test referral and spectrum bias were excluded.10 Indeed, using RSBI <105 as a criterion for entry into a study tends to exclude patients with rapid shallow breathing, resulting in spectrum bias or test-referral bias.10,17 In certain extreme conditions, the test-referral bias resulted in as little as 3% of a clinically relevant population being included in a study,10 accounting at least partly for the low specificities of weaning indices reported in previous studies.18,19
Consistent with previous studies,12,20,21 we demonstrated poor predictive performance of RSBI, even when measured at multiple time points. In the present study, the RSBI30 <105 had a sensitivity of 89%, specificity of 72.2%, PPV of 92.1%, and NPV of 61.9%. We observed that 8 patients with a RSBI-30 >105 were safely extubated. In a recent study that included 73 patients, Teixeira et al.21 demonstrated that RSBI at any given time point was not predictive of ES. Previous studies have evaluated the performance of P0.1 and P0.1 × RSBI as extubation indices, showing P0.1 × RSBI to be a good predictor of ES. Sassoon and Mahutte5 studied 45 male patients, showing that P0.1 × RSBI <450 had the sensitivity of 0.97 and specificity of 0.60 in predicting successful weaning. Nemer et al.22 found P0.1 <4.2 cm H2O to be the moderately accurate predictor of ES with a sensitivity of 86.5%, specificity of 61.1%, DA of 80%, and AUC of 0.76. In our study, P0.1 × RSBI30 ≤328 had moderate accuracy in predicting successful extubation with a sensitivity of 89%, specificity of 88.9%, and AUC of 0.89.
However, weaning is usually delayed by clinicians,13 and a weaning index should be more predictive of EF (for example, it should be more specific, and have a good NPV). In our study, we observed that 8 patients, even with a P0.1 × RSBI30 value >328, were safely extubated because of the good tolerance to SBT. Thus, using either P0.1 × RSBI-30 ≤328 or RSBI ≤105 as a predictor tends to leave some patients on a ventilator for a longer period of time than is necessary. Because of our results and those from other recent studies,6,18,22 we suggest that neither RSBI nor P0.1 × RSBI, even measured at multiple time points during the SBT, should be used in clinical decision making.
The low predictive performances of the evaluated static indices (RSBI, P0.1 × RSBI) could be explained by the fact that they were measured on a one-time basis, whereas weaning is a dynamic process during which extubation failures display a more severe deterioration in lung mechanics than do successes.10,23 We hypothesize that a change of RSBI during SBT, which may reflect the patient's increased effort to cope with the increase in mechanical load and capture the change in a patient's condition over time,23,24 may have improved accuracy in predicting extubation outcome. In this prospective cohort study, we found that although the ▵RSBI30 was significantly higher in the EF group (118% ± 34%) than in the ES group (93% ± 35%, P = 0.01), the predictive performance of the index was not as good as was expected with AUC of 0.76 and NPV of only 39.5%. Our findings were supported by Segal et al.'s study,12 in which the authors demonstrated that the RSBI increase (<5% as the threshold) at the 30th minute of SBT could predict ES with a sensitivity of 83%, specificity of 78%, PPV of 96%, NPV of 38%, and AUC of 0.83. Indeed, the low NPV values reported in both Segal et al.'s study and our study would make it inappropriate as a predictor of extubation outcome in clinical practice.
It should be borne in mind that a single test can examine only one aspect of physiological function, failing to reflect all the pathophysiological processes affecting extubation outcome.9,10 In this context, we attempted to construct a decision-tree model, which included multiple variables and considered the changes of these variables, to predict the extubation outcome more accurately. CART analysis was chosen, because it can handle large numbers of variables without any assumptions regarding their distribution or their nature.25,26 The CART selected 3 variables to develop a simple predictive model that could predict ES with an AUC of 0.94. The fact that all included variables are readily available in most ICUs enhances the clinical applicability of the decision tree considerably.
In the decision-tree model, patients were first divided into different risk groups for EF according to P0.1 × RSBI values at the 30th minute. We found that the ΔRSBI30, with the threshold of <98%, has the best predictive power in patients with a high risk of EF (defined by a P0.1 × RSBI30 >474), splitting patients in this subgroup into 2 zero-overlap categories (success and failure; Fig. 3). However, ΔRSBI30 was not selected by CART as discriminative in patients with a low risk of EF (defined by a P0.1 × RSBI30 ≤328), which is not surprising in view of the highly predictive performance of the SBT in this subgroup. As is suggested by our decision tree, it is important to emphasize that different predictors should be selected for different subgroups of patients.
The high predictive power of the decision-tree model in our study may also be explained by the fact that none of the EFs were reintubated because of upper airway obstruction or insufficient cough. Consistent with our study, previous studies also reported the low frequency of weak cough, abundant amount of secretions, or both as the causes of reintubation in EFs.3,12 Indeed, the model developed in our study did not consider the cough strength and was not designed to identify EFs because of factors of airway incompetence, such as mucus plugging, aspiration, and weak cough. Therefore, it is not surprising that low predictive value was observed in certain previous studies,18,21,27 when weaning indices (such as RSBI and P0.1), failing to consider the factors of airway competence, were used in predicting EF because of airway incompetence.
There are several limitations to our study that should be mentioned. First, a selection bias may have been present because the use of ATC could potentially allow more marginal patients to tolerate the SBT, who would then develop ventilatory failure after extubation. However, the accuracy of ATC in commercially available mechanical ventilators, such as are used in our study, has been formally assessed by Elsasser et al.28 They found that the built-in commercial ATC may provide adequate inspiratory tube compensation with little, if any, overassistance. Second, a further selection bias might have been introduced by the clinical decision to stop the SBT in the study. Indeed, using objective criterion (i.e., RSBI >105) for stopping the SBT tends to exclude patients with rapid shallow breathing, leading to test referral and spectrum bias. However, in our cohort, the decision to stop the trial was protocolized, and made by the physicians in charge who were blind to the results of the measurements except the respiratory rate. Third, as identified by one of our reviewers, the reintubation criteria were not predefined in our study, which would also have introduced a selection bias. However, decisions regarding reintubation in our study were made by the primary physician using the standard policy. Moreover, the reintubation rate in our study (19.8%) was similar to those found in other studies, varying from 13.5% to 25%.2,19,29 It has been accepted that an EF rate of 15% to 20% implies an acceptable balance between performing premature extubation and unnecessarily prolonging MV.30 For these reasons, we believe that the impact of selection bias on the results was minimal. Although the optimal decision tree in our study was determined as what would result in the minimum cross-validation error, and the final tree was validated by 5-fold cross-validation, the small sample size of our study is a limitation. Indeed, some subgroups in our decision tree had only a small number of patients because of the low incidence of EF, and their identification may be the result of pure chance. Thus, our study should be interpreted as exploratory only. Indeed, before this model can be used in clinical decision making, it should be validated in an external population with a larger sample size.
In conclusion, in a prospective observational cohort of mechanically ventilated elderly patients passing a SBT, we demonstrated that none of the single-point weaning indices was powerful enough to predict ES. A decision-tree model constructed by the CART algorithm selected 3 discriminators (P0.1 × RSBI30, RSBI1, ▵RSBI30), predicting ES with an accuracy of 96.7% and AUC of 0.94. If the current model is confirmed by a prospective study with a larger sample size, it could be used for physicians to guide extubation decisions.
We thank the medical and the nursing staffs at the participating site for efficient help, and Yu-Ming Li, MD, PhD, for his advice on the study design. We are indebted to Mao-Ti Wei, PhD , for kindly providing statistical guidance. We also thank the reviewers for their careful reviews and insightful comments.
1. MacIntyre NR, Cook DJ, Ely EW Jr, Epstein SK, Fink JB, Heffner JE, Hess D, Hubmayer RD, Scheinhorn DJ. Evidence-based guidelines for weaning and discontinuing ventilatory support: a collective task force facilitated by the American College of Chest Physicians; the American Association for Respiratory Care; and the American College of Critical Care Med. Chest 2001;120:375S–95S
2. Esteban A, Alia I, Tobin MJ, Gil A, Gordo F, Vallverdu I, Blanch L, Bonet A, Vazquez A, de Pablo R, Torres A, de La Cal MA, Macias S. Effect of spontaneous breathing trial duration on outcome of attempts to discontinue mechanical ventilation. Spanish Lung Failure Collaborative group. Am J Respir Crit Care Med 1999;159:512–8
3. Cohen JD, Shapiro M, Grozovski E, Singer P. Automatic tube compensation-assisted respiratory rate to tidal volume ratio improves the prediction of weaning outcome. Chest 2002;122:980–4
4. Meade M, Guyatt G, Cook D, Griffith L, Sinuff T, Kergl C, Mancebo J, Esteban A, Epstein S. Predicting success in weaning from mechanical ventilation. Chest 2001;120:400S–24S
5. Sassoon CS, Mahutte CK. Airway occlusion pressure and breathing pattern as predictors of weaning outcome. Am Rev Respir Dis 1993;148:860–6
6. Conti G, Montini L, Pennisi MA, Cavaliere F, Arcangeli A, Bocci MG, Proietti R, Antonelli M. A prospective, blinded evaluation of indexes proposed to predict weaning from mechanical ventilation. Intensive Care Med 2004;30:830–6
7. Krieger BP. Respiratory failure in the elderly. Clin Geriatr Med 1994;10:103–19
8. Krieger BP, Isber J, Breitenbucher A, Throop G, Ershowsky P. Serial measurements of the rapid-shallow-breathing index as a predictor of weaning outcome in elderly medical patients. Chest 1997;112:1029–34
9. Yang KL, Tobin MJ. A prospective study of indexes predicting the outcome of trials of weaning from mechanical ventilation. N Engl J Med 1991;324:1445–50
10. Tobin MJ, Jubran A. Weaning from mechanical ventilation. In: Tobin MJ, ed. Principles and Practice of Mechanical Ventilation. 2nd ed. New York: McGraw-Hill, 2006: 1185–220
11. Jubran A, Grant BJ, Laghi F, Parthasarathy S, Tobin MJ. Weaning prediction: esophageal pressure monitoring complements readiness testing. Am J Respir Crit Care Med 2005;171:1252–9
12. Segal LN, Oei E, Oppenheimer BW, Goldring RM, Bustami RT, Ruggiero S, Berger KI, Fiel SB. Evolution of pattern of breathing during a spontaneous breathing trial predicts successful extubation. Intensive Care Med 2010;36:487–95
13. Epstein SK. Decision to extubate. Intensive Care Med 2002;28:535–46
14. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 1988;44:837–45
15. Molinaro AM, Simon R, Pfeiffer RM. Prediction error estimation: a comparison of resampling methods. Bioinformatics 2005;21:3301–7
16. Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning. New York: Springer, 2001
17. Tobin MJ, Jubran A. Meta-analysis under the spotlight: focused on a meta-analysis of ventilator weaning. Crit Care Med 2008;36:1–7
18. Fernandez R, Raurich JM, Mut T, Blanco J, Santos A, Villagra A. Extubation failure: diagnostic value of occlusion pressure (P0.1) and P0.1-derived parameters. Intensive Care Med 2004;30:234–40
19. Uusaro A, Chittock DR, Russell JA, Walley KR. Stress test and gastric-arterial PCO2 measurement improve prediction of successful extubation. Crit Care Med 2000;28:2313–9
20. Lee KH, Hui KP, Chan TB, Tan WC, Lim TK. Rapid shallow breathing (frequency–tidal volume ratio) did not predict extubation outcome. Chest 1994;105:540–3
21. Teixeira C, Zimermann Teixeira PJ, Hoher JA, de Leon PP, Brodt SF, da Siva Moreira J. Serial measurements of f/VT can predict extubation failure in patients with f/VT ≤105? J Crit Care 2008;23:572–6
22. Nemer SN, Barbas CS, Caldeira JB, Guimaraes B, Azeredo LM, Gago R, Souza PC. Evaluation of maximal inspiratory pressure, tracheal airway occlusion pressure, and its ratio in the weaning outcome. J Crit Care 2009;24:441–6
23. Jubran A, Tobin MJ. Pathophysiologic basis of acute respiratory distress in patients who fail a trial of weaning from mechanical ventilation. Am J Respir Crit Care Med 1997;155:906–15
24. Tobin MJ, Laghi F, Brochard L. Role of the respiratory muscles in acute respiratory failure of COPD: lessons from weaning failure. J Appl Physiol 2009;107:962–70
25. Muller R, Mockel M. Logistic regression and CART in the analysis of multimarker studies. Clin Chim Acta 2008;394:1–6
26. Trujillano J, Sarria-Santamera A, Esquerda A, Badia M, Palma M, March J. [Approach to the methodology of classification and regression trees]. Gac Sanit 2008;22:65–72
27. Frutos-Vivar F, Ferguson ND, Esteban A, Epstein SK, Arabi Y, Apezteguia C, Gonzalez M, Hill NS, Nava S, D'Empaire G, Anzueto A. Risk factors for extubation failure in patients following a successful spontaneous breathing trial. Chest 2006;130:1664–71
28. Elsasser S, Guttmann J, Stocker R, Mols G, Priebe HJ, Haberthur C. Accuracy of automatic tube compensation in new-generation mechanical ventilators. Crit Care Med 2003;31:2619–26
29. Esteban A, Alia I, Gordo F, Fernandez R, Solsona JF, Vallverdu I, Macias S, Allegue JM, Blanco J, Carriedo D, Leon M, de la Cal MA, Taboada F, Gonzalez de Velasco J, Palazon E, Carrizosa F, Tomas R, Suarez J, Goldwasser RS. Extubation outcome after spontaneous breathing trials with T-tube or pressure support ventilation. The Spanish Lung Failure Collaborative group. Am J Respir Crit Care Med 1997;156:459–65
30. Siner JM, Manthous CA. Liberation from mechanical ventilation: what monitoring matters? Crit Care Clin 2007;23:613–38