Acute exacerbations are common in patients with

Predicting the Result of Noninvasive Ventilation in Severe Acute Exacerbations of Patients With Chronic Airflow Limitation* Antonio Antón, MD; Rosa Güell, MD; Juan Gómez, MD; José Serrano, MD; Abilio Castellano, MD; Jose Luis Carrasco, BSC; and Joaquín Sanchis, MD Objective: To analyze prospectively the factors related to the success of noninvasive ventilation (NIV) in the treatment of acute exacerbations of chronic airflow limitation (CAFL) and to generate a multiple regression model in order to detect which patients can be successfully treated by this method. Setting: A respiratory medicine ward of a referral hospital. Methods and principal results: Initially, we examined 44 episodes of acute respiratory failure in 36 patients with CAFL in whom mechanical ventilation was advisable. In 34 of 44 episodes (77%), NIV was used successfully. Patients in whom NIV succeeded had a lower FEV 1 prior to admission, a higher level of consciousness (LC), and significant improvements in PaCO 2, ph, and LC after 1 h of NIV. A logistic regression model consisting of baseline FEV 1 and PaCO 2 values, initial PaCO 2, ph, and LC values on admission, and PaCO 2 values after1hofnivallowed us to correctly classify > 95% of the 44 episodes in which the outcome was successful. In the second part of the study, we prospectively validated the equation in another 15 consecutive CAFL patients with acute hypercapnic respiratory failure. NIV successfully treated 12 patients (80%), and the model correctly classified 14 patients (93%). Conclusion: Good LC at the beginning of NIV and improvements in ph, PaCO 2, and LC values after 1 h of NIV are associated with successful responses to NIV in COPD patients with acute hypercapnic respiratory failure. Our validated multiple regression model confirms that these variables predict the result of NIV in acute hypercapnic failure in CAFL patients. (CHEST 2000; 117:828 833) Key words: acute respiratory failure; chronic airflow limitation; noninvasive ventilation; pressure support ventilation Abbreviations: APACHE acute physiology and chronic health evaluation; CAFL chronic airflow limitation; LC level of consciousness; MV mechanical ventilation; NIV noninvasive ventilation; OTI oral-tracheal intubation; PSV pressure support ventilation Acute exacerbations are common in patients with advanced chronic airflow limitation (CAFL). Oral-tracheal intubation (OTI) and mechanical ven- For editorial comment see page 625 *From the Department of Respiratory Medicine (Drs. Antón, Güell, Gómez, Serrano, Castellano, and Sanchis), Hospital de la Santa Creu i Sant Pau, Universitat Autónoma de Barcelona, Barcelona, Spain; and the Department of Biostatistics (Mr. Carrasco), School of Medicine, Universitat de Barcelona, Barcelona, Spain. Manuscript received March 10, 1999; revision accepted September 30, 1999. Correspondence to: Antonio Antón, MD, Department of Respiratory Medicine, Hospital de la Santa Creu i Sant Pau, Sant Antoni Maria Claret 167, 08025 Barcelona, Spain; e-mail: panton@ hsp.santpau.es tilation (MV) are sometimes needed despite the associated risks. 1 5 To avoid such risks, interest in noninvasive ventilation (NIV), defined as a technique that assists ventilation without the need for an artificial airway, has increased in recent years. 6 12 Pressure support ventilation (PSV) with a face mask is a mode of NIV that provides ventilatory assistance during inspiration, allowing respiratory muscles to work less, increasing the volume inspired per minute, and improving arterial blood gas levels. 9,13 16 In patients in whom MV to treat severe exacerbations of COPD was advisable, NIV was used successfully to reduce rates of intubation and MV. 17,18 In some patients, however, NIV is inadequate, and OTI and invasive ventilation cannot be avoided. The proportion of such patients varies from study to study, 8,9 and, therefore, it seems important to de- 828 Clinical Investigations in Critical Care

velop a way to identify patients who can be treated successfully with NIV. The few studies dealing with possible factors that predict the outcome of NIV were retrospective and mainly focused on other objectives, 19,20 used NIV modes that were probably less tolerated than PSV, 21,22 or were performed in a small number of patients. 23 To our knowledge, there is no previous study that objectively proves the value of some clinical variables selected by generating and validating a multiple regression model that predicts the outcome of NIV in acute respiratory failure of CAFL patients. Our aim was to analyze prospectively the clinical and lung-function variables associated with the success or failure of NIV in the treatment of acute hypercapnic failure in patients with CAFL and to generate and validate a predictive model using multiple regression analysis. Materials and Methods The study was carried out between February 1995 and March 1998. In the first part of the study, 44 consecutive episodes of acute hypercapnic respiratory failure in 34 patients with an earlier diagnosis of CAFL were studied. They had not had previous treatment with NIV at home, were admitted to our hospital because of acute exacerbations of their condition, and were considered candidates for the study. Initially, all received conventional treatments, 1 4 including bronchodilators, corticosteroids, antibiotics (if necessary), and oxygen therapy with a Venturi mask, and controlled oxygen inspiration (fraction of inspired oxygen, 30%). Their response, however, was poor, as shown by the deterioration of arterial blood gas levels and clinical signs of respiratory failure. All patients had clinical symptoms (severe respiratory distress with dyspnea, use of accessory muscles of respiration, and/or abdominal paradoxical movements) and arterial blood gas levels (Pao 2, 55 mm Hg with supplementary oxygen; and/or ph, 7.30) that suggested a need for MV. 1 Patients with systolic pressure 90 mm Hg, unstable coronary disease, facial deformity, tracheostomy, or in whom intubation was necessary to remove excessive respiratory secretions, were excluded. All patients were treated in a conventional respiratory medicine ward, where nurses and physicians were trained in NIV techniques. A portable ventilator (BiPAP; Respironics; Murrysville, PA) was used in the spontaneous mode. A nose mask (Respironics) was initially used. If air leaks were excessive, a full-face mask (Respironics) was substituted. The pressure support level was initially set at 10 cm H 2 O and was gradually increased until the respiratory rate fell to 25 breaths per minute and no excessive leakage from the mouth was observed. Expiratory pressure was set at 6 cm H 2 O. A nasogastric tube was placed if gastric distension was observed. Arterial oxygen saturation was continuously monitored. Oxygen was administered through the mask until arterial oxygen saturation was 90%. The duration of the NIV session was similar to that used in other studies 18 and depended on the patient s tolerance to ventilation. Nighttime sessions were continuous, provided that patient tolerance permitted. Daytime sessions lasted from 3 to 12 h with pauses of 3hto allow administration of conventional medication, respiratory physiotherapy, feeding, and general patient care. Clinical signs and arterial blood gas levels were checked 1, 3, and 24 h after starting NIV. If the physician in charge detected persistence or worsening of clinical symptoms and pulmonary function variables, NIV was stopped and the patient was transferred to the ICU. Weaning from NIV was started when no clinical signs of respiratory distress were observed during the last pause. Then, if arterial blood gas levels were stable (ph, 7.35) after a 12-h pause, therapy was considered terminated, provided that the original cause of exacerbation was under control. We considered treatment with NIV to be successful if clinical and functional improvement had been achieved, OTI had been avoided, and the patient could be discharged. Relationships between the results of NIV and anthropometric, clinical, and lung-function variables were initially analyzed by a univariate analysis. The patients were classified as suffering from CAFL-emphysema, CAFL-chronic bronchitis, bronchiectasis with CAFL, CAFL post-tuberculosis, and unclassified CAFL, which were diagnosed on the basis of history, symptoms, physical examination, chest films, and lung function tests. An arbitrary scoring system to assess the level of consciousness (LC) was established as follows: 1, normal; 2, presence of flapping tremor; 3, confusion; and 4, stupor or coma. The acute physiology and chronic health evaluation (APACHE) II 24 score also was calculated for each patient before NIV. Nominal variables were analyzed by a 2 test. Fisher s Exact Test was employed if the number of cases was below five. For ordinal variables, the Kruskal-Wallis test was used. A t test was used for continuous variables, and a Mann-Whitney U test was used if continuous variables did not have a Gaussian distribution. The level of significance was set at 5%. A logistic multivariate regression model then was used to analyze the relationship between significant variables of the univariate study and to calculate a regression equation. A backward stepwise procedure, using log-likelihood ratio statistics as a selection criteria with two-tailed t test and a level of significance set at 0.1 (instead of 0.05 to minimize error, thereby using a conservative criterion for removing variables), was used to construct the final model. Missing data after 1 h of ventilation (for patients who did not tolerate NIV) were estimated from the means of the other data. In the second part of the study, we applied the regression equation to another 15 consecutive patients admitted to our hospital with acute hypercapnic failure due to exacerbation of CAFL for whom the same inclusion criteria were used. Results NIV was used to successfully treat 46 of 59 episodes (77%) of acute respiratory failure in the 49 patients included in the two consecutive parts of our study. The data related to clinical symptoms and lung function variables of 44 episodes of acute hypercapnic failure in the 34 patients included in the first part of the study are shown in Table 1. Their clinical status before and after 1 h of NIV is summarized in Table 2. In 34 of the 44 episodes (77%), OTI was avoided and the outcome was favorable. Patients with good initial response to NIV did not experience a recurrence of respiratory failure during their stay in the hospital. Ten patients did not respond to NIV, 5 of whom were transferred to the ICU where OTI and MV were performed. The other five patients were not transferred to the ICU; four died later in the conventional respiratory-medicine ward, and one CHEST / 117 / 3/ MARCH, 2000 829

Table 1 Clinical Characteristics and Results of Pulmonary Function Tests* Characteristics and Tests Success (n 34) Failure (n 10) p Value Age, yr 72 6 71.3 10 NS FEV 1, % predicted 27 11 38 11 0.01 Positive BRs, No. 12 5 FVC, % predicted 47 12 55 21 NS Pao 2 57 12 57 15 NS Paco 2 56 9 52 10 NS ph 7.38 0.03 7.39 0.03 NS Cause of CAFL, No. Emphysema 9 2 NS Chronic bronchitis 11 4 NS Bronchiectasis 5 1 NS Tuberculosis 4 1 NS Not classified 5 2 NS *NS not significant; BR bronchodilator response. Values given as mean SD, unless otherwise indicated. Success indicates successful treatment with NIV; failure indicates unsuccessful treatment with NIV. Defined as an improvement 15% of absolute FEV 1 after receiving bronchodilator. Table 2 Clinical Status Prior to and 1 h After NIV* Success (n 34) Failure (n 10) Status Pre-NIV 1 h After NIV Pre-NIV 1 h After NIV No. of patients with signs of respiratory infection (%) 19 (55) 4 (40) APACHE II 19 2 20 3 LC 2.44 0.89 1.58 0.65 2.8 1.13 3 1.24 Respiratory rate 29 6 22 6 28 5 23 7 Pao 2 at Fio 2 0.3 54 17 65 18 52 9 68 17 Paco 2 81 16 68 10 81 11 84 8 ph 7.27 0.03 7.34 0.04 7.28 0.04 7.28 0.03 *Fio 2 fraction of inspired oxygen. Values given as mean SD, unless otherwise indicated. See Table 1 for other definitions. Significant differences between failed and successful groups prior to NIV. Significant differences between failed and successful groups with NIV. improved with prolonged conventional treatment and was discharged. The mean duration of total NIV treatment was 2.22 0.6 days, and the mean inspiratory pressure employed was 16 2cmH 2 O. In all cases, supplementary oxygen was administered during NIV. Two cases of nostril cutaneous sores and one case of gastric distension were the only complications; and in none of the three cases was it necessary to discontinue NIV because of the complications. Before NIV, the successful patients had better LC scores (2.44 0.89 vs 2.8 1.13; p 0.05). However, APACHE II scores, signs of respiratory infection, respiratory rates, and the values for Pao 2, Paco 2, and ph at the beginning of NIV did not differ between patients in whom treatment with NIV was successful and not successful (Table 2). After 1 h of NIV the patients in whom NIV was successful also had better LC scores (p 0.05); they also had higher ph (p 0.001) and greater reductions in Paco 2 (p 0.001). In all patients in whom treatment with NIV was successful, Pao 2 changed similarly (Table 2). Changes in LC and ph with ventilation were related (r 0.37; p 0.05). We also observed that the LC prior to ventilation correlated with changes in ph (r 0.31; p 0.05) and LC with NIV (r 0.32; p 0.05). As expected, changes in Paco 2 and ph with NIV were highly correlated (r 0.93; p 0.001). The final regression model included the following: initial ph, Paco 2, and LC values; baseline FEV 1 and Paco 2 values under stable conditions; and the change in Paco 2 with NIV. The regression equation obtained was Y p p 1 where p ex, e is the base of natural logarithms, and X 223.76 (0.60 Dif Paco 2 ) (27.48 initial ph) (2.16 LC) (0.13 baseline Paco 2 ) (0.27 initial Paco 2 ) (0.25 baseline FEV 1 ). Dif 830 Clinical Investigations in Critical Care

Paco 2 was the difference between Paco 2 before NIV and after 1hofventilation (initialpaco 2 Paco 2 1h NIV), and baseline FEV 1 and Paco 2 were the last recorded values under stable conditions prior to admission. To make the calculation of the equation easier, LC was recodified on two levels: 0 for the LC scores of 1 and 2, and 1 for scores 3 and 4. The model correctly classified 95.5% of the initial 34 patients with a sensitivity of 0.97 (range, 0.91 to 1) and a specificity of 0.9 (range, 0.7 to 1) when the cutoff was set at 0.5. In the 15 patients included in the second part of the study to validate the multiple regression model, treatment with NIV was successful in 12 (80%), and the model correctly classified 14 patients (93%) (Table 3). Discussion NIV has been used successfully to treat acute exacerbations of CAFL 13,15 23,25 27 and has reduced the OTI and mortality rates in such patients who have severe acidosis. 17,18 In our study, 77% of patients were successfully treated with NIV in our conventional respiratory medicine ward, a proportion similar to that of studies performed in ICUs. 13,19,20,23,25,26 However, in some patients NIV is known to be ineffective. 18,27 Better screening criteria, based on predictive variables, are needed to identify which patients are theoretically least likely to respond to NIV and should, therefore, undergo MV immediately. 28 In the first part of the study, we evaluated prospectively the clinical and functional respiratory variables related to the efficacy of NIV provided by way of PSV in 44 episodes of acute hypercapnic failure in 34 patients with CAFL. The variables showing the highest correlation with NIV outcome were changes in Paco 2, ph, and LC after 1 h of ventilation. Like other authors, 19,20,23 we found that patients with an increase in ph and a decrease in Paco 2 soon after starting treatment with NIV responded successfully to it. The improvement in LC that we observed in successfully treated patients, which probably is attributable to the decrease in Paco 2, 5 suggested that continuous monitoring of LC may be useful for assessing response to ventilation. We acknowledge, however, that the integrity of the proposed model might be called into question because change in Paco 2 with NIV was included in the equation, and the worsening of clinical signs of respiratory failure or arterial blood gas measurement also was used as a criterion for stopping ventilation. We therefore designed a test of the model with 15 additional patients. This test confirmed the validity of the model, which is consistent with the fact that discharge to home rather than change in Paco 2 constituted the grounds for ultimate success. Moreover, the variables in the model were not the only bases for stopping NIV and switching to OTI. Thus, some patients were switched merely because of intolerance to NIV regardless of other factors. As for the screening criteria applied before NIV, we observed lower LC scores (indicating less altered consciousness) prior to ventilation in patients successfully treated with NIV, which is in disagreement with two smaller studies 22,25 in which no correlation between initial LC and success of treatment with NIV was found. Ambrosino et al, 20 on the other Table 3 Validation of the Multiple Regression Model* Patient No. Paco 2 b FEV 1 b LC phi Paco 2 i Paco 2 dif Y Result of NIV Agreement 1 61 46 0 7.26 74.6 19 0.99 Success Yes 2 57 10 0 7.24 83 14 0.99 Success Yes 3 51 24 1 7.27 78 1 0.59 Failure No 4 58 31 0 7.31 84 34 0.99 Success Yes 5 52 50 1 7.27 70 3 0.16 Failure Yes 6 62 20 1 7.26 96 7 0.95 Success Yes 7 62 34 0 7.21 91 12 0.99 Success Yes 8 60 22 0 7.32 84 17 0.99 Success Yes 9 52 32 0 7.3 74 9 0.99 Success Yes 10 73 28 1 7.28 80 11 0.99 Success Yes 11 44 51 1 7.3 65 3 0.002 Failure Yes 12 58 21 0 7.31 76 7 0.99 Success Yes 13 73 33 1 7.28 92 16 0.99 Success Yes 14 56 27 0 7.25 75 35 0.99 Success Yes 15 56 55 0 7.29 72 6 0.93 Success Yes *b baseline data available prior to admission with patients in clinically stable condition; i initial data obtained just before beginning of treatment with NIV; dif value obtained by subtracting Paco 2 value after 1 h of NIV from Paco 2 i; Y result of regression equation ( 0.5 predicted success of NIV treatment). See Table 1 for other definitions. CHEST / 117 / 3/ MARCH, 2000 831

hand, studying a series similar in number of patients to ours, also observed that initial LC scores were lower in patients in whom treatment with NIV was successful. These results support the consensus of the American Respiratory Care Foundation, 10 which stresses that altered consciousness should be a relative contraindication for NIV, given that confused patients are likely to adapt poorly to NIV as a result of impaired collaboration. In our patients who did not respond to NIV, baseline FEV 1 was higher, meaning, somewhat surprisingly, that patients with greater airway obstruction may respond better to NIV. In patients with less severe CAFL (higher FEV 1 ) the presence of severe acute respiratory failure may imply greater alteration and workload of respiratory muscles generally. Thus, the presence of previously severe airflow limitation should not be used to rule out the application of this technique. The observation of Soo Hoo et al 23 that baseline FEV 1 was similar in all patients regardless of outcome may be attributable to either the very small number of patients enrolled in their study or to the use of the volume control mode of NIV, which has been described as less effective and less well tolerated than PSV. 29 33 Finally, the nature of the respiratory obstruction seems to be unimportant for the selection of patients for NIV, given that the distribution of types of CAFL was similar in our patients treated successfully and unsuccessfully with NIV. Benhamou et al 22 similarly observed no differences in success of treatment with NIV for different patterns of CAFL. Age in our study also was unrelated to outcome, a fact consistent with the observations of other authors, 23,25,26 indicating that NIV needs not be denied to older patients. Signs of respiratory infection also were similar in our patients in whom NIV succeeded and failed, as in previous studies. 19,22,25 No other initial variable recorded in our study (respiratory rate, APACHE II scores, and arterial blood gas levels) was related to outcome. Our validated multiple regression model, combining baseline and initial variables with their evolution after 1hofNIV,allowed our model to correctly classify 95% of patients in the first series with a higher sensitivity and specificity than does the only other model described previously, 20 which does not include Paco 2 change after a short session of NIV as a main variable. The second part of our study, the validation of the model by confirming its predictive power in 15 consecutive patients, correctly classified 93% of them. The clinical use of the equation can be complex if appropriate software is not available. However, in order to predict the success of treatment with NIV in CAFL patients with hypercapnic failure, a sound alternative could be to carry out a brief trial (1 h or so) of ventilation with monitoring of LC, Paco 2, and ph. ACKNOWLEDGMENT: We thank M. E. Kerans for the English translation and editing of the manuscript. References 1 American Thoracic Society. Standards for the diagnosis and care of patients with chronic obstructive pulmonary disease. Am J Respir Crit Care Med 1995; 152:S7 S120 2 American College of Chest Physicians. Mechanisms and management of COPD. Chest 1998; 113(suppl):S233 S276 3 Siafakas NM, Vermerie P, Pride NB, et al. Optimal assessment and management of chronic obstructive pulmonary disease. Eur Respir J 1995; 8:1398 1420 4 British Thoracic Society. Guidelines for the management of chronic obstructive pulmonary disease. Thorax 1997; 52(suppl 5):S1 S27 5 Derenne JP, Fleury B, Pariente R. Acute respiratory failure of chronic obstructive pulmonary disease. Am Rev Respir Dis 1988; 138:1006 1033 6 American College of Chest Physicians. Mechanical ventilation beyond the intensive care unit. Chest 1998; 113(suppl): S289 S344 7 Meyer T, Hill N. Noninvasive positive pressure ventilation to treat respiratory failure. Ann Intern Med 1994; 120:760 770 8 Ambrosino N. Noninvasive mechanical ventilation in acute respiratory failure. Eur Respir J 1996; 9:795 807 9 Meduri GU. Noninvasive positive-pressure ventilation in patients with acute respiratory failure. Clin Chest Med 1996; 17:513 553 10 Shneerson JM. The changing role of mechanical ventilation in COPD. Eur Respir J 1996; 9:393 398 11 American Respiratory Care Foundation Consensus Conference. Non-invasive positive pressure ventilation. Respir Care 1997; 42:364 369 12 Hillberg RE, Johnson DC. Noninvasive ventilation. N Engl J Med 1997; 337:1746 1752 13 Brochard L, Isabey D, Piquet J, et al. Reversal of acute exacerbations of chronic obstructive lung disease by inspiratory assistance with a face mask. N Engl J Med 1990; 323:1523 1530 14 Nava S, Ambrosino N, Rubini F, et al. Effect of nasal pressure support ventilation and external PEEP on diaphragmatic activity in patients with severe stable COPD. Chest 1993; 103:143 150 15 Confalonieri M, Parigi P, Scartaballati A, et al. Non-invasive mechanical ventilation improves the immediate and longterm outcome of COPD patients with acute respiratory failure. Eur Respir J 1996; 9:422 430 16 Díaz O, Iglesia R, Ferrer M, et al. Effects of noninvasive ventilation on pulmonary gas exchange and hemodynamics during acute hypercapnic exacerbations of chronic obstructive pulmonary disease. Am J Respir Crit Care Med 1997; 156:1840 1845 17 Kramer N, Meyer T, Meharg J, et al. Randomized, prospective trial of noninvasive PPV in acute respiratory failure. Am J Respir Crit Care Med 1995; 151:1799 1806 18 Brochard L, Mancebo J, Wysocky M, et al. Noninvasive ventilation for acute exacerbations of chronic obstructive pulmonary disease. N Engl J Med 1995; 333:817 822 19 Meduri GU, Turner E, Abou-Shala N, et al. Non-invasive positive pressure ventilation via face mask: first-line intervention in patients with acute hypercapnic and hypoxemic respi- 832 Clinical Investigations in Critical Care

ratory failure. Chest 1996; 109:179 193 20 Ambrosino N, Foglio K, Rubini F, et al. Non-invasive mechanical ventilation in acute respiratory disease due to chronic obstructive pulmonary disease: correlates for success. Thorax 1995; 50:755 757 21 Bott J, Carroll M, Conway JH, et al. Randomised controlled trial of nasal ventilation in acute ventilatory failure due to chronic obstructive airways disease. Lancet 1993; 341:1555 1557 22 Benhamou D, Girault C, Faure C, et al. Nasal mask ventilation in acute respiratory failure: experience in elderly patients. Chest 1992; 102:912 917 23 Soo Hoo G, Santiago S, Williams A, et al. Nasal mechanical ventilation for hypercapnic respiratory failure in chronic obstructive pulmonary disease: determinants of success and failure. Crit Care Med 1994; 22:1253 1261 24 Knauss WA, Draper EA, Zimmerman JE. APACHE II: a severity of disease classification system. Crit Care Med 1985; 13:818 829 25 Wysocky M, Tric L, Wolff M, et al. Noninvasive pressure support ventilation in patients with acute respiratory failure. Chest 1993; 103:907 913 26 Pennock B, Kaplan P, Carlin B, et al. Pressure support ventilation with a simplified ventilatory support system administered with a nasal mask in patients with respiratory failure. Chest 1991; 100:1371 1376 27 Barbé F, Togores M, Rubí M, et al. Non-invasive ventilatory support does not facilitate recovery from acute respiratory failure in chronic obstructive pulmonary disease. Eur Respir J 1996; 9:1240 1245 28 Hess D. Noninvasive positive pressure ventilation: predictors of success and failure for acute care applications. Respir Care 1997; 42:424 431 29 Brochard L. Non-invasive ventilation: practical issues. Intensive Care Med 1993; 19:431 432 30 Mancebo J, Isabey D, Lorino H. Comparative effects of pressure support ventilation and intermittent positive pressure breathing in non-intubated healthy subjects. Eur Respir J 1995; 8:1901 1909 31 Girault C, Richard JC, Chevron V. Comparative physiologic effects of noninvasive assist-control and pressure support ventilation in acute hypercapnic respiratory failure. Chest 1997; 111:1639 1648 32 Tejeda M, Boix JH, Alvarez F. Comparison of pressure support ventilation and assist-control ventilation in the treatment of respiratory failure. Chest 1997; 111:1322 1325 33 Cinnella G, Conti G, Lofaso F. Effects of assisted ventilation on the work of breathing: volume-controlled versus pressurecontrolled ventilation. Am J Respir Crit Care Med 1996; 153:1025 1033 CHEST / 117 / 3/ MARCH, 2000 833