Baseline characteristics of the 80 enrolled patients are presented in stratified according to the final expert diagnosis. Medical and surgical ICUs were equally represented (58% and 42% of the patients, respectively). Patients were high risk based on the APACHE II score and other clinical criteria. Subjects with CPE were more likely to be older, to have a history of CHF, and to have worse kidney function. There was no statistical difference in the frequency of prior cardiac disease, and the overall severity of illness was indistinguishable as captured by the APACHE II score. In addition, there were no significant differences in temperature, systolic BP, heart rate, WBC count, lactate levels, proportion of patients with documented bacteremia, or proportion of patients receiving one or multiple pressors between the ALI/ARDS and CPE groups. The most frequent ICU admitting diagnoses were sepsis (34%) and pneumonia (15%), followed by respiratory failure (12%), abdominal surgery (9%), CHF (8%), pancreatitis (6%), and ischemic bowel (5%).
Baseline Characteristics at Enrollment by Expert Diagnosis*
Subjects were enrolled at a median of 3 days (25 to 75% interquartile range [IQR], 1 to 9 days) following admission to the hospital, and at a median of 0 days (IQR, 0 to 2 days) from the time of admission to the ICU. The median fluid input/output from the time of admission to the ICU until the performance of RHC was + 4,100 mL (IQR, 1,000 to 9,250 mL). The median concentrations of BNP and echocardiographic, radiographic, ventilatory, and hemodynamic data are presented in . The radiographic and ventilatory data presented were collected at the time of hemodynamic measurements. Echocardiographic data collected and presented were the most recent data in relation to the hemodynamic assessment, of which 54% were within 48 h between the two studies. Subjects with a diagnosis of ARDS had a higher median ejection fraction (p = 0.03) and lower mean right atrial (p = 0.0006) and end-expiratory wedge pressures (p = 0.001) compared to subjects with CPE, and were more likely to be managed with higher levels of positive end-expiratory pressure (p = 0.004).
Radiologic, Echocardiographic, Hemodynamic, and BNP Data*
BNP and Diagnosis of ARDS
The initial clinical diagnosis of CPE by the treating physician prior to RHC was in agreement with the experts’ final diagnosis in only 48.6%, with the remainder classified as ARDS by expert final diagnosis (). In contrast, the initial diagnosis of ARDS by the clinician was more frequently in agreement with the experts’ final diagnosis (). Patients with a diagnosis of ARDS had significantly lower median levels of BNP compared to subjects with a diagnosis of CPE (325 pg/mL vs 1,260 pg/mL, p = 0.0001) . This observation remained consistent when restricted to those for whom the initial clinical diagnosis was CPE (388 pg/mL vs 1,280 pg/mL, p = 0.01). In addition, BNP was modestly correlated with end-expiratory PCWP (r = 0.27, p = 0.02). Of note, when PCWP was used alone to categorize patients as ALI/ARDS vs CPE, patients categorized with ARDS (PCWP < 18 mm Hg) similarly had significantly lower median levels of BNP compared to those with PCWP ≥ 18 mm Hg (550 pg/mL vs 1,192 pg/mL, p = 0.0003).
Concordance of clinician pre-RHC diagnosis with expert diagnosis (including RHC data).
The area under the ROC curve for BNP in relation to the expert’s final diagnosis of ARDS was 0.80. Including patients with a mixed diagnosis as CPE had minimal impact on the area under the ROC curve (C-statistic, 0.79). The diagnostic performance of BNP at several cut points is presented in , . At a cut point ≤ 200 pg/mL, BNP provided a specificity of 91% and a positive predictive value of 91% for the diagnosis of ARDS. At a cut off point ≥ 1.200 pg/mL, BNP had a specificity of 92% and a positive predictive value of 75% for the diagnosis of CPE. The use of the lower cut-off point ≤ 200 pg/mL for BNP measurement would have reclassified 8 of 36 patients (22%) with a pre-RHC assessment of CPE as ARDS. The use of the higher cut-off point ≥ 1,200 pg/mL would have reclassified 1 of 35 subjects (3%) with a pre-RHC assessment of ARDS instead as CPE. Including patients with a mixed diagnosis as CPE had minimal impact on the diagnostic performance of BNP (, ). Moreover, use of the PCWP alone to categorize patients with ARDS vs CPE did not qualitatively alter the results (C-statistic, 0.75; specificity, 93% for ARDS using BNP ≤ 200 pg/mL; specificity, 87% for CPE using BNP ≥ 1,200 pg/mL).
Performance Characteristics of Various Cut Points of BNP Excluding Patients With a “Mixed” Diagnosis*
Performance Characteristics of Various Cut Points of BNP Including Mixed Diagnosis Patients as CPE*
In multivariable analyses, a higher concentration of BNP was associated with a decreased odds of the diagnosis of ARDS (odds ratio [OR], 0.4 per log increase in BNP; p = 0.027) after adjustment for baseline covariates. The final covariates in the model included log BNP, right atrial mean pressure, and a history of CHF. The overall C-statistic for the model was 0.88, with adequate fit as assessed by the Hosmer-Lemeshow test (p > 0.3). In these analyses, indexes of renal function (both BUN as well as GFR) or the pulmonary artery systolic pressure were not associated with the diagnosis of ARDS.
In exploratory analyses, log BNP was as discriminatory a covariate as PCWP (C-statistic of model substituting PWCP for BNP, 0.88) and provided additional explanatory power to that of PCWP alone (p = 0.017 by likelihood ratio test). Further sensitivity analyses including adjustment for echocardiographic parameters (LVEF and presence of a wall-motion abnormality) in patients in whom these data were available resulted in only a modest shift in the estimate of the association between BNP and ARDS (adjusted OR, 0.5 [confidence interval, 0.2 to 1.0] per log increase in BNP).
BNP and In-hospital Mortality
BNP showed a strong graded relationship with mortality risk in subjects with ARDS and CPE (p = 0.03; ). This association remained independent after considering APACHE II score and other covariates associated with mortality in univariate analyses. Only two parameters remained in the final model as independently associated with mortality: BNP (OR, 1.6 for each log increase; p = 0.03) and the APACHE II score (OR, 1.1 for each point increase; p = 0.002), with an overall C-statistic of the model of 0.76 and adequate goodness of fit (p > 0.6). Of note, the independent association between BNP and mortality remained after further adjustment for PCWP.
In-hospital mortality stratified by BNP.