Establishing the relative accuracy of three new definitions of the adult respiratory distress syndrome

Abstract
Over the last few years, new definitions of the adult respiratory distress syndrome (ARDS) have been introduced that potentially identify patients earlier in their course of acute lung injury. However, these definitions have never been compared with any of the older and potentially stricter definitions of ARDS to determine if similar patients are eventually identified. We compared new definitions of ARDS--as represented by the Lung Injury Score, a modified Lung Injury Score, and the American-European Consensus Conference definition--against a stricter definition of ARDS to determine their accuracy.Prospective.Intensive care unit (ICU) patients in a tertiary, university-affiliated city hospital.ICU patients with clearly defined at-risk diagnoses for ARDS (group 1, n = 111) and general medical ICU patients without clearly defined at-risk diagnoses for ARDS (group 2, n = 125).Measurements of hypoxemia, static respiratory system compliance, positive end-expiratory pressure, radiographic changes, and general demographic information were collected. The sensitivity, specificity, positive-predictive value, negative-predictive value, and accuracy of all three new definitions were determined. Accuracy was defined as the true-positive plus the true-negative results divided by the total number of patients. When compared with a stricter definition of ARDS, all three definitions maintained a high degree of accuracy in those patients with a clearly defined at-risk diagnosis (group 1): Lung Injury Score 90.0% (95% confidence interval 84-96); modified Lung Injury Score 97.3% (95% confidence interval 94-100), and the American-European Consensus Conference definition 97.3% (95% confidence interval 94-100). For these at-risk patients, the accuracy of the modified Lung Injury Score and the American-European Consensus Conference definition was significantly better than the Lung Injury Score when compared with the strict definition (p = .027 for both comparisons). Although all three definitions maintained an accuracy of > 90% for general medical ICU patients (group 2), the low frequency of ARDS in these patients (3.4%) produced a low positive-predictive value for all three definitions.We conclude that the Lung Injury Score, the modified Lung Injury Score, and the American-European Consensus Conference definition identify similar patients, provided that these methods are applied to patients with clearly defined at-risk diagnoses for ARDS.