Objective: To assess the predictive value of the endotoxin level in the bronchoalveolar lavage (BAL) and to propose to the clinician a guide in the diagnosis of gram-negative bacterial (GNB) pneumonia.
Design: Retrospective and prospective studies to investigate the relation between endotoxin level and quantitative bacterial culture of BAL and to test the predictive value of a defined threshold.
Setting: University hospital general intensive care unit.
Patients: In the first part of the study, 77 consecutive ventilated patients with clinical suspicion of nosocomial pneumonia between January 1995 and January 1996. In the second part of the study, 93 consecutive ventilated patients studied prospectively between February 1996 and April 1997.
Measurements and main results: Quantitative cultures for aerobic bacteria were performed directly from the fluid. Bacterial species were determined with standard techniques. The detection of endotoxin in BAL was made using a quantitative chromogenic Limulus assay. In the retrospective analysis, a significant correlation between quantitative GNB cultures and BAL endotoxin levels was observed (r2 = 0.60, p < .0001). An endotoxin level > or = 4 endotoxin units/mL (EU/mL) distinguishes patients with a significant GNB count from colonized patients with a sensitivity of 92.6%, a specificity of 81.4% and a correct classification rate of 84.9%. In the prospective analysis, the 4 EU/mL threshold permits identification of infected patients with a sensitivity of 82.2%, a specificity of 95.6%, and a correct classification rate of 90.3%. The receiver operating characteristic curve analysis showed that the Limulus assay still had a good discrimination power in the prediction of significant bacterial count in BAL fluid.
Conclusions: Endotoxin detection immediately after bronchoscopy is a distinct advantage to the clinician because antimicrobial gram-negative therapy may be immediately justified according to the results.