Diagnostic accuracy of bronchoalveolar lavage samples in immunosuppressed patients with suspected pneumonia: analysis of a protocol

Abstract

Background: Fast and accurate etiologic diagnosis of pneumonia in immunocompromised patients is essential for a good outcome. Utility of bronchoalveolar lavage (BAL) samples has already been established, but studies about them are scarce and limited to few countries. We aimed to evaluate the accuracy of a diagnostic protocol, emphasizing on local epidemiology, rapidity, and yield of different techniques.

Methods: One year prospective study of 101 consecutive immunosuppressed patients admitted with suspected pneumonia to a university hospital. They all had bronchoscopic BAL (n=109) and respiratory sampling. Conventional microbiological studies, cytomegalovirus pp65 antigenemia and transbronchial biopsy (TBB), whenever considered pertinent, were done. Results were analyzed along with other diagnostic procedures, clinical course and final outcome.

Results: HIV/AIDS infection was the most frequent cause of inclusion (n=80). Infections accounted for 79 out of 122 final diagnoses (64.8%). Our protocol identified 60 infectious and 3 noninfectious pathologies (general yield: 51.6%). Sensitivity in pulmonary infections was 75.9% (IC95%: 64.8-84.6%), specificity 86.0% (72.6-93.7%), positive predictive value 89.6% (79.1-95.3%), negative predictive value 69.4% (56.2-80.1%), accuracy 79.8% (71.7-86.2%). Mycobacterium spp. (n=27), bacteria (n=19), Pneumocystis jirovecii (n=18) and other fungi (histoplasmosis: 6, aspergillosis: 5, cryptococosis: 3) were the most common infectious pathogens. Direct microscopy allowed an early definite/presumptive diagnosis in 36/49 fungal and mycobacterial infections (73.5%). Up to 30% of mycobacterial infections were missed.

Conclusions: Systematical study of BAL samples has a high diagnostic yield in our immunocompromised patients with suspected pneumonia. As economical and epidemiological conditions of regions are different, it should be tried everywhere.