Rationale The accuracy and impact of new tuberculosis (TB) tests, such as Xpert MTB/RIF, when performed on bronchoalveolar lavage fluid (BALF) obtained from patients with sputum-scarce or smear-negative TB is unclear.

Methods South African patients with suspected pulmonary TB (n=160) who were sputum-scarce or smear-negative underwent bronchoscopy. MTB/RIF was performed on uncentrifuged BALF (1 ml) and/or a resuspended pellet of centrifuged BALF (∼10 ml). Time to TB detection and anti-TB treatment initiation were compared between phase one, when MTB/RIF was performed as a research tool, and phase two, when it was used for patient management.

Results 27 of 154 patients with complete data had culture-confirmed TB. Of these, a significantly lower proportion were detected by smear microscopy compared with MTB/RIF (58%, 95% CI 39% to 75% versus 93%, 77% to 98%; p<0.001). Of the 127 patients who were culture negative, 96% (91% to 98%) were MTB/RIF negative. When phase two was compared with phase one, MTB/RIF reduced the median days to TB detection (29 (18–41) to 0 (0–0); p<0.001). However, more patients initiated empirical therapy (absence of a positive test in those commencing treatment) in phase one versus phase two (79% (11/14) versus 28% (10/25); p=0.026). Consequently, there was no detectable difference in the overall proportion of patients initiating treatment (26% (17/67; 17% to 37%) versus 36% (26/73; 26% to 47%); p=0.196) or the days to treatment initiation (10 (1–49) versus 7 (0–21); p=0.330). BALF centrifugation, HIV coinfection and a second MTB/RIF did not result in detectable changes in accuracy.

Conclusions MTB/RIF detected TB cases more accurately and more rapidly than smear microscopy and significantly reduced the rate of empirical treatment.