Trial-of-antibiotic algorithm for the diagnosis of tuberculosis in a district hospital in a developing country with high HIV prevalence

OBJECTIVE: To evaluate a diagnostic algorithm for pulmonary tuberculosis ba sed on smear microscopy and objective response to trial of antibiotics. SETTING: Adult medical wards, Hlabisa Hospital, South Africa, 1996-1997. METHODS: Adults with chronic chest symptoms and abnormal chest X-ray had sp utum examined for Ziehl-Neelsen stained acid-fast bacilli by light microsco py. Those with negative smears were treated with amoxycillin for 5 days and assessed. Those who had not improved were treated with erythromycin for 5 days and reassessed. Response was compared with mycobacterial culture. RESULTS: Of 280 suspects who completed the diagnostic pathway, 160 (57%) ha d a positive smear, 46 (17%) responded to amoxycillin, 34 (12%) responded t o erythromycin and 40 (14%) were treated as smear-negative tuberculosis. Th e sensitivity (89%) and specificity (84%) of the full algorithm for culture -positive tuberculosis were high. However, 11 patients (positive predictive value [PPV] 95%) were incorrectly diagnosed with tuberculosis, and 24 case s of tuberculosis (negative predictive value [NPV] 70%) were not identified . NPV improved to 75% when anaemia was included as a predictor. Algorithm p erformance was independent of human immunodeficiency virus status. CONCLUSION: Sputum smear microscopy plus trial of antibiotic algorithm amon g a selected group of tuberculosis suspects may increase diagnostic accurac y in district hospitals in developing countries.