Algorithm for the diagnosis of smear-negative pulmonary tuberculosis in high-incidence resource-constrained settings.

OBJECTIVES Diagnosis of smear-negative pulmonary tuberculosis (SNPT) remains a challenge, particularly in resource-constrained settings. We evaluated a diagnostic algorithm that combines affordable laboratory tools and a clinical prediction rule (CPR). METHODS We derived, based on published evidence, a diagnostic algorithm for SNPT. Sputum concentration constitutes its first step. In suspects with negative results, SNPT probability is classified with a CPR as low (excluded), high (confirmed) or intermediate. For intermediate patients, sputum Middlebrook 7H9 liquid culture is performed, and they are assessed after 2 weeks. If clinically deteriorated, with still negative liquid culture, bronchoscopy is offered. Otherwise, results of Middlebrook 7H9 culture are awaited. We prospectively evaluated this algorithm against a reference standard of solid and liquid cultures in two reference hospitals in Lima, Peru. RESULTS 670 SNPT suspects were included from September 2005 to March 2008. The prevalence of SNPT was 27% according to the reference standard. The algorithm's overall accuracy was 0.94 (95% CI 0.91-0.95), its sensitivity was 0.88 (95% CI 0.82-0.92) and its specificity, 0.96 (95% CI 0.94-0.98). Sputum concentration, the CPR, Middlebrook 7H9 sputum culture and bronchoscopic samples defined a diagnosis of SNPT according to the algorithm in 57 (37%), 25 (16%), 63 (41%) and 8(5%) of patients, respectively. 65% of patients were diagnosed within 3 weeks. CONCLUSIONS The algorithm was accurate for SNPT diagnosis. Sputum concentration, CPR and selective Middlebrook 7H9 culture are essential components.