Suboptimal diagnostics for pulmonary tuberculosis drive the use of the so-called trial of antibiotics, a course of broad-spectrum antibiotics without activity against Mycobacterium tuberculosis that is given to patients who are mycobacteriology negative but symptomatic, with the aim of distinguishing pulmonary tuberculosis from bacterial lower respiratory tract infection. The underlying assumption—that patients with lower respiratory tract infection will improve, whereas those with pulmonary tuberculosis will not—has an unclear evidence base for such a widely used intervention (at least 26·5 million courses are prescribed per year). We aimed to collate available evidence on the diagnostic performance of the trial of antibiotics.

In this systematic review and meta-analysis we searched the MEDLINE, Embase, and Global Health databases for studies published up to March 15, 2019, that investigated the sensitivity and specificity of the trial of antibiotics against mycobacteriology tests in adults (≥15 years) with tuberculosis symptoms. We used the QUADAS-2 tool to assess the risk of bias. We estimated pooled values for sensitivity and specificity of trial of antibiotics (as the index text) versus mycobacteriology tests (as the reference standard) using random-effects bivariate modelling, and we used the I² statistic to assess heterogeneity between studies contributing to these estimates. This study is registered with PROSPERO, number CRD42017083915.

Of the 9410 articles identified by our search, eight studies were eligible for inclusion. The studies were from seven countries in Africa, South America, and Asia, and involved 2786 participants. Six studies used mycobacterial culture as the reference standard, and six used penicillins for the trial of antibiotics. The treatment duration, number of antimicrobial courses, and definition of what constituted response to treatment varied substantially between studies. The pooled sensitivity (67%, 95% CI 42–85) and specificity (73%, 58–85) of the trial of antibiotics versus mycobacteriology tests were below internationally defined minimum performance profiles for tuberculosis diagnostics and had substantial heterogeneity (I² was 96% for sensitivity and 99% for specificity). Each included study failed on one or more domain of the QUADAS-2 tool.

Current policy and practice regarding the trial of antibiotics appear inappropriate, given the weak evidence base, poor diagnostic performance, potential contribution to the global antimicrobial resistance crisis, and adverse individual and public health consequences from the misclassification of tuberculosis status. Antibiotic strategies during tuberculosis investigations should instead optimise clinical outcomes, ideally guided by clinical trials in both inpatient and outpatient groups.

Helse Nord RHF, Wellcome Trust, and the UK Commonwealth Scholarship Commission.