Background: Drug resistance is a challenge for the global control of tuberculosis. We examined mortality in tuberculosis patients from high-burden countries, according to concordance or discordance of results from drug susceptibility testing (DST) done locally and in a reference laboratory. Methods: We collected Mycobacterium tuberculosis isolates from adult patients in Côte d’Ivoire, Democratic Republic of the Congo, Kenya, Nigeria, South Africa, Peru, and Thailand, stratified by HIV status and tuberculosis drug resistance. Molecular or phenotypic drug susceptibility testing (DST) was done locally and at the Swiss tuberculosis reference laboratory. We examined mortality during treatment according to DST results and treatment adequacy in logistic regression models adjusting for sex, age, sputum microscopy and HIV status. Findings: 634 tuberculosis patients were included; median age was 33.2 years, 239 (37.7%) were female, 272 (42.9%) HIV-positive and 69 (10.9%) patients died. Based on the reference laboratory DST, 394 (62.2%) strains were pan-susceptible, 45 (7.1%) mono-resistant, 163 (25.7%) multidrug-resistant (MDR-TB), and 30 (4.7%) had pre-extensive or extensive drug resistance (pre-XDR/XDR-TB). Results of reference and local laboratories were discordant in 121 (19.1%) cases. Overall, sensitivity and specificity to detect any resistance were 90.8% and 84.3%, respectively. Mortality ranged from 6.0% (20/336) in patients with pan-susceptible tuberculosis treated according to WHO guidelines to 57.1% (8/14) in patients with resistant strains who were under treated. In logistic regression, compared to concordant DST results, the adjusted odds ratio of death was 7.33 (95% CI 2.70–19.95) for patients with discordant results potentially leading to under treatment. Interpretation: Inaccurate DST by comparison to a reference standard led to under treatment of drug resistant tuberculosis and increased mortality. Rapid molecular DST of first- and second-line drugs at diagnosis is required to improve outcomes in patients with MDR-TB and pre-XDR/XDR-TB.

BACKGROUND: Diagnosis of tuberculosis should be improved in children infected with HIV to reduce mortality. We developed prediction scores to guide antituberculosis treatment decision in HIV-infected children with suspected tuberculosis. METHODS: HIV-infected children with suspected tuberculosis enrolled in Burkina Faso, Cambodia, Cameroon, and Vietnam (ANRS 12229 PAANTHER 01 Study), underwent clinical assessment, chest radiography, Quantiferon Gold In-Tube (QFT), abdominal ultrasonography, and sample collection for microbiology, including Xpert MTB/RIF (Xpert). We developed 4 tuberculosis diagnostic models using logistic regression: (1) all predictors included, (2) QFT excluded, (3) ultrasonography excluded, and (4) QFT and ultrasonography excluded. We internally validated the models using resampling. We built a score on the basis of the model with the best area under the receiver operating characteristic curve and parsimony. RESULTS: A total of 438 children were enrolled in the study; 251 (57.3%) had tuberculosis, including 55 (12.6%) with culture- or Xpert-confirmed tuberculosis. The final 4 models included Xpert, fever lasting >2 weeks, unremitting cough, hemoptysis and weight loss in the past 4 weeks, contact with a patient with smear-positive tuberculosis, tachycardia, miliary tuberculosis, alveolar opacities, and lymph nodes on the chest radiograph, together with abdominal lymph nodes on the ultrasound and QFT results. The areas under the receiver operating characteristic curves were 0.866, 0.861, 0.850, and 0.846, for models 1, 2, 3, and 4, respectively. The score developed on model 2 had a sensitivity of 88.6% and a specificity of 61.2% for a tuberculosis diagnosis. CONCLUSIONS: Our score had a good diagnostic performance. Used in an algorithm, it should enable prompt treatment decision in children with suspected tuberculosis and a high mortality risk, thus contributing to significant public health benefits.