Background: Mortality within the first 6 months after initiating antiretroviral therapy is common in resource-limited settings and is often due to tuberculosis in patients with advanced HIV disease. Isoniazid preventive therapy is recommended in HIV-positive adults, but subclinical tuberculosis can be difficult to diagnose. We aimed to assess whether empirical tuberculosis treatment would reduce early mortality compared with isoniazid preventive therapy in high-burden settings. Methods: We did a multicountry open-label randomised clinical trial comparing empirical tuberculosis therapy with isoniazid preventive therapy in HIV-positive outpatients initiating antiretroviral therapy with CD4 cell counts of less than 50 cells per μL. Participants were recruited from 18 outpatient research clinics in ten countries (Malawi, South Africa, Haiti, Kenya, Zambia, India, Brazil, Zimbabwe, Peru, and Uganda). Individuals were screened for tuberculosis using a symptom screen, locally available diagnostics, and the GeneXpert MTB/RIF assay when available before inclusion. Study candidates with confirmed or suspected tuberculosis were excluded. Inclusion criteria were liver function tests 2·5 times the upper limit of normal or less, a creatinine clearance of at least 30 mL/min, and a Karnofsky score of at least 30. Participants were randomly assigned (1:1) to either the empirical group (antiretroviral therapy and empirical tuberculosis therapy) or the isoniazid preventive therapy group (antiretroviral therapy and isoniazid preventive therapy). The primary endpoint was survival (death or unknown status) at 24 weeks after randomisation assessed in the intention-to-treat population. Kaplan-Meier estimates of the primary endpoint across groups were compared by the z-test. All participants were included in the safety analysis of antiretroviral therapy and tuberculosis treatment. This trial is registered with ClinicalTrials.gov, number NCT01380080. Findings: Between Oct 31, 2011, and June 9, 2014, we enrolled 850 participants. Of these, we randomly assigned 424 to receive empirical tuberculosis therapy and 426 to the isoniazid preventive therapy group. The median CD4 cell count at baseline was 18 cells per μL (IQR 9-32). At week 24, 22 (5%) participants from each group died or were of unknown status (95% CI 3·5-7·8) for empirical group and for isoniazid preventive therapy (95% CI 3·4-7·8); absolute risk difference of -0·06% (95% CI -3·05 to 2·94). Grade 3 or 4 signs or symptoms occurred in 50 (12%) participants in the empirical group and 46 (11%) participants in the isoniazid preventive therapy group. Grade 3 or 4 laboratory abnormalities occurred in 99 (23%) participants in the empirical group and 97 (23%) participants in the isoniazid preventive therapy group. Interpretation: Empirical tuberculosis therapy did not reduce mortality at 24 weeks compared with isoniazid preventive therapy in outpatient adults with advanced HIV disease initiating antiretroviral therapy. The low mortality rate of the trial supports implementation of systematic tuberculosis screening and isoniazid preventive therapy in outpatients with advanced HIV disease.