Causal effect of loss to follow-up on mortality in a population-based tuberculosis cohort in Brazil

Background Loss to follow-up (LTFU) during tuberculosis treatment is a major programmatic gap, but its causal effect on mortality has been difficult to quantify. We estimated this effect using a sequential landmark cohort analysis. Methods and Findings We constructed a retrospective cohort of individuals aged [≥]15 years initiating their first tuberculosis treatment in Sao Paulo State, Brazil (2013-2023), using the State registry (TBweb) linked to the national mortality system (SIM). To account for immortal-time bias, we compared mortality between LTFU and continued-treatment patients in time-aligned monthly cohorts, applying a symmetric 30-day grace period to align eligibility. Cause-specific Cox models estimated adjusted hazard ratios for late (6-24 month) mortality, with non-TB mortality as a within-cohort negative control. We also computed standardized (g-formula) absolute mortality risk differences over the same window. Effect modification was assessed across pre-specified subgroups (age, sex, HIV, homelessness, drug-resistance status, calendar period). Of 171,048 individuals initiating tuberculosis therapy, 20,830 (12.2%) experienced LTFU. LTFU at any month substantially increased late mortality (adjusted hazard ratios [aHR] 1.83 [95% CI 1.27-2.64] to 2.85 [2.34-3.48] by month of LTFU), corresponding to standardized late-window mortality risk differences of up to 1.6 percentage points. The excess was concentrated in TB-attributable deaths (aHR 1.60-4.36) and was essentially null for non-TB mortality (0.96-1.48). Relative effects were largest in younger, stably housed individuals with low baseline mortality, whereas the largest absolute excess fell on people living with HIV (risk difference 3.0 percentage points). Conclusions LTFU at any point in tuberculosis therapy substantially increased late TB-attributable mortality, consistent with a causal pathway through interrupted treatment. Preventing LTFU should be a programmatic priority.