Researchers followed 171,048 individuals aged 15 years or older who initiated a first course of tuberculosis treatment in São Paulo State between 2013 and 2023. They defined loss to follow-up (LTFU) as an interruption of treatment for at least 30 consecutive days; 20,830 people (12.2%) met that definition. The study team used a clone-censor-weight target trial emulation to contrast two strategies over 60 months from treatment start: disengage at any time in the first 6 months versus remain in care throughout. Analyses were adjusted for demographic and clinical covariates including age, sex, race, education, HIV status, diabetes, alcohol and drug use, incarceration, homelessness, hospitalization at diagnosis, clinical form of disease, directly observed therapy (DOT), and municipality.
Data indicated that LTFU increased the standardized 5-year risk of death by 2.49 percentage points (95% CI 1.94–3.17). That magnitude corresponds to approximately 25 additional deaths per 1,000 people over five years attributable to disengagement under the stated identification assumptions. The effect estimate derives from observational registries but uses causal-emulation weighting to address several sources of bias inherent to routine data.
The investigators decomposed excess deaths by cause using linked mortality records and high-level ICD-10 groupings available in the registry. Cause attribution indicated that tuberculosis accounted for 61% of the 5-year excess mortality associated with disengagement. Other causes contributed the remaining share; the source reported only aggregated cause-group results and did not provide granular cause-specific event counts in the abstract. The dataset available via Zenodo contains reduced ICD-10 classifications but not the detailed individual-level records, which remain confidential.
A rolling landmark analysis examined whether the timing of disengagement modified the mortality excess. The study team observed that the excess risk was similar for disengagement occurring during any of the first four months after treatment initiation. Effects attenuated for disengagement occurring later in the standard treatment window. In short, interruptions in earlier treatment months carried a comparable mortality penalty; interruptions after that period were associated with reduced incremental effect on 5-year mortality in these analyses.
Relative increases in risk were largest among younger, stably housed individuals because their baseline mortality was low; a given absolute increment therefore produced a larger relative change. The largest absolute excess difference in 5-year mortality fell on people living with HIV: the excess was 5.16 percentage points (95% CI 3.76–6.60) in that subgroup. These subgroup results were obtained within the same target-trial emulation framework and adjusted for the same covariates described above.
Among people classified as LTFU, 65% had received directly observed therapy prior to disengagement. High DOT coverage did not preclude substantial disengagement and an associated mortality burden. The authors therefore proposed that tuberculosis programs may need interventions that extend beyond observed medication intake—measures that support sustained engagement across the treatment course—to reduce mortality. The paper recommended testing such interventions for effectiveness but did not evaluate any specific implementation strategies within the registry analysis.
The authors listed several limitations that affect interpretation. Key clinical severity markers—such as radiographic extent of disease and bacillary load—were absent from the surveillance registry, so residual confounding by disease severity cannot be excluded. The registry-based design also meant some patients were excluded because of incomplete records or transfers, and the linked mortality data used reduced cause categories rather than full ICD-10 detail. The analysis code and a minimal deidentified dataset are available publicly, but the underlying identifiable records require formal approval from the São Paulo State Health Department and the local ethics committee for access. These documented constraints define concrete gaps: whether unmeasured disease severity or in-treatment health changes explain part of the observed excess remains unresolved and requires data with richer clinical detail or prospective interventional evaluation.
The study ends on specific, research-oriented open questions rather than broad generalities. Which support interventions—social, clinical, or structural—most effectively prevent disengagement and reduce downstream mortality remains to be established in trials or implementation studies. Also unresolved is the extent to which unmeasured severity markers explain the excess deaths; answering that will require datasets that include radiographic and microbiological burden or prospective cohorts with granular clinical follow-up. Those are the next concrete steps implied by the registry findings and the limitations the authors reported.