Excess mortality in people living with HIV who received tuberculosis (TB) treatment with negative TB tests: a multinational study of data quality and validation

Background: Observational studies typically use routinely collected (RC) data to investigate TB treatment outcomes, but these data may be error-prone. Using RC data together with validated data on a subsample, we explored the association of TB testing results with mortality and TB treatment outcomes in persons living with HIV (PLWH). Method: We used RC data from two large HIV observational cohorts to evaluate the association of TB test result (positive/negative/unknown) at treatment start with death and the composite unfavorable outcome (treatment failure, loss-to-follow-up [LTFU], recurrence, or death) within 1.5 years of TB treatment start. We designed and implemented an optimal multi-wave validation study on a subsample of PLWH. We fitted logistic regression models using RC data only and combining them with the chartvalidated data using a generalized raking approach, controlling for age, sex, body mass index (BMI), antiretroviral therapy, CD4 count, and region. Results: RC data were extracted from 22,587 PLWH; 1122 were selected for validation, with 842 validated. We observed large discrepancies between validated and unvalidated data. We found that PLWH who started TB treatment with negative test results had higher odds of death compared to those who started treatment with a positive test: adjusted odds ratio (aOR)=1.22 (95% confidence interval [CI]=[1.03-1.45]) and aOR=3.94 (95% CI=[1.73-9.00]), using RC data only and RC plus validated data, respectively. Results for the composite outcome were inconclusive. Conclusion: In PLWH treated for TB, bacteriological confirmation was associated with lower mortality. Data validation for observational research using RC clinical data care is needed.