The authors responded to critique of their original dispatch describing observed tuberculosis (TB) incidence after TB preventive therapy (TPT) among persons living with HIV who recently initiated antiretroviral therapy (ART). The initial analysis used routinely collected programmatic data with a descriptive intent rather than to establish causal effects of TPT on TB. In light of concerns raised about potential analytic bias, the team undertook a re-analysis to better classify exposure and person-time.
The correspondence acknowledges that comparing groups defined by TPT completion can be vulnerable to immortal time bias. This bias arises when periods during which an outcome cannot occur are incorrectly attributed to an exposure group, which can spuriously amplify apparent protective associations for those classified as having completed therapy. The authors note that immortal time bias is a common issue in published observational studies of TB after TPT.
To address this concern, the investigators treated TPT exposure as a time-varying variable in the revised analysis. This approach allowed follow-up person-time to be allocated appropriately across three exposure states as they occurred during follow-up: not started, incomplete, and complete TPT. By classifying person-time dynamically rather than assigning individuals to a single static exposure category, the analysis reduces misattribution of event-free follow-up that would otherwise cause immortal time bias.
After implementing the time-varying exposure specification, the effect estimates changed in magnitude but maintained the same overall pattern observed in the original analysis. The adjusted incidence rate ratios (IRRs) reported in the re-analysis were smaller than in the earlier report, consistent with bias correction, but remained statistically significant. Specifically, compared with time periods following TPT completion, TB incidence was higher during incomplete TPT (adjusted IRR 1.4, 95% CI 1.3–1.5) and higher among those who did not initiate TPT (adjusted IRR 2.2, 95% CI 1.9–2.5).
The reduction in effect size after correcting for immortal time bias is consistent with expectations when that bias is addressed. The persistence of the same directional pattern—higher TB incidence during incomplete or no TPT compared with periods after completion—supports confidence in the original descriptive findings. The authors emphasize two main implications: routine programmatic data can be valuable for understanding real-world TB risk after TPT in people with HIV starting ART, and analytic methods must be chosen carefully when interpreting observational program data to avoid misleading inferences.
The response lists the author team and their institutional affiliations, including US Centers for Disease Control and Prevention staff in Maputo and Atlanta, members of the US Agency for International Development in Maputo, and officials from the Ministry of Health, Mozambique. The correspondence cites the original dispatch and related literature on TB after TPT and methodological considerations in observational analyses.
References cited in the source include the authors' earlier dispatch describing TB after TPT in Mozambique, an observational study from India on long-term protection after TPT, and other recent reports of TB following TPT completion. The letter makes no new clinical recommendations and clarifies that the re-analysis was intended to address analytic bias rather than to produce causal estimates of TPT effectiveness.