Why have trials of isoniazid preventive therapy among people with HIV infection not demonstrated an effect on mortality?

Several randomized controlled trials, mostly conducted in African countries in the early 1990s, demonstrated the effectiveness of isoniazid preventive therapy (IPT) in preventing new episodes of tuberculosis among people living with HIV, particularly among individuals with a positive tuberculin skin test (TST) [1]. However, only one trial, conducted in Haiti, found a major effect of IPT on mortality [2] and in meta-analyses, the effect of IPT on death was modest [among TST positive, risk ratio 0.74, 95% confidence interval (CI) 0.55–1.00; overall: risk ratio 0.95, 95% CI 0.85–1.06] [1]. Is it plausible that IPT could have an effect on mortality that was not evident in randomized controlled trials? Most of the randomized controlled trials of IPT included in the most recent meta-analysis [2–8] were conducted prior to availability of antiretroviral therapy (ART). Most trials had inclusion criteria which selected relatively healthy individuals (Table 1), presumably intending to minimize the risk that individuals with undiagnosed active tuberculosis would be enrolled. In line with this, although few of the studies measured participants' CD4 cell count at baseline, in the Kenyan study [4], the median CD4 cell count was around 330 cells per μl. Nonetheless, all-cause mortality was high, approaching 10 per 100 person-years, and exceeded tuberculosis incidence rates in all studies (Table 1). Therefore, it is unlikely that an effect on all-cause mortality went undetected because death was too rare an event.Table 1: Tuberculosis screening methods, tuberculosis incidence and mortality in randomized controlled trials of isoniazid preventive therapy.Isoniazid has a narrow spectrum of antibacterial activity, and so logically any direct effect of isoniazid on mortality must be by reducing death among individuals with tuberculosis. Tuberculosis case fatality (death during tuberculosis treatment) among people with HIV infection in the pre-ART era generally exceeded 10% [9–12]. Earlier detection of active tuberculosis results in lower tuberculosis case fatality: in South African gold mines, prior to ART availability, individuals whose tuberculosis was detected by the mine's radiological screening programme had lower case fatality than those self-presenting with symptomatic tuberculosis, even after controlling for confounding factors [9]. Examination of the methods of the randomized trials of IPT reveals the investigators' vigorous efforts to detect active tuberculosis both before enrolment and during follow-up (Table 1). In all studies, screening prior to enrolment included a symptom screen and chest radiograph, and in some studies one or more sputum samples were sent for culture. Follow-up procedures were not always described, but participants were reviewed monthly in most studies, presumably with a symptom screen, and in some studies chest radiographs were taken periodically during follow-up. Follow-up was less intensive in the study by Pape et al., comprising review every 3 months, and it is interesting that this was the only study in which lower mortality was seen in the IPT arm (reduction in mortality >50% among all patients regardless of TST status, with wide confidence intervals). In addition to this intensive screening for active tuberculosis, participants in the IPT trials in developing countries very likely had enhanced access to healthcare, as a benefit of the study. This is likely to have resulted in lower tuberculosis case fatality among study participants, compared to nonstudy tuberculosis patients, because of earlier detection of active tuberculosis, but this effect would have been seen equally in all study groups, including those allocated placebo. Screening for active tuberculosis is an essential part of an IPT programme, which should be seen as a package comprising both components. However, in randomized controlled trials, the intensive screening component was implemented in both arms and, thus, only the IPT component was tested. Individually-randomized controlled trials may, therefore, have underestimated the impact of an IPT programme on tuberculosis case fatality. Further, the effect of an IPT programme at clinic level may extend beyond those individuals who receive IPT, if effective screening for active tuberculosis prior to IPT results in earlier diagnosis and treatment of active tuberculosis among tuberculosis suspects who are identified and excluded from IPT, thus reducing tuberculosis case fatality. The effect of the ‘package’ would only be observed if clusters (for example, clinics) were randomized to implement an IPT programme vs. standard of care, and the outcomes were determined among all individuals with HIV, a trial which would now be difficult to justify ethically, given the clear evidence that IPT reduces tuberculosis incidence. It is also plausible that the association between IPT and mortality could be different in the presence of combination ART. ART rapidly reduces the risk of HIV-related mortality [13], of which tuberculosis is the most important in developing countries; other major causes of death identified at autopsy include severe bacterial infections and cryptococcal disease [14–17]. Tuberculosis remains an important cause of early death on ART [13]. ART and IPT have complementary roles in reducing tuberculosis incidence [18,19]; it is possible that the effect of ART in reducing mortality may be more rapid with respect to diseases that reflect profound immunosuppression, such as cryptococcosis, than tuberculosis, which continues to occur at high rates even after initiation of ART [20,21]. Thus, as initiation of ART rapidly reduces nontuberculosis causes of HIV-related death, an effect of IPT on tuberculosis case fatality may emerge. This could be particularly evident during the first 3 months on ART, at the time when latent tuberculosis may reactivate and subclinical active tuberculosis may be unmasked, in the context of immune reconstitution [22]. In our paper in this supplement [23], we analysed data from individuals starting ART with a median CD4 cell count around 155, thus a sicker population than that in the pre-ART era trials, and also with a higher median age. All-cause mortality in the first year on ART was 11.1 per 100 person-years among individuals who did not receive IPT, similar to other cohorts initiating ART in sub-Saharan Africa [13], but was substantially lower (3.75 per 100 person-years) in those who received IPT, and this difference persisted after adjusting for known confounders and in sensitivity analyses exploring possible biases. The great majority of individuals started IPT and ART at almost the same time. As this analysis was based on programme data, we cannot compare the intensity of screening for active tuberculosis among those who did or did not get IPT, but it is likely that this activity was done more diligently among individuals prior to and while taking IPT, and this could have played a role in reducing mortality. Randomized controlled trials are in progress to investigate whether the addition of IPT to ART reduces tuberculosis incidence [1], and the results of these studies will help to guide policy. However, if both intervention and control group participants are intensively screened for active tuberculosis in the same way, these trials will not detect a survival benefit attributable to early detection of tuberculosis cases. Close examination of the trees may have obscured our view of the wood: IPT is a package of care comprising screening for tuberculosis, treatment for those with active disease and IPT for those without. By focussing more narrowly on the IPT component, clinical trials may have underestimated the benefits of the package and, thus, people living with HIV may have more to gain from this intervention than has been previously recognized. Conflicts of interest: None.