Reducing the Burden of Tuberculosis Presenting during the Initial Months of Antiretroviral Therapy in Resource‐Limited Settings

To the Editor—The high burden of tuberculosis within antiretroviral therapy (ART) programs in resource-limited settings is challenging [1, 2]. We read with interest the article by Koenig et al [3] from Haiti in which they reported a very high mortality rate (27% at 1 year) among patients with tuberculosis diagnosed during the first 3 months of ART. This observation is important and is entirely consistent with other data from sub-Saharan Africa [1, 2]. However, it was also reported that, in both adjusted and unadjusted analyses, mortality in this patient group was substantially greater than that among patients with either prevalent tuberculosis diagnosed before the initiation of ART or incident tuberculosis diagnosed at some time beyond 3 months of ART [3]. Data from the latter 2 groups were pooled to form a combined comparator group with a low overall tuberculosis risk. The much higher mortality risk among patients with tuberculosis presenting during early ART compared with that among patients with tuberculosis at baseline differs from the findings of previous studies [1, 2], and we are concerned that this observation could be misinterpreted as showing that ART has a deleterious impact on survival. We wonder whether this may be the result of patient selection. The mortality rate among patients with advanced immunodeficiency just before starting ART in resource-limited settings is extremely high, and tuberculosis during this period is difficult and time-consuming to diagnose [4–7]. Thus, during this period many deaths occur among patients with tuberculosis (whether or not it has been diagnosed). Only selected patients in whom tuberculosis diagnosis was possible and who survived long enough to start ART could form the pre-ART group. Such patient selection may have substantially diminished the observed mortality risk in the pre-ART group. This would be consistent with the Kaplan-Meier plot, which shows an unusually low frequency of death during the first 6 weeks of tuberculosis treatment in the comparator arm containing this patient group. Notwithstanding this potential limitation, we agree that tuberculosis is very likely to be an important contributor to early mortality during ART [4]. The key question is how to tackle this problem. Koenig and colleagues suggested that many patients with incident tuberculosis during early ART are likely to have active tuberculosis that was present at baseline but remained undiagnosed. In agreement with this suggestion, we recently estimated that, in our South African ART cohort, ∼40% of the tuberculosis cases presenting during this period were due to the “unmasking” of asymptomatic or minimally symptomatic disease that was present at baseline [8]. In a more recent study, moreover, all patients entering our program without a preexisting tuberculosis diagnosis were (regardless of symptoms) systematically screened by culturing induced sputum samples. The yield of culture-confirmed tuberculosis diagnoses among these patients at baseline was extremely high (25%) [9] and was 12-fold higher than what had been diagnosed during routine program screening in preceding years [1]. Of note, during follow-up the proportion of patients who presented with tuberculosis during the first 3 months of ART was, correspondingly, 12-fold lower than that previously observed. Thus, following the observation by Koenig and colleagues of a very high mortality rate among patients presenting with tuberculosis during the first 3 months of ART, we have found that effective baseline screening substantially reduces incident tuberculosis during this period. This suggests a need for routine pre-ART tuberculosis screening with high-sensitivity tests (such as automated liquid culture of sputum) and highlights an urgent need for much more rapid and sensitive diagnostics for use in this clinical setting. It remains to be demonstrated, however, whether this strategy would also reduce mortality risk. Financial support. S.D.L. is funded by the Wellcome Trust, London, United Kingdom. R.W. is funded in part by the US National Institutes of Health (RO1 grant A1058736-01A1 and Comprehensive International Program of Research on AIDS grant 1U19AI53217-01). D.J.E. is funded by the US National Institutes of Health through the International Clinical Research Fellows Program at Vanderbilt (R24 TW007988). Potential conflicts of interest. All authors: no conflicts.