Tuberculosis incidence is high in HIV-infected African children but is reduced by co-trimoxazole and time on antiretroviral therapy

The ARROW trial design, methods, and results have been described elsewhere [16, 18]. Briefly, ARROW was a randomized controlled factorial trial comparing outcomes of children randomized to either clinically- or laboratory-driven monitoring and (simultaneously in a factorial design) to either a standard three-drug or an induction-maintenance first-line ART regimen. Children were randomized to one of three arms: Arm A (lamivudine/abacavir/non-nucleoside reverse transcriptase inhibitor (NNRTI) throughout); Arm B (lamivudine/abacavir/zidovudine/NNRTI until week 36, with subsequent lamivudine/abacavir/NNRTI); or Arm C (lamivudine/abacavir/zidovudine/NNRTI until week 36, with subsequent lamivudine/abacavir/zidovudine). A total of 1,206 children and adolescents, median age 6 years (range, 3 months to 17 years), eligible for ART following WHO 2006 [19] guidelines, were enrolled between March 2007 and November 2008 from three centres from Uganda and one in Zimbabwe.

All children started once daily co-trimoxazole at enrolment if they were not already taking it, as per the WHO 2006 [19] guidelines (200 mg sulfamethoxazole and 40 mg of trimethoprim, 400 mg sulfamethoxazole and 80 mg of trimethoprim, or 800 mg of sulfamethoxazole and 160 mg of trimethoprim for a body weight of 5–15, 15–30, or >30 kg, respectively). Isoniazid preventative therapy was not routinely used, and TB screening pre-enrolment was based on standard elicitation of symptoms. After enrolment, children were followed at weeks 4, 8, and 12, and then 12 weekly for up to 5 years. Children saw a nurse at every visit who solicited specific symptoms and referred the child to the doctor if unwell. Height and weight were measured at every visit and converted to Z-scores using UK reference standards [20]. Children saw a doctor routinely every 12 weeks for assessments including a full blood count and CD4 T-cell count. Only 3 % of children were lost to follow-up; 95 % remained on first-line ART over a median of 4 years and 96 % were alive at the end of the trial. The trial included an additional (open-label) randomization 96 weeks after ART initiation to either stop or continue taking co-trimoxazole prophylaxis. Children over 3 years of age who had been receiving ART for at least 96 weeks and were using insecticide-treated bed-nets (in malaria-endemic areas) were eligible for this randomization. Children with previous P. jirovecii pneumonia were excluded. For the combined primary endpoint of death and hospitalizations, the trial showed that prolonged use of co-trimoxazole was beneficial [16].

All caregivers provided written informed consent to participate in the study and for future publication of data: children and adolescents provided consent or assent, depending on age and knowledge of HIV status. The trial was approved by ethics committees in Uganda (Joint Clinical Research Centre IRB Office), Zimbabwe (Medical Research Council of Zimbabwe), and the UK (UCL Research Ethics Committee).

Data on TB diagnosis were collected on a standardized form, from enrolment and throughout follow-up. TB was reported as either definitive or presumptive (as per WHO diagnostic criteria for HIV-infected children) [21]. A diagnosis of TB was made based on suggestive clinical features with available supportive investigations, including the tuberculin skin test, chest X-ray or other imaging, and sputum microscopy with or without culture. TB was categorized by site of infection as pulmonary, disseminated extrapulmonary, and tuberculous lymph node disease.

All TB diagnoses and causes of death were adjudicated blind to randomized arm by an independent Endpoint Review Committee (ERC) using clinical summaries of the event provided in real-time by clinicians managing the children, and all routine and non-routine laboratory data including history of TB contact, clinical presentation, microbiological and radiological investigations, and response to TB treatment. Deaths were classified as TB-related when TB was adjudicated as one of the causes of death. The ERC also adjudicated whether, in their opinion, the TB event was likely due to immune reconstitution inflammatory syndrome (IRIS); there are no validated IRIS definitions in children.

In order to exclude the possibility of relapses from previous TB infection, children with a reported past history of TB (including prevalent TB at enrolment) were excluded from the analyses.

For the co-trimoxazole analysis, time zero was the date of the co-trimoxazole randomization, which occurred more than 96 weeks after ART initiation. Time to TB diagnosis in this analysis was defined as the time from co-trimoxazole randomization to diagnosis of TB, including any recurrences in children who had had a previously confirmed TB diagnosis during follow-up, but excluding children with prevalent or prior TB at ART initiation (since these diagnoses could not be confirmed). Kaplan–Meier and multivariable Cox proportional hazards models were fitted to estimate the unadjusted and adjusted effect of stopping compared to continuing co-trimoxazole on the risk of TB diagnosis. Covariates considered for a multivariable adjusted model were as for the TB incidence analysis, measured at the time of co-trimoxazole randomization.

All analyses were performed in Stata (version 13).

Of the 1,206 children in the ARROW trial, 237 had a history of TB and were excluded, leaving 969 in the analysis (Fig. 1a). Over a median of 4 years follow-up from ART initiation, 87 children (9 %) were diagnosed with TB; 18 TB diagnoses (14 pulmonary and 4 extra-pulmonary) were adjudicated as not TB by the ERC based on established alternative diagnoses, leaving 69 incident TB cases (61 pulmonary, 3 extra-pulmonary, and 5 lymph node TB) and 900 children with no TB at the end of follow-up (Fig. 1a). A total of 55 (80 %) of the 69 TB cases adjudicated by the ERC were presumptive diagnoses; 14 (20 %) were definitive.

Based on the 69 ERC-adjudicated TB cases, the overall incidence of TB was 1.9 per 100 child-years (95 % CI, 1.5–2.4), which varied over time: 8.8/100 child-years for the first 12 weeks, 2.7/100 child-years between 12 and 52 weeks, and 1.2/100 child-years thereafter (Fig. 2). The smoothed hazard function indicates a peak in the first few weeks following ART initiation with the risk falling rapidly thereafter.

Table 1 shows the characteristics of children at ART initiation. Overall, the median age was 5 years (IQR, 2–9), 488 (50 %) were male, and 712 (73 %) were recruited from sites in Uganda and 257 (27 %) from Zimbabwe. Children with an ERC-adjudicated incident TB diagnosis were marginally more likely to be aged <3 years at ART initiation (45 % vs. 34 % without TB, P = 0.07), and to be female (59 % vs. 49 %, P = 0.09); they were significantly more likely to have a lower CD4 percent (median 8 % vs. 13 % for those without TB, P <0.001), lower height- and weight-for-age Z scores (median −3.0 and −2.8, respectively, vs. −2.3, and −1.8, respectively, both P <0.001); and to already have WHO stage 3 or 4 HIV disease (75 % vs. 62 %, P = 0.03) at ART initiation. Additional file 1 shows the same characteristics including children (n = 237) with a history of TB, indicating broadly similar results. Of note, a history of TB was not found to be associated with future TB in this cohort.

In univariable analysis, a 3- or 4-drug ART regimen that included nevirapine (compared to efavirenz) was associated with an increased risk of TB (P = 0.03). A lower proportion of children in Arm C (initial 4-drug regimen decreased to 3-drug NRTI-only regimen after 36 weeks) developed TB compared to other groups (P = 0.06). There was no difference in TB incidence between patients randomized to either clinical- or laboratory-driven monitoring.

Table 2 shows unadjusted and adjusted hazard ratios from the multivariable Cox model, including effects of the time-updated covariates. The following factors were independently associated with the risk of developing TB: younger age at ART initiation (P = 0.04), female sex (P = 0.01), lower current weight-for-age (P = 0.001), and lower current CD4 percent (P <0.001). Current height-for-age and pre-ART WHO stage did not independently predict TB risk. Baseline weight-for-age, height-for-age, and CD4 percent were not significant when included with the corresponding time-updated variables (data not shown).

Although a nevirapine-containing first-line ART regimen was univariably associated with risk of TB, this was no longer significant when age and current weight were included in the model, likely due to confounding, as all children under 3 years started nevirapine. However, a reduced risk of TB remained for those randomized to ART regimen C (induction-maintenance strategy) compared to regimen A (3-drug NNRTI-based regimen throughout) (P = 0.01). To investigate this further we restricted the analysis to TB incidence within 12 weeks of starting ART (with appropriate censoring) when Arms B and C were the same (i.e. a 4-drug ART regimen). Here we found no evidence of a difference for Arm B versus A (P = 0.6), with some limited evidence of a difference for Arm C versus A (P = 0.06), suggesting that this could be a chance effect.

Twenty-two TB cases (32 %) were reported as likely due to TB-IRIS, with a median time to diagnosis of 20 weeks following ART initiation (IQR, 4–42) versus 100 (24–151) in non-IRIS cases. Of the IRIS cases, 17 (77 %) were presumptive TB, and notably 17 (77 %) occurred in females.

Censoring non-IRIS cases in the same multivariable Cox model to estimate the impact of factors on the cause-specific hazard of developing TB-IRIS before TB non-IRIS, showed broadly similar results to Table 2, although power was lower with small numbers of events; the only significant associations were with weight and CD4 percent (data not shown). Censoring TB-IRIS cases in the same multivariable Cox regression analysis to estimate the impact of factors on the cause-specific hazard of developing TB non-IRIS before TB-IRIS, showed similar results in terms of identifying risk factors, except that sex was no longer significant: males versus females (HR = 1.28; 95 % CI, 0.72–2.29; P = 0.39).

After 96 weeks on ART, 760 children were eligible for the co-trimoxazole randomization; of these, 138 had a previous history of TB and were excluded, leaving 622 children in the analysis (Fig. 1b); 310 (50 %) children were randomized to continue co-trimoxazole and 312 (50 %) to stop, and baseline characteristics of children were well balanced by randomized group (Additional file 2: Table S2). Overall, 51 % were female and median age was 4 years. The median weight-for-age and height-for-age scores were −1.3 and −1.9, respectively, and CD4 percent was 32 %.

There were a total of 22 TB diagnoses over a median of 2 years follow-up after co-trimoxazole randomization. Two TB diagnoses were adjudicated as not TB by the ERC leaving 20 incident TB cases and 602 children with no TB at the end of follow-up. Of 20 adjudicated TB cases, 5 were definitive (25 %) and 15 (75 %) were presumptive.

Time to TB diagnosis by randomized group is shown in Fig. 3. There were 5 adjudicated TB cases in the continue arm and 15 in the stop arm (HR = 3.0; 95 % CI, 1.1–8.3; P = 0.028), comparing stopping to continuing co-trimoxazole. Of the 5 cases in the continue arm, 1 was definitive and 4 were presumptive; of the 15 cases in the stop arm, 4 were definitive and 11 were presumptive. Adjusting for risk factors identified from the main analysis (including age, sex, CD4 percent, weight and current ART regimen) gave similar results (HR = 2.7; 95 % CI, 1.0–7.6; P = 0.055). Table 3 shows the characteristics of the children at the time of co-trimoxazole randomization by future TB diagnosis. On multivariable adjustment, along with stopping co-trimoxazole, only a lower current CD4 percent (P <0.001) was independently associated with a higher risk of TB. Repeating the analysis including children (n = 138) who had a history of TB gave similar results (data not shown).

The ARROW trial data are held at MRC CTU at UCL, which encourages optimal use of data by employing a controlled access approach to data sharing [42]. All requests for data are considered and can be initiated by contacting mrcctu.ctuenquiries@ucl.ac.uk.