During 2004–2008, children aged 0–14 years diagnosed with World Health Organization (WHO) Stage IV disease, WHO Stage III disease if <12 months old, or with certain clinical conditions including TB, were eligible for ART regardless of CD4⁺ T-cell (CD4) % or count. Children not yet eligible by clinical criteria could be eligible for ART based on age-dependent CD4% or count criteria, as described in the 2006 WHO treatment guidelines.¹⁰ Recommended ART regimens for children with HIV-1 infection include two nucleoside reverse transcriptase inhibitors (NRTI) and either nevirapine (NVP) or efavirenz (EFV). For children with TB co-infection, NVP was replaced with either EFV or a third NRTI, depending on the age of the child. For first-line ART for HIV-2-infected children, two NRTIs and a ritonavir-boosted protease inhibitor were recommended.

Côte d’Ivoire guidelines recommend that pediatric ART patients be seen frequently initially (at 2 weeks, monthly for 3 months and quarterly thereafter) until stable, and then at least every 6 months,¹⁰ to evaluate disease progression or improvement. At each visit, standard Ministry of Health (MOH) recommended medical records are completed. Documentation of whether TB screening was performed and results of any TB diagnostic tests is recommended at all clinic visits. Recommended TB screening practices for children included assessment for 1) poor weight gain, 2) current fever, 3) chronic cough, 4) TB contact history, and 5) night sweats. Recommended TB tests for all child TB suspects included 1) Mantoux skin test, 2) two sputum samples (or gastric aspirates) sent for smear microscopy, 3) one sputum (or gastric aspirate) sent for TB culture, and 4) chest X-ray. If extra-pulmonary TB was suspected, lymph node aspiration, lumbar puncture and abdominal ultra-sound were additional diagnostic tests available at some facilities. In reality, only two laboratories provided TB culture; very few TB suspects would therefore have had sputum samples or gastric aspirates sent for TB culture.

This was a nationally representative retrospective cohort study. By 1 January 2008, about 3000 children had initiated ARTat 64 health facilities.¹¹ To improve study feasibility, only facilities with >10 pediatric ART enrollees by 1 January 2008 were considered eligible. Of 30 eligible facilities, 29 agreed to participate. According to MOH records, these 29 facilities had enrolled 2820 (94%) of all 3000 children enrolled nationally during 1998–2008. Because the medical records of ART enrollees before 2004 had considerable missing data, the protocol excluded 427/2820 enrollees (15%) who started treatment before 2004. A further 195 (7%) records could not be found, and 88 (3%) had been transferred with the child to another facility. All remaining 2110 records were included in the study. Data were collected from the MOH-recommended ART medical records by trained data abstractors from November 2009 to March 2010.

Children taking anti-tuberculosis treatment at ART initiation were considered prevalent TB cases. Children starting anti-tuberculosis treatment (for pulmonary or extra-pulmonary TB) during ART were considered incident TB cases.

All variables on MOH-recommended ART records were assessed as possible risk factors for TB (Table 1). Weight was recoded as weight-for-age Z-score (WAZ score), using Centers for Disease Control and Prevention (CDC) growth curves for children aged 5–14 years, and WHO curves for children aged 0–<5 years.

Data were analyzed using Stata 11 (StataCorp, 2009, Stata Statistical Software, Release 11, College Station, TX, USA). Missing data are reported for each covariate of interest and were assumed to be missing at random (MAR). If <30% of observations were missing data for a baseline covariate of interest, multiple imputation with chained equations was used to impute the missing data.¹² The ICE (imputation by chained equations)¹³ procedure in Stata was used to create 20 imputed data sets for the key outcome of interest: TB incidence. The imputation model included the event indicator, all study variables and the Nelson-Aalen estimate of cumulative hazard.¹⁴ In all analyses involving imputed data, estimates were combined across data sets according to Rubin’s rules.¹² Using imputed data, associations between baseline covariates and prevalent TB were assessed using bivariate logistic regression, with random effects specified for each facility.

As ART medical records did not routinely document the date of anti-tuberculosis treatment cessation, time at risk for incident TB among prevalent TB cases could not be estimated. As in other studies, in time-to-event analysis, patients with prevalent TB were excluded from the cohort at risk.¹⁵ Patients who were included in the cohort at risk, but were not diagnosed as having incident TB during follow-up, were censored at the most recent visit, date of death if death occurred, or date of transfer if transferred.

Cox proportional hazards regression models were used to estimate non-adjusted and adjusted hazard ratios (HR and aHR), with random effects specified for ART facility. The proportional hazards assumption was assessed using visual methods and the Grambsch and Therneu test.16 Kaplan-Meier curves were used to examine the cumulative probability of remaining undiagnosed with TB over time stratified by baseline variables.

This study was approved by the Ivorian Ethics Review Committee and the Institutional Review Boards of the United States CDC and the Harvard School of Public Health.

Among 2110 children starting ART during 2004–2008, the median age was 5.1 years, and 150 (7%) were receiving anti-tuberculosis treatment (prevalent TB cases). The median age of the prevalent TB cases was older than that of children considered TB-free (6.7 vs. 5.0 years, P < 0.001) (Table 1). The majority of the children (99%) were HIV-1-infected; only 15 (0.7%) were HIV-2-infected, and five (0.2%) were dually HIV-1 and −2 seroreactive. HIV type was not associated with prevalent TB (P = 0.553).

Markers of advanced HIV disease were common among ART enrollees; 29% had WHO Stage IV, 42% had severe under-nutrition (WAZ <−3), and median CD4% was 11%. Compared with children considered TB-free at ART initiation, children with prevalent TB were more likely to have WHO Stage IV disease (72% vs. 25%, P < 0.001), to have severe under-nutrition (59% vs. 40%, P < 0.001), and to have a lower median CD4% (9.0% vs. 11.0%, P = 0.037).

Compared with children assessed as TB-free, prevalent TB cases were more likely to start ART at tertiary care facilities (51% vs. 32%, P = 0.011) and at facilities lacking nutritional support programs (83% vs. 65%, P = 0.007).

Overall, 1012 (48%) children had documentation of TB screening before ART initiation; 723 (34%) were screened for chronic cough, 565 (27%) for fever, 391 (19%) for TB contact, 366 (17%) for weight loss, and 228 (11%) for night sweats. Only 10% of children had documentation of screening for all five symptoms. The proportion of medical records with documentation of any TB screening before the start of ART declined from 63% to 46% during 2004–2008. The proportions of patients screened for at least one TB symptom before ART varied widely by clinic, from 0% to 93%. At 25/29 clinics, <80% of ART enrollees had some documentation of screening for TB (Figure 1).

Among the 1960 children assessed as TB-free at the start of ART, 56 (3%) were documented to start TB treatment during 4190 child-years (cy) of follow-up, at a rate of 1.34 cases/100 cy (95% confidence interval [CI] 1.03–1.74). Of 56 TB cases, 8 were extra-pulmonary TB (0.19 cases/100 cy, 95%CI 0.09–0.37) and 48 were pulmonary TB (1.14 cases/100 cy, 95%CI 0.86–1.52). The overall TB incidence rate was highest in days 0–90 of ART, at 6.28/100 cy (95%CI 4.31–9.16). This rate declined to 2.52/100 cy (95%CI 1.36–4.69) during days 91–180 of ART, and 0.56/100 cy (95%CI 0.36–0.89) thereafter.

A 1-year increase in age was borderline associated with increased TB incidence (aHR 1.08, 95%CI 0.99–1.15, P = 0.067) (Table 2). Each one-unit increase in WAZ was associated with a 13% reduction in TB incidence (aHR 0.87, 95%CI 0.77–1.00, P = 0.047) (Figure 2). Maternal and paternal orphan status and severe anemia were not associated with TB incidence risk.

TB prevalence at enrollment on ART (7%) is higher than that reported from previous studies in Western Kenya (3.6%),¹⁷ Côte d’Ivoire (2–4%),^8,9 and Zambia (5.7%),¹⁸ but is lower than that reported from Johannesburg, South Africa (29%).¹⁹ There could be many reasons for the variation in TB prevalence at ART initiation, including variations in TB screening and diagnostic algorithms,²⁰ variations in the degree of immune suppression,⁷ and variations in background TB incidence in the general population.¹⁹ Earlier ART enrollment for HIV-infected children could reduce the burden of TB among ART enrollees.^7,17 To facilitate earlier ART enrollment, expansion of early infant diagnostic and referral services would be needed.¹¹

Our reported TB incidence (1.34/100 cy), equivalent to about 1340 per 100 000 population, is considerably higher than the national average of 399/100000.²¹ Our TB incidence rate is similar to previous Ivorian reports (1.3–2.1/100 cy),^8,9 but lower than reported South African incidence rates among children starting ART (6.4/100 cy).²² Higher background TB incidence in South Africa (971 vs. 399/100 000),²¹ or better TB screening and diagnostic capability in South Africa, might explain the different rates.⁷

As in other studies,^17,23,24 TB incidence in our cohort was highest in days 0–90 of ART, and declined thereafter. Several mechanisms might explain the high early TB incidence rates.²⁴ Some early incident TB may be new active TB disease.²⁴ However, the majority of TB cases detected and treated in early ART are probably due to 1) ‘unmasking’ of subclinical disease present at ART initiation, or 2) worsening of symptomatic disease that went unscreened or undiagnosed before the start of ART.²⁴ Active disease may be subclinical at ART initiation because TB symptoms are dependent on both bacillary burden and immune response. The immune reconstitution associated with early ART can unmask new TB signs and symptoms, which in turn will initiate TB diagnostic or trial treatment protocols.²⁴ Among symptomatic children who went undiagnosed through failure to screen or diagnose pediatric TB,⁷ ART could worsen TB symptoms, prompting more intensive TB diagnostics or a clinical trial of anti-tuberculosis treatment.²⁴ As in other retrospective cohort studies,²² our reported TB prevalence and incidence rates may be underestimates due to failure to screen for and/or diagnose TB disease, or overestimates due to incorrect trials of anti-tuberculosis treatment in children with clinical TB diagnoses.

TB screening practices were suboptimal, appeared to worsen over time, and varied widely by clinic. Operational research assessing TB screening practices among pediatric ART enrollees is limited; one Thai study reported documentation of TB screening before pediatric ART for 60–90% of patients.²⁵ Poor documentation of, and declining clinician compliance with, TB screening are worrying, because undiagnosed TB at the start of ART is an important cause of morbidity and mortality.²⁶ Targeting the 25 clinics where <80% of pediatric ART enrollees were screened for at least one TB symptom with intensified training and supervision could reduce TB-associated morbidity and mortality among children starting ART.²⁰

Markers of advanced HIV disease (advanced WHO stage, under-nutrition and low CD4%) were predictive of prevalent TB,^8,17 and higher WAZ at enrollment was protective against incident TB. Earlier initiation of ART, before advanced HIV disease, could reduce the TB burden among HIV-infected children.¹⁷

The association between lower WAZ and incident TB has been reported previously.^17,23 Possible explanations include the following: 1) active TB at ART enrollment, a possible cause of low WAZ, went undiagnosed due to insensitive diagnostic tests or failure to screen for TB;²⁷ 2) clinicians made presumptive diagnoses of TB after ART initiation because WAZ gain was suboptimal;¹⁷ or 3) under-weight TB-free children at ART initiation are more at risk of developing active TB during follow-up due to malnutrition-induced impairment of cell-mediated immunity.²⁸

To address failure to diagnose active TB at start of ART, it is essential to monitor and ensure TB screening before ART initiation. In addition, the rollout of the GeneXpert^® MTB/RIF assay (Xpert, Cepheid, Sunnyvale, CA, USA), whose sensitivity in diagnosing culture-positive TB in HIV-infected children is twice that of smear microscopy (76% vs. 38%), should be accelerated.²⁷ To address the possibility that malnutrition is a root cause of the development of active TB during ART, there may be a role for expanding existing nutrition programs.²⁸ In our study, lack of a nutrition program at the clinic was associated with prevalent TB at ART initiation (P = 0.007), and was marginally associated with TB incidence in crude (P = 0.075), but not adjusted (P = 0.323), analysis. Further evaluation of pediatric HIV food supplementation programs are needed to assess their benefit and cost-effectiveness.²⁸

As in other studies,¹⁷ increased age was associated with prevalent TB and incident TB during follow-up. This may be due to more advanced immune suppression in older children at ART enrollment, or difficulties in diagnosing TB in younger children.¹⁷

This report has several limitations. First, due to the lack of data on the timing of anti-tuberculosis treatment cessation among prevalent TB cases, prevalent TB cases were excluded from the cohort at risk to estimate TB incidence. However, in a sensitivity analysis (data not shown), estimates of TB incidence with all patients included in the cohort at risk (n= 2110) did not differ from our reported TB incidence rate (n= 1960). Second, as described earlier, reported TB prevalence and incidence rates may be underestimates or overestimates of true TB burden due to difficulties in diagnosing pediatric TB. Third, missing data on covariates probably introduced non- differential measurement error. Fourth, some incident TB may have been prevalent cases, for whom the results of diagnostic tests only returned after ART initiation. Finally, lack of information on which children screened positive for TB, and, among TB suspects, which children were tested for and diagnosed with TB, limits our ability to assess clinician compliance with TB diagnosis guidelines.