Mycobacterium tuberculosis infection and tuberculosis disease in the first decade of life: a South African birth cohort study

Abstract

Background: Paediatric tuberculosis leads to more than 200 000 deaths annually. We aimed to investigate the incidence of Mycobacterium tuberculosis infection and tuberculosis disease in the first decade of life in the Drakenstein Child Health Study (DCHS), a South African cohort in a community with high tuberculosis and HIV incidence.

Methods: In this prospective birth cohort study, we enrolled pregnant women aged 18 years or older who were between 20 and 28 weeks' of gestation in a peri-urban setting outside of Cape Town, South Africa. We followed up their children for tuberculosis until age 10 years. To measure M tuberculosis infection tuberculin skin tests were administered to children at age 6 months, 12 months, and then annually in children with a negative test, and at the time of a lower respiratory tract infection. Tuberculin skin test conversion was defined by an induration reaction of 10 mm or more. To measure tuberculosis disease, active surveillance was done throughout follow-up. Each episode of presumed tuberculosis disease was investigated using sputum induction, tested with Xpert MTB/RIF and liquid culture for M tuberculosis. Survival analyses were performed and multivariable Cox regression was used to measure factors associated with M tuberculosis infection or disease.

Findings: Between March 5, 2012, and March 31, 2015, 1137 women and their 1143 children (248 [21·7%] of 1143 children were HIV-exposed, two [0·2%] children with HIV) were included in the analysis. Children were followed up for 8870 person-years (median follow-up 9·1 years [IQR 8·2-10·2]). The annual risk of tuberculin conversion during follow-up was 6·6 infections per 100 person-years (95% CI 5·8-7·3) but ranged from 4-9 infections per 100 person-years over the follow-up period. 98 children developed tuberculosis (1105 cases per 100 000 person-years; 95% CI 906-1347). The cumulative hazard of tuberculin conversion was 36% (95% CI 32-41) at age 8 years and the cumulative hazard of tuberculosis disease was 10% (8-12) at age 10 years. Preventive treatment was associated with a reduction in tuberculosis disease among children who had tuberculin conversion (adjusted hazard ratio 0·23 [95% CI 0·12-0·47]). Most cases of tuberculosis disease (78 [79%; 95% CI 69-86] of 98 children) occurred among children who had tuberculin skin test conversion but were not administered preventive treatment.

Interpretation: In this prospective South African birth cohort, M tuberculosis transmission was consistently high throughout the first decade of life leading to approximately 10% of children developing tuberculosis disease. A multipronged approach to decrease paediatric tuberculosis is needed that combines preventive treatment for children at risk, reducing community M tuberculosis transmission, and active case finding.

Funding: Bill & Melinda Gates Foundation, Medical Research Council South Africa, and National Research Foundation South Africa.

Figure 1

Study flow *Loss of pregnancy due to miscarriage, stillbirth, or intrauterine death (23 infants [including one set of twins]). †Including four pairs of twins and one set of triplets.

Figure 2

The annual percentage risk of tuberculin skin test conversion (A) and tuberculosis disease incidence per 100 000 person-years (B) Panel A describes the probability that a tuberculin skin test conversion will occur at a given age conditional on no tuberculin skin test conversion having occurred before that. Because this was a non-parametric model, we were unable to compute results preceding the first observed event. Moreover, some minimum number of events are needed before an instantaneous hazard function can be computed. Many children did not have tuberculin skin test results after the 8-year visit; therefore, panel A stops before the end of total study follow-up. Panel B shows tuberculosis disease incidence per 100 thousand person-years over the entire cohort through a hazard function or the probability that a tuberculosis diagnosis will occur at a given age conditional on no tuberculosis diagnosis having occurred before that. This specifies a kernel function to be used in calculating the weighted kernel-density estimate smoothing of the estimated hazard over age. The solid line in each graph represents modelled estimates and the shaded bands are 95% CIs.

Figure 3

Tuberculosis-free survival in the first decade of life Kaplan–Meier estimates of tuberculosis-free survival in children from the Drakenstein Child Health Study. The solid line represents modelled estimates and the shaded bands are 95% CIs. The hazard ratio and CI were calculated using a Cox proportional-hazards model. There was no indication of non-proportional hazards.