Risk factors for Mycobacterium tuberculosis infection in 2–4 year olds in a rural HIV-prevalent setting

SUMMARY BACKGROUND: Mycobacterium tuberculosis infection in children acts as a sentinel for infectious tuberculosis. OBJECTIVE: To assess risk factors associated with tuberculous infection in pre-school children. METHOD: We conducted a population-wide tuberculin skin test (TST) survey from January to December 2012 in Malawi. All children aged 2–4 years residing in a demographic surveillance area were eligible. Detailed demographic data, including adult human immunodeficiency virus (HIV) status, and clinical and sociodemographic data on all diagnosed tuberculosis (TB) patients were available. RESULTS: The prevalence of M. tuberculosis infection was 1.1% using a TST induration cut-off of 15 mm (estimated annual risk of infection of 0.3%). The main identifiable risk factors were maternal HIV infection at birth (adjusted OR [aOR] 3.6, 95%CI 1.1–12.2), having three or more adult members in the household over a lifetime (aOR 2.4, 95%CI 1.2–4.8) and living in close proximity to a known case of infectious TB (aOR 1.6, 95%CI 1.1–2.4), modelled as a linear variable across categories (>200 m, 100–200 m, <100 m, within household). Less than 20% of the infected children lived within 200 m of a known diagnosed case. CONCLUSION: Household and community risk factors identified do not explain the majority of M. tuberculosis infections in children in our setting.

Accurately identifying M. tuberculosis infection in children can highlight recent failures in control measures in the community, 1 and identify a population at high risk of progression to active disease. 5 Untreated M. tuberculosis infection in children will ultimately form the reservoir from which future infectious cases will arise. 2,6 Understanding and managing this reservoir is critical to achieving the ambitious Stop TB Partnership target of TB elimination by 2050. 7 Most M. tuberculosis transmission to young children is thought to result from household contact. 1 While household contacts of TB cases are at high risk for infection and disease, 8 in endemic areas most adult TB appears to occur from transmission outside the household. [9][10][11] This may also be true for paediatric TB, but there is little evidence to date, 12 and results are conflicting. Wide variations in estimates of the proportion of M. tuberculosis infection in children attributable to contact with infectious adults within the household have been reported, ranging from 75% in Cape Town 13 to ,1% in a peri-urban shanty town in Peru. 14 Differences in settings, such as community human immunodeficiency virus (HIV) and TB prevalence, population density and housing, and differences in age groups, with varying degrees of susceptibility to infection 15 and disparate social contact patterns, 16 contribute to differences in the estimates reported.
Within the context of a well-implemented TB control programme, we aimed to assess household and community risk factors associated with M. tuberculosis infection in pre-school children to help elucidate factors driving M. tuberculosis transmission that are not being addressed by current prevention strategies. and an incidence of new smear-positive TB of 87 per 100 000 adults per year. 17 Bacille Calmette-Guérin (BCG) vaccination is administered to children on first health system contact (usually birth) as part of the Expanded Programme on Immunisation.
The present study was conducted in the Karonga demographic surveillance site (DSS), with a population of approximately 36 000, which has been described in detail elsewhere. 18 The entire population is seen annually for re-census and related surveys, and 97% of all births and 99% of all deaths are reported by key informants. 18 Key informants are responsible community individuals who are trained to report vital events at a monthly meeting at which they are given a nominal sum (of about US$3) as compensation. 19 The demographic data include detailed information on family relationships, household socio-economic status, global positioning system (GPS) co-ordinates, dwelling structure, screening for chronic cough (72 weeks among those seen) and vaccination history of children aged 65 years. Adult HIV status (age 715 years) was collected in HIV prevalence surveys from 2007 to 2011 and other studies.

Participants
We conducted a population-wide tuberculin skin test (TST) survey in 2012 among all children aged 2-4 years residing within the DSS at the time of household recruitment. Children whose mothers were known to be HIV-positive at the time of delivery were defined as HIV-exposed, while children whose mothers were HIV-negative after or up to 1 year before their birth were defined as non-HIV-exposed.
A composite dwelling index was generated based on the building materials used and the presence or absence of glass windows. 20 A composite asset index was constructed by summing the ownership of household items, weighted by the monetary value of each item, to create an asset score. 21 The lowest 10-15% scores were coded as '1', the highest 15% as '4' and the middle groups were divided into '2' and '3' to create the respective socio-economic indices.

Diagnosed tuberculosis cases
The Karonga Prevention Study (KPS) collaborates with the district National Tuberculosis Programme to support core TB prevention and care activities. Bacteriological (including smear and culture status), demographic (including GPS co-ordinates of TB case household/s) and clinical (including HIV status) data were available from an ongoing prospective cohort of all patients starting treatment for TB in the district. 17 A total of 108 adult (aged 715 years) residents of the DSS were diagnosed with smear-positive pulmonary TB during 2007-2012.
The average annual smear-positive notification rate for each predefined residential area of approximately 450 households was calculated as the number of smear-positive TB cases reported per year per 100 000 population for the period 2007-2012. 22 Distance to the nearest smear-positive pulmonary TB case was calculated using ArcGIS 10 w software (Environmental Systems Research Institute, Redlands, CA, USA). A categorical variable was generated taking into account the variation in distance to the closest neighbour within the DSS. Preparatory work with parents identified that children aged ,5 years did not generally venture further than 'calling' distance from home, which was estimated to be about 200 m. The variable was therefore categorised to include household contact, neighbourhood contact (resident within 100 m; resident within 100-200 m) and children who lived .200 m from a known smear-positive TB case (baseline category).

Study procedures
Field staff were trained in TST placement and reading according to standard international guidelines, 23 using 2 international units of RT23 (Statens Serum Institute, Copenhagen, Denmark) and measuring the induration 48-72 h later. Periodic blinded comparison readings were done against the same reference reader to ensure consistency of TST reading.
Children with TST 710 mm were assessed for TBrelated symptoms by field staff, and the results were recorded in the child's health passport. Any child with symptoms suggestive of TB (fever, weight loss, failure to thrive, night sweats or cough) was reviewed by a clinician and referred to the district hospital. The HIV status of the children was not determined unless clinically indicated at the hospital. All children with TST 715 mm were commenced on 6-month isoniazid preventive therapy (IPT) (10 mg/kg once daily) after active disease had been excluded.

Case definition
A positive TST was defined as an induration of 715 mm, based on a previous mixture analysis of a tuberculin survey conducted in Karonga in 1980-1984 to estimate infection prevalence where exposure to environmental mycobacteria is common. 24 To check the suitability of this cut-off, the analysis was repeated in the new data set.

Statistical analysis
Mixture analysis of tuberculin data was based on implementation of the Expectation Maximization (EM) algorithm in R (R Foundation for Statistical Computing, Vienna, Austria) by fitting a twocomponent model to the observed profiles of the non-zero indurations, 24,25 Annual risk of infection was calculated using the formula: 2 Tuberculous infection in preschool children where R ¼ annual risk of infection, P ¼ proportion of children with 'positive' TST and a ¼ the mean age of the children.
A random effects logistic regression model taking into account clustering within the residential area was used to assess the relationship between risk factors and TST positivity. The likelihood ratio test was used to assess the overall significance of risk factors, tests for trend and departures from linearity, unless otherwise specified. To prevent over-parameterisation, 26 only those risk factors most strongly associated with the outcome were tested in the multivariable analysis (P , 0.1) and limited to a maximum of 4-5 parameters in the fully adjusted model. The population attributable fraction (PAF) was calculated using the formula: 27,28 where P 0 ¼ proportion of cases exposed and h ¼ the adjusted odds ratio (OR) of the association of M. tuberculosis infection in children with the risk factor. Analyses were performed using Stata v. 13.1 for Mac (Stata Corporation, College Station, TX, USA).

Ethics approval
The study was approved by the Malawi National Health Sciences Research Committee, Lilongwe, Malawi, and the London School of Hygiene & Tropical Medicine Ethics Committee, London, UK. Written informed consent was provided by a parent or guardian of each participating child. Tuberculin data An inter-rater reliability analysis on a sample of 215 TSTs of reading induration size to within 1 mm was found to be excellent (j ¼ 0.85, 95% conference interval [CI] 0.78-0.92), although this does not preclude digit bias; 91.6% of children aged 2-4 years, the majority (n ¼ 2340, 81%) of whom had been BCG-vaccinated within the first year of life, had no induration in response to TST. The frequency distribution of non-zero indurations, shown in Figure  2, shows a bimodal distribution with modes at 8 mm and 15 mm, and some evidence of digit preference (at 5 mm, 8 mm and 15 mm).

RESULTS
Mixture analysis of non-zero induration data in this study ( Figure 3A) generated a two-component  distribution, with a mean at 7.9 mm representing sensitisation to environmental mycobacteria and/or BCG vaccination, and a mean of 15.9 mm representing M. tuberculosis infection. Mixture analysis of non-zero induration data from Karonga District in 1980-1984, restricted to those in the same age group with a BCG scar, revealed a strikingly similar distribution, with means at 7.7 mm and 16.0 mm ( Figure 3B).
Of 2782 non-HIV-exposed children, 2555 (91.8%) had zero indurations, compared to 89.7% (78/87) of HIV-exposed children (v 2 , P ¼ 0.5). The median and interquartile range (IQR) of the non-zero indurations were similar in HIV-exposed (n ¼ 9, median 8 mm, IQR 8-15) and non-HIV-exposed children (n ¼ 227, median 8 mm, IQR 6-11). Risk factors associated with prevalent M. tuberculosis infection Distance from the nearest known TB case during the child's lifetime, being resident in an area with an average smear-positive notification rate of .30/ 100 000 population/year (community M. tuberculosis exposure), being born to an HIV-positive mother, cohabiting with 73 adult (age 715 years) household members over the child's lifetime, dwelling score and age at BCG vaccination were associated with a positive TST on univariable analysis (Table 1). Adjusting for age and sex did not change the effect estimate for any of the risk factors examined. There was evidence of a dose-response effect of decreasing distance from the nearest known infectious TB case (P trend across categories ¼ 0.02), with the highest ORs in those living within the same household, compared to children who lived .200 m from the nearest infectious case. In a multivariable model, distance from the nearest known infectious TB case, cumulative number of adult household members and maternal HIV status at birth were the only risk factors that remained strongly associated with the risk of a positive TST (P , 0.05; Table 2). The adjusted ORs for these three risk factors were similar to the crude ORs.
The odds of M. tuberculosis infection was two-fold

DISCUSSION
Despite the known lack of specificity of TST, tuberculin surveys remain a valuable epidemiological tool and provide important information on trends in tuberculous infection. 29 These surveys are usually undertaken in school-aged children (6-11 years), and there is thus a paucity of data on the risk of recent M.
tuberculosis infection as inferred by TST positivity in the very young. 30,31 In our study, the prevalence of M. tuberculosis infection in children aged 2-4 years was approximately 1.1%, using a TST cut-off of 15 mm, yielding an estimated ARTI of 0.3% (95%CI 0.2-0.5).
Using the tuberculin data from Karonga District in 1980-1984, and restricting this study to a comparable population (aged 2-4 years with BCG scar) and the same methods, the estimated prevalence of M. tuberculosis infection was 2.3% (95%CI 1.8-2.8) and the ARTI was 0.7% (95%CI 0.5-0.8). Replication of surveys, with identical antigens and procedures, in the same district and segments of the population (as attempted in this study) is probably the most accurate way to assess changes in the risk of M. tuberculosis infection. 32 Mixture analysis of tuberculin data from this study and from the earlier survey demonstrated strikingly similar two-component distributions, supporting the use of the 15 mm cut-off. The results suggest a reduction in the risk of infection in this age group over time, which is consistent with observed trends in incidence of smear-positive adult TB in the district. 17 Our ARTI estimate of 0.3% may not reflect the ARTI in older children and adults in this setting, due to different social mixing patterns, 33,34 although the determinants of the ARTI in this study population of very young children may in fact reflect the social mixing patterns and transmission dynamics of the mother or main carer. 35 The association with neighbourhood TB was expected. A study in a township in Cape Town, South Africa, found a similar doubling in the odds of TST positivity (using TST 710 mm) in children who lived on the same plot as an adult with smear-positive TB, compared to children with no plot-related exposure. 3 In a setting with a low-to-moderate TB burden such as Karonga, the relative importance of household transmission might be expected to be greater than in Cape Town, where the very high TB incidence might make community transmission more important. It should be noted that despite the association with distance to the nearest infectious  TB case, the PAF in our setting was very low, at 2.3% for household contact and 17.0% for neighbourhood (within 200 m), excluding household contact. This suggests that the maximum public health impact of treating all neighbourhood child contacts (aged ,5 years) of known adult infectious TB cases with IPT in this population would not be substantial. Children born to HIV-positive mothers had a 3.6fold increased odds of M. tuberculosis infection, after adjusting for distance to known TB cases and the number of adult household members. An increased risk of M. tuberculosis infection in HIV-exposed, non-infected children compared to non-HIV-exposed children in early childhood was also seen in a Ugandan study, which used both TST and an interferon-gamma release assay to diagnose M. tuberculosis infection status. 36 We did not know the HIV status of the children or the maternal M. tuberculosis infection status. It is unclear in our study whether the increased risk of M. tuberculosis infection seen in HIV-exposed children is due to increased susceptibility to M. tuberculosis infection following exposure and/or increased exposure, via either occult infectious TB within the household or unmeasured lifestyle factors associated with maternal HIV, such as increased attendance at health facilities.
Interestingly, no association between M. tuberculosis infection and population density was found. A plausible explanation may be that M. tuberculosis infections in young children in this community occur as a result of social mixing, which is unrelated to the population density of the area in which the child resides. There may be more undiagnosed infectious TB cases among females (the usual care givers of young children in rural Malawi), particularly in areas with lower population density, which are further from primary health care facilities than areas with the highest population density. A study is currently ongoing in the DSS to see if it is possible to identify the source of incident M. tuberculosis infection in children. Incident M. tuberculosis infection is identified by a repeat TST 1 year later in children with a previously negative TST. Household, neighbourhood and family contacts of children identified with incident M. tuberculosis infection are then screened for TB. It is hoped that this type of targeted case finding may identify a subset of undiagnosed cases that are responsible for onward M. tuberculosis transmission to the youngest members within the community.
The 1% of young M. tuberculosis-infected children in this population act as a sentinel for infectious TB in the community. In this setting, the majority of transmission events take place outside the household, either from unidentifiable casual contacts (with known or unknown TB) or from undiagnosed close contacts. This has major implications for strategies aiming to interrupt transmission. This study high-lights the need to identify people with undiagnosed infectious TB in the community, the need to identify congregate transmission settings and the need to develop strategies to improve community infection control, irrespective of the presence of people with known TB. IPT for household contacts of smearpositive TB cases, although justified given the high risk to individual children, is predicted to have little effect on the population reservoir of infection, even with effective implementation in settings such as this.

Tuberculous infection in preschool children
i