Incidence of Tuberculosis Among Young Children in Rural Mozambique

Background: Tuberculosis (TB) contributes significantly to child morbidity and mortality. This study aimed to estimate the minimum community-based incidence rate of TB among children <3 years of age in Southern Mozambique. Methods: Between October 2011 and October 2012, in the Manhiça District Health and Demographic Surveillance System, we enrolled prospectively all presumptive TB cases younger than 3 years of age through passive and active case finding. Participants included all children who were either symptomatic or were close contacts of a notified adult smear-positive pulmonary TB. Children were clinically evaluated at baseline and follow-up visits. Investigation for TB disease included chest radiography, HIV and tuberculin skin testing as well as gastric aspirate and induced sputum sampling, which were processed for smear, culture and mycobacterial molecular identification. Results: During the study period, 13,764 children <3 years contributed to a total of 9575 person-year. Out of the 789 presumptive TB cases enrolled, 13 had TB culture confirmation and 32 were probable TB cases. The minimum community-based incidence rate of TB (confirmed plus probable cases) was 470 of 100,000 person-year (95% confidence interval: 343–629 of 100,000). HIV co-infection was present in 44% of the TB cases. Conclusion: These data highlight the huge burden of pediatric TB. This study provides one of the first prospective population-based incidence data of childhood tuberculosis and adds valuable information to the global effort of producing better estimates, a critical step to inform public health policy.

T uberculosis (TB) is an underrecognized but potentially important cause of morbidity and mortality in children in TB endemic settings. 1,2 Infants and young children (<3 years) and those with immunodeficiency caused by HIV or severe malnutrition are at highest risk of developing TB disease following infection. 3 Delay of diagnosis and treatment in these children increases the risk of rapid disease progression and mortality. 4 TB diagnosis is particularly challenging in this population, given the lack of specific symptoms, the difficulty in obtaining samples for microbiological examination and the often paucibacillary disease. The diagnostic yield of samples is often <20% under TB program conditions. 5,6 These diagnostic difficulties result in delayed and underdiagnosis of the disease, contributing to the hidden burden of TB in children.
Child TB is receiving more attention 3 as the World Health Organization (WHO) post-2015 TB strategy seeks to engage the wider health sector including the child health-care sector. 7 The WHO Global Tuberculosis Report 2014 estimates that 550,000 children developed tuberculosis during 2013, representing 6% of the global TB burden. 8 However, several factors suggest that the true burden of disease may be higher as these estimates assume an equal ratio of notified cases in children and adults (whereas underreporting in children is very common 9 ), and estimated deaths only include those in HIV-negative children. 3 As a setting's total TB burden increases, there tends to be a rise in the proportion of TB cases attributable to children. 10 Thus, in high TB burden settings, children may represent up to 10-20% of TB cases, with increased TB incidence in <5 and >15 years. 4,8,11 Mozambique is one of the high TB burden countries listed by the WHO but has a very low reported case-detection rate of 37%. 8 Improved reliable estimates are required to quantify the hidden burden of disease and measure future progress toward the control of TB in the country, especially for vulnerable populations such as children. 8,12,13 We, therefore, aimed to determine the minimum community-based incidence rate (IR) of childhood TB.

Setting
The study was conducted in the Manhiça District (rural southern Mozambique), where the Manhiça Health Research Center (Centro de Investigação em Saúde de Manhiça) runs a Health and Demographic Surveillance System (HDSS) including the Manhiça District Hospital (MDH) and other peripheral health posts in the area. The HDSS links demographic and clinical data and covers a population of around 92,000 inhabitants, of which approximately 11% are <3 years. 14 A full description of the site can be found elsewhere. 14 In 2011, the <5 years mortality rate was 70 of 1000 live births. Severe malnutrition is common with an estimated IR of 35 of 1000 person-year among children from 1 to 2 years. 15 TB treatment is offered free of charge at the health units, and children are routinely vaccinated at birth with Bacille Calmette-Guérin (BCG), with estimated coverage ranging from 86% to 90%. 16,17 The 2013 WHO TB incidence estimates for the country is 552 of 100,000 population. 8 The HIV prevalence in the district is among the highest in the world, reaching 39.9% in the community among individuals aged 18-47 years and 29.4% for women attending the antenatal clinic. 18 A prospective study was designed to recruit participants through passive and active case finding in the community, MDH and peripheral health centers during a 1-year period (2011)(2012). Participants included all children from the HDSS who were <3 years at the time of enrolment and had either TB symptoms or were close contacts of a notified adult smear-positive pulmonary TB (PTB) case. Relapse or recurrent cases were excluded.

Clinical Procedures
Presumptive TB cases were identified through 2 strategies: (a) passive case detection of children presenting to the health facility with ≥1 symptoms compatible with TB (see Table 1 for a complete list of symptoms). Those not recruited at the time of the visit to the clinic were later identified through the clinical data collected at the health unit by the HDSS. (b) Active case finding consisted of linking the adult smear-positive PTB cases registered at the district National TB Program (NTP) in the previous 24 months to the HDSS database to identify all household contacts <3 years. At enrolment, demographic and clinical information was collected through interviewing of parents and physical examination. Participants had a chest radiography (CXR) performed, followed by HIV antibody testing and tuberculin skin testing (TST). CXR were performed with a digital radiogram machine and included posteroanterior and lateral projections. For clinical purposes, an initial reading was performed on site by the clinician. Subsequently, all CXR were reviewed and reassessed by an experienced pediatric radiologist (J.R.) who was blinded to the clinical information. TST was performed with intradermal injection of 2 units (Serum Staten´s Institute, Denmark) and reading at 48-96 hours, according to the study protocol. For symptomatic cases, in the same day, 2 ambulatory samples were obtained in a negative pressure facility available at the MDH: 1 gastric aspirate (GA) and 1 induced sputum (IS) with nasopharyngeal suction, following WHO recommendations. 19 Asymptomatic cases with abnormal CXR did not undergo sampling but were reevaluated at further visits. For suspected extrapulmonary TB (EPTB), appropriate samples were obtained.
All case managements were performed by the NTP according to established national clinical guidelines. Those patients with clinical or microbiological diagnosis of TB were started on TB treatment at the NTP with the standard 3 or 4 first-line regimens according to WHO category. Other symptomatic patients were referred for specific treatment and follow-up including antibiotics or nutritional supplementation if indicated. Presumptive cases had a follow-up visit within the next 6 months regardless of initial disease classification to assess resolution of symptoms without anti-TB treatment and/or clinical response to alternative therapy (if any). If persistently symptomatic, further evaluation and testing including CXR and samples were performed to rule out TB. Contacts had a follow-up visit, which included physical examination and CXR, as well as GA and IS samples for those symptomatic or with an abnormal CXR.

Laboratory Procedures
Samples were transported within 4 hours of collection and processed in the Biosafety Level III TB laboratory at Centro de Investigação em Saúde de Manhiça. Following NaCl/NaOH digestion and concentration through centrifugation, all samples were processed for acid-fast bacilli smear testing using LED Microscopy and Ziehl-Neelsen staining and inoculated into liquid culture media (BACTEC MGIT 960-automated; Becton Dickinson Microbiology Systems, Sparks, MD) and solid media (Lowenstein-Jensen). Positive cultures were confirmed using Ziehl-Neelsen staining and rapid test as well as Xpert MTB/RIF and identified through mycobacterial molecular identification (GenoType Mycobacterium CM/AS; Hain Lifescience). First-line drug sensitivity testing was performed either on liquid culture or line probe assays. The laboratory is subject to an external quality assurance program.

Study Definitions
• Exposure to TB was defined as either documented (identified through active case finding) or reported contact (household or regular contact during child lifetime).   (1) compatible symptoms unresolved at last clinical follow-up visit (before any TB treatment initiation) plus (2) compatible CXR (for children with ≥1 CXR, the latter was used given the likelihood of seeing resolving pneumonias) plus (3) at least one of the following: TB exposure, positive TST or positive response to TB treatment. EPTB cases followed the same definition except for the requirement of having an abnormal CXR. The study TB case definition was adapted a standardized clinical case definition of intrathoracic TB disease and included confirmed plus probable cases 21 (see Fig. 1 for complete case definition).

Ethical Approval
The study protocol was approved by the Mozambican National Bioethics Committee and the Hospital Clinic of Barcelona Ethics Review Committee.

Data Analysis and Statistical Considerations
Clinical data were double entered in an electronic data capture system (OpenClinica, www.openclinica.org) and checked for discrepancies. Statistical software for analysis was Stata 13.0 (StataCorp. 2013. Stata: Release 13, StataCorp LP, Statistical Software, College Station, TX). We calculated Z scores for weight-for-age, height-for-age and weight-for-height using WHO 2006 reference data. 22 The minimum community-based IR was calculated as a density rate with the age-specific yearly number of TB cases (according to the study case definition) among study participants divided by the total age-specific population at risk during a period of 12 months (person-time at risk). Time at risk was individually measured using demographic surveillance system (DSS) data taking into account demographic events (births, deaths and migrations) of all children included in the study. The IR is considered to be minimum as the case detection system cannot ensure that all TB cases are detected. For each IR, 95% exact Poisson confidence interval (CI) was calculated. Proportions were compared using the Pearson or Fisher exact χ 2 test, and odds ratio and 95% CIs were estimated using logistic regression. Variables at a significance level below 0.2 were chosen and placed on stepwise backward multivariate logistic regression. Only factors with a P value on likelihood ratio tested were retained on the model.

RESULTS
During the study period, 13,764 children <3 years contributed to a total of 9575 person-year in the Manhiça DSS (Fig. 2). A total of 789 presumptive TB cases were enrolled (42 and 747 identified through active and passive case finding, respectively). Forty-five children fulfilled the TB case definition-13 microbiologically confirmed plus 32 probable TB (Fig. 1). Thus, the minimum community-based IR was 470 of 100,000 person-year (95% CI: 343-629 of 100,000) for confirmed plus probable cases and 135 of 100,000 person year for confirmed cases (95% CI: 72-232 of 100,000; Table 2).

FIGURE 2. Study profile.
Flowchart showing the number of children younger than 3 years of age in the study area and those enrolled in the study. A total of 1483 children were identified with at least 1 compatible TB symptom, and 747 presumptive TB cases were enrolled in the study for further workup. Among the 329 adult smear-positive cases registered at the NTP between October 2010 and October 2012, we identified 123 belonging to the study area and 180 contacts <3 years, of whom 102 accepted to participate in the study yielding an additional 42 presumptive TB cases. Among the remaining 60 contacts, 7 were lost to follow-up and the rest had at least 1 follow-up visit. Eighty-eight percent presumptive TB cases enrolled had at least 1 follow-up visit and 632 of 697 completed follow-up (had follow-up visits until alternative diagnosis was made or became asymptomatic).
Baseline characteristics of presumptive TB cases are presented in Table 1. Fifty-four percent were males, and the age distribution showed a predominance of children between the ages 12 and 24 months (51%). The most frequent clinical feature at enrolment was severe malnutrition, which was the only symptom in 72% of cases. Nutritional assessment found that almost a quarter had severe undernutrition (weight for age Z score <3). Of the 1347 total CXR performed during the study, 27% had only one projection. Twentyone percent of all presumptive TB cases had a CXR compatible with TB. Thirty percent of presumptive cases had a second TST of which 9% had a positive TST. Among all presumptive TB cases, 9 had a positive smear, although none of the 9 had a positive Mycobacterium tuberculosis culture (4 were non-TB mycobacteria and 5 were culture negative). Non-TB mycobacteria were isolated in 27% of all cultures of presumptive cases. We found 7 EPTB cases-4 lymph node and 3 disseminated-and no TB meningeal cases. A total of 104 children were diagnosed as HIV positive (13%).
We identified 13 confirmed TB cases (7 in GA, 4 in IS and 2 both in GA and IS). The percentage of confirmed cases among TB cases was highest for those <1 year (40% vs. 29% and 22% among children with 1-2 and 2-3 years, respectively), and statistically significantly lower for HIV-TB coinfected cases (10% vs. 44%, P = 0.02). Confirmed cases presented a higher frequency of cough or fever when compared with probable cases. Furthermore, the confirmed cases appeared to be more symptomatic at enrolment than did the probable cases (53.8% vs. 15.6% presenting with ≥1 TB symptom, respectively, P < 0.001). Probable cases had a higher proportion of HIV infection (P = 0.01), positive TST (P = 0.001) or BCG scarring (P = 0.08) when compared with confirmed cases.
Multivariate logistic regression analysis for TB risk factors showed that HIV infection and number of previous outpatient consultations were predictors of TB disease when compared with unlikely TB cases. After adjusting for other variables, HIV-infected children were 6 times more likely to have TB disease than uninfected ones (odds ratio: 8.4; 95% CI: 4-17; see Table, Supplemental Digital Content 1, http://links.lww.com/INF/C101).
Fifty-two patients were started on TB treatment based on clinical or microbiological criteria, and 67% fulfilled the study TB case definition (Table 3). A total of 97 children initiated isoniazide preventive treatment (IPT; 71 based on exposure history, 21 on TST results and 5 unspecified) and 5 were later diagnosed as TB cases while on IPT. Because of drug supply shortages, isoniazide was not always available, and 47% of children with criteria did not initiate IPT. The mortality rate for all presumptive cases at 12 months after enrolment was 5.2% and increased with decreasing age (10.9%, 5.7% and 0.8% of children in the first, second and third year of life, respectively, P < 0.001). Mortality was also higher in TB cases when compared with non-TB (13% vs. 5%, respectively, P = 0.02) as it was in HIVinfected children when compared with HIV-uninfected (14.4% vs. 3.8%, P < 0.001). The case fatality rate was 9% (n = 4 of 45 TB cases), all deaths taking place in the first 6 months after enrollment.

DISCUSSION
This study provides one of the first prospective populationbased incidence estimates of childhood tuberculosis in a high TB-HIV endemic setting and shows a consistently high IR across all ages. These results underscore the hypothesis of a gross underdetection and underreporting of childhood TB in Mozambique and globally. 23 Mozambique has almost half its population below the age of <15 years, and yet, pediatric TB only accounted for 7% of all new cases notified in 2012, much lower than the expected 10-20% of the total burden of TB disease seen in high burden countries. In the Manhiça District, the notified IR for children <1 year and between 1 and 4 years in 2011 was 163 and 399 of 100,000, respectively (García-Basteiro et al, personal communication, 2014 24 ). This corresponds to half the TB IR reported in this study and may suggest underdetection and underdiagnosing. Furthermore, underdetection is common in the wider Mozambican context, where WHO  estimates that only 37% of actual TB cases are detected. 8 Although the latest WHO country incidence estimates are 552 per 100,000 population, data from Manhiça suggest that the burden of disease in Southern Mozambique might be much higher. In fact, while the national 2011 notified rate was 186 of 100,000 population, sameyear data from the Manhiça suggest the TB IR of smear-positive cases could be as high as 456 per 100,000 population among adults aged 18-47 years. 25 Globally, several pediatric TB incidence estimates have been published recently, with results varying from less than 200,000 new cases in 2013 26 to 970,000 in 2010. 6 The large variation in the estimates highlights the challenges in estimating the burden of pediatric TB and the need for population-based data to inform predictive models.
There are few studies reporting age-specific pediatric TB incidence in high burden countries. Most studies are based on hospital-based retrospective reviews of notification rates, and to our knowledge, none have reported community IR using DSS. 11,[27][28][29][30][31][32][33][34][35] However, in areas where health seeking behavior strongly modifies the pattern of attendance, community-based studies that use active case detection rather than notified TB rates are necessary to provide accurate estimates. Moreover, childhood mortality and frequent migration are potential causes of disease underestimation if DSS person-years are not available. Inconsistency in TB clinical definitions among studies is a challenge for comparability and has been a limitation in obtaining data for meta-analyses. The recently proposed definition, applied in this study, may pave the road for future comparisons. 9 The IR we report is significantly higher than in other high burden African countries, such as Malawi (notified IR <1 year of 78 of 100,000) or Tanzania (theoretical IR <5 years based on likelihood of disease progression of 134.5-308.5 of 100,000) and similar to data from Gabon (extrapolated IR <15 years 366 of 100,000) or neighboring South Africa (notified IR <5 years 770 of 100,000). [30][31][32] There are several limitations to this study, mostly leading to a possible underestimation of the true TB incidence. First, only single-day samples were obtained as most patients would not accept overnight admission, decreasing the chances of microbiological confirmation. Second, for study purposes CXR were read by a single blinded experienced pediatric radiologist rather than the 2 independent CXR readers are often recommended to prevent bias, given the pivotal role of CXR in case definition, and the poor interobserver and intraobserver agreement among reviewers. 36 Third, contact tracing could not be fully implemented mainly because of difficulties in patient identification and poor recording. Fourth, the percentage of EPTB cases was lower than the 20-30% expected and reported by others. 11,30,31,[37][38][39] Although BCG protection may have a role, it is likely that some EPTB cases in this study were missed; the reason may be because of a stronger focus on PTB in the study design, errors in classification (disease localization, including disseminated TB, may be confounding in young children in the absence of CT-scan) or lost cases because of the fact that severely ill children are often transferred to the tertiary reference hospital in the capital for specific diagnostic procedures. Finally, there is a risk of overestimating TB IR by either including TB prevalent cases at enrollment or adding new incident cases during follow-up beyond the 1-year enrollment period. Although this possibility cannot be ruled out, we believe the effect would be minimal and probably overweighed by the above-mentioned risk of underestimation.
In this study, HIV prevalence was high regardless of disease classification, reaching 56% of probable cases. The fact that significantly fewer HIV-infected children had TB confirmation reflects the diagnostic difficulties in this group. Given the overlap between symptoms from both TB and HIV, HIV-infected cases of TB pose the greatest ascertainment bias with the highest risk of over or under estimation. Even though IPT is indicated in all HIV-positive children, the implementation of IPT among African NTP remains very poor, 40 and in this study, the high proportion of missed opportunities for chemoprophylaxis in HIV and/or TB exposed should raise alert. There may also be missed preventable child deaths in HIV-infected presumptive TB cases and possible TB cases. It has been reported that many children who die of diseases such as malnutrition or respiratory infections may have, in fact, undiagnosed TB. 41 There is thus a need for widespread recognition that TB control is crucial for childhood survival. 1 This study highlights the huge burden of pediatric TB under detection in children younger than 3 years of age. These data add valuable information to the global effort of producing better estimates of childhood TB burden, a critical step to inform public health policy.