Prospective Observational Study of Incidence and Preventable Burden of Childhood Tuberculosis, Kenya

Prospective data on childhood tuberculosis (TB) incidence and case detection rates (CDRs) are scant, and the pre ventable burden of childhood TB has not been measured in prospective studies. We investigated 2,042 children (<15 years of age) with suspected TB by using enhanced sur veillance and linked hospital, demographic, notification, and verbal autopsy data to estimate the incidence, CDR, risk factors, and preventable burden of TB among children in Kenya. Estimated TB incidence was 53 cases/100,000 children/year locally and 95 cases/100,000 children/year nationally. The estimated CDR was 0.20–0.35. Among chil dren <5 years of age, 49% of cases were attributable to a known household contact with TB. This study provides much needed empiric data on TB CDRs in children to inform national and global incidence estimates. Moreover, our find ings indicate that nearly half of TB cases in young children might be prevented by implementing existing guidelines for TB contact tracing and chemoprophylaxis. of


Investigation of Children Enrolled through Active Contact Tracing
New cases of smear positive pulmonary TB resident within the Kilifi Health and Demographic Surveillance System (KHDSS) were identified in the KCH TB outpatient clinic, and child household contacts of these index cases identified on the KHDSS population register.
For pragmatic reasons, and in keeping with Kenyan national guidelines, contact tracing focused on children under 5 years of age resident in the same household as a case of smear positive pulmonary TB, as smear positive cases are the most infectious and young children are most vulnerable to developing active TB following infection (5). Each index case was then invited to bring all children under 5 years in the household (symptomatic or asymptomatic) to the pediatric TB outpatient clinic for further assessment, and given sufficient money to cover the return fare to hospital.
All children identified through active contact tracing underwent a structured history and examination, anthropometry, CXR, and a TST. Those with symptoms or signs of possible TB (Technical Appendix Box 2), an abnormal CXR, or a positive TST were further investigated for suspected TB as described below.

Specimen Collection for Mycobacteriology
Appropriate clinical specimens were collected for AFB microscopy and mycobacterial culture from all children with suspected TB. Children who were able to expectorate provided three spontaneous sputum samples. Sputum induction was performed on the remainder. If sputum induction was contraindicated (e.g., due to severe respiratory distress), gastric aspiration was performed. Sputum induction and gastric aspiration were performed according to international recommendations (6). Further investigations including fine needle aspiration (FNA) of lymph nodes, mycobacterial culture of CSF, urine, pleural/ascitic/joint fluid, or biopsy material, or repeat sampling, were performed at the discretion of the clinical team caring for the patient according to clinical indications in individual cases. Specimens were transported to the laboratory at 2-8°C and processed the same day.

Application of Published Clinical Diagnostic Tools to Estimate Childhood TB Incidence
To compare crude incidence estimates generated using the study case definitions with incidence estimates derived using other published clinical definitions we included clinical diagnostic tools published in the peer reviewed medical literature, and guidelines from the WHO and Kenya National TB Programme. We excluded published tools that failed to present diagnostic criteria in sufficient detail to apply them to the dataset. For those tools that included a category of confirmed TB based on microbiological diagnosis we confined our analysis to categories defined by clinical criteria alone, to explore their performance under normal programmatic conditions with limited availability of mycobacterial culture. We also retrospectively applied new consensus definitions for childhood TB research that were published after completion of our study (4), and derived incidence estimates for the consensus definitions of both Definite and Probable TB to facilitate comparison with future studies.
We created variables for each diagnostic score and/or diagnostic categories with close reference to the published definitions of each variable. In instances where the exact definition of a clinical variable was not clearly specified in the original publication we chose what we judged to be the most likely intended definition for application in the relevant setting and reported the definition we used. Thus 'unexplained fever' was defined as a fever for >14 days in the absence of malaria parasitaemia or evidence of focal infection; a cutoff of at least 1 week's duration was used for a history of night sweats; and 'bulky lymphadenopathy' was defined pragmatically as the presence of lymph nodes sufficiently large to perform a fine needle aspirate (usually ≥2cm diameter). 'Malnutrition not responding to treatment' was defined as death, or failure to regain 10% bodyweight (in the case of marasmus) or failure of edema to resolve (kwashiorkor), in a child admitted with severe malnutrition.
A 'suggestive symptom complex of TB' was included in the Ghidey-Habte diagnostic tool (7) but only vaguely defined as "non-specific symptoms such as fever, night sweats and loss of weight, and specific symptoms related to the site invaded, e.g. cough, swelling of lymph nodes, abdominal distension, difficulty in walking, etc ". For the purposes of our analysis we included in this definition fever, cough, night sweats, and weight loss (each for at least 2 weeks), bulky lymphadenopathy, signs of pleural effusion or ascites, gibbus, and a change in temperament or reduced level of consciousness. We then compared incidence estimates using a requirement for either ≥2 or ≥3 of these clinical features to define a 'suggestive symptom complex of TB'.
In keeping with published definitions (4,8,9), a 'suggestive CXR' for TB was defined as the presence of a Ghon focus or complex, miliary infiltrate, cavities, or a pleural or pericardial effusion -unless an alternative definition was clearly presented for a particular clinical tool in which case the definition presented was used.
Using each of these published clinical definitions we calculated TB incidence using as the numerator the number of KHDSS-resident children fulfilling each definition during the study period.

Using TB Notification Data
We linked National Tuberculosis Programme (NTP) notification data with KHDSS census data to estimate the CDR in both the passive and the active case detection arms of the study.

Passive Case Detection
We used notification data from the Kilifi District TB Register to estimate the proportions of KHDSS resident child TB cases captured at KCH through passive case detection. Data from the register were double entered into a bespoke electronic database using Filemaker Pro version 10 (Filemaker Inc, CA, USA). The KHDSS residence status (resident or non-resident) of each patient in the register was then coded manually by a senior demographer with several years of local experience and detailed knowledge of the KHDSS area (CN), using the address documented in the register. All KHDSS resident childhood TB cases notified between August 2009 and July 2011 were identified from this database. We then manually cross-referenced the name, age and treatment date of each of these cases against the KIDS TB Study database to identify children that had also been captured by passive case detection at KCH. To limit disease misclassification among young children we limited the analysis to smear positive cases, and calculated the case detection rate as: We derived 95% confidence intervals based on the variance of the product of these two proportions using standard methods.

Using Hospital-Based Mortality Surveillance
We used the unique personal identification number (PID) of each KHDSS resident child to link vital status data from KHDSS census rounds with KCH pediatric admission outcome data.
We then calculated the case detection rate as the proportion of all childhood deaths in the KHDSS area that were captured at KCH during the study period:

=
No. KHDSS-resident children who died at KCH Total no. childhood deaths among KHDSS residents

Using Verbal Autopsy
A better approach to estimating the sensitivity of hospital-based surveillance is to use disease specific mortality data to calculate the proportion of childhood TB deaths captured by the study. Poor quality vital registration data in Kilifi District make these data unsuitable for this analysis. We therefore made use of data from an ongoing verbal autopsy (VA) study within the KHDSS.
Details of the Kilifi verbal autopsy study, including validation of the methodology using hospital records of the cause of death, have been published elsewhere (10). Deaths among In the case of a discrepancy between the two clinicians, a third clinician reviews the case blind to adjudicate, and if there is no agreement between the three reviewers they meet to discuss the case to form a consensus.
Using each child's unique KHDSS personal identification (PID) number we merged VA and KIDS TB records to calculate the proportion of TB deaths among KHDSS resident children with that were captured by the KIDS TB study. We defined TB deaths as those whose cause was coded as TB or which occurred in a patient with documented tuberculosis according to the respondent and/or any available supporting documentation, including death certificates, burial permits and post mortem reports.
We then estimated the case detection rate as =

No. TB deaths in VA study that occurred in children captured by KIDS TB Study No. TB deaths in VA study
Although the true mortality burden of TB among children in Kilifi District was not known, we predicted that the number of child TB cases diagnosed by VA was likely to be small (since TB is responsible for a minority of childhood deaths and is even more difficult to diagnose retrospectively by VA than in clinical practice); and that the precision of our case detection rate estimate was therefore likely to be poor.
To mitigate this, we also used the VA study to identify the much larger group of children whose reported clinical features before death met the KIDS TB Study criteria for suspected TB.
Healthcare-seeking behavior in Kilifi is usually determined by the clinical features of an illness, rather than the diagnosis per se (11,12

Risk Factors for Childhood TB
We summarized the distribution among cases and controls of each putative risk factor, and derived crude odds ratios (OR) and 95% confidence intervals (CI) in each case. Likelihood ratio tests for a general association were performed and p values reported.
To explore associations with TB contact variables we used children with no TB contact as the baseline group for comparison. Since some children had a history of more than one TB contact, we assumed an individual child had an equal probability of acquiring TB from each contact; created a separate record for each child-contact pair; and weighted each of these pairs in the analysis by the reciprocal of the number of TB contacts reported for each child.
We then derived multivariable logistic regression models to identify independent risk factors for TB. Categorical variables with at least a weak association with TB in the univariable analysis (likelihood ratio test p value ≤0.1) were included in the model. We performed backward stepwise logistic regression using standard selection criteria, such that variables that were not significantly associated with TB (Wald p value <0.5) were sequentially dropped from the model.
Likelihood ratio tests were used to test for potential interactions in the final model. Based on this model adjusted odds ratios and 95% confidence intervals were derived for the associations with TB of each variable included; p values for each association were derived using the Wald test.
We estimated the population attributable fraction (PAF) of childhood TB due to close contact with a known case of adult TB. We confined this analysis to children under 5 years for two reasons. First, it is well documented from natural history studies in the pre-chemotherapy era that >90% active TB disease in this age group occurs within 2 years of infection (13). The number of contacts identified during the 2 year recruitment period therefore provides a good estimate of the likely number of contacts putting children at risk, since the overall rate of TB notifications in the study population is constant. Second, this is the group targeted for isoniazid chemoprophylaxis since they are the most vulnerable (1). • All clinical features explained by a definitive alternative diagnosis, making TB highly unlikely; and/or insufficient clinical indication for a trial of TB treatment and no clinical deterioration during follow up in the absence of TB therapy Not classifiable • Children who did not meet criteria for confirmed, highly probable or possible TB, and in whom TB could not confidently be excluded, for example because they died or were lost to follow up Technical Appendix Table 3. Estimated annual national case burden and incidence of childhood TB by age group (2010