In areas with a high prevalence of tuberculosis, children contribute a significant proportion of the disease burden and experience considerable morbidity and mortality related to tuberculosis.^1,2 This is demonstrated by the fact that pulmonary tuberculosis rivals bacterial pneumonia as a respiratory cause of death in African children older than 6 months of age.³ The diagnosis of tuberculosis in children is difficult because bacteriological confirmation is rarely achieved and is often not even attempted.^4–,⁸ Chest radiography is regarded as a valuable diagnostic tool, but it is often impossible to identify hilar adenopathy, considered to be the most consistent sign of primary pulmonary tuberculosis, with certainty.⁹ In addition, the distinction between recent primary infection and active disease is highly problematic.¹⁰ Most diagnostic algorithms are partly symptom based, but these algorithms are poorly validated and lack standard symptom definitions.¹¹

To our knowledge, no previous study has documented the prevalence of symptoms associated with tuberculosis in randomly selected children living in a high burden community. In these communities a large number of individuals become infected with Mycobacterium tuberculosis (M tuberculosis) during childhood,¹² and symptom based diagnostic approaches are often promoted to screen children for active tuberculosis. The literature on tuberculosis related symptomatology in children is almost exclusively hospital based, which does not reflect the prevalence of symptoms in the general population.

A recent review of studies from the pre-chemotherapy era documented the natural history of tuberculosis in childhood.¹⁰ It confirmed that immune competent children, older than 2 years of age, are at low risk of disease progression following primary infection.¹⁰ The presence of symptoms associated with tuberculosis had definite diagnostic value in these low risk children, in whom the principal diagnostic challenge was to distinguish self contained infection from progressive disease.¹⁰

The aims of this community based study were (i) to document the prevalence of symptoms associated with tuberculosis among randomly selected children living in a high burden community, and (ii) to compare the prevalence of these symptoms in children without tuberculosis to those in children with newly diagnosed tuberculosis.

METHODS

A cross sectional community based survey was carried out between July and December 2002. A questionnaire was completed and a tuberculin skin test (TST) administered to each child (<15 years of age). Chest radiographs (CXR) were performed according to the South African National Tuberculosis Control Program (NTCP) guidelines.

RESULTS

In total, 1593 children were enumerated at the selected addresses. Questionnaire and TST results, available in 1415 (88.9%) children, were included in the analysis. Table 2⇓ summarises the demographic characteristics of the whole group. A positive TST was recorded in 451 (31.9%) children. Seventy five children reported previous HIV testing, with only one reporting a positive test result. CXR were performed in 239 (16.9%) children, applying current NTCP guidelines. The main reasons for performing the CXR were as follows: (i) 177 (74.1%) <5 years of age with household contact ever, (ii) 33 (13.8%) <5 years of age and with TST ⩾15 mm, and (iii) 29 (12.1%) with suspect symptoms. Eighteen (1.3%) children were newly diagnosed with probable or confirmed tuberculosis. Table 3⇓ summarises the disease characteristics (type of disease) in these 18 children. Bacteriological confirmation (a positive M tuberculosis culture) was achieved in two cases (11.1%). None of the children with newly diagnosed tuberculosis had clinical signs suggestive of HIV disease.

Table 4⇓ describes the demographics of the 1397 children without tuberculosis, compared to the 18 children with newly diagnosed tuberculosis. Most children with newly diagnosed tuberculosis (14; 77.8%) were less than 5 years of age. Household contact with a recent sputum smear positive source case was reported in three (16.7%) children with newly diagnosed tuberculosis, compared to 50 (3.6%) children without tuberculosis (OR = 5.4; 95% CI 1.5 to 19.2). The TST result was equal to or greater than 15 mm (median 17 mm; range 0–31 mm) in all but one child with newly diagnosed tuberculosis.

Table 5⇓ compares the symptoms reported in children without tuberculosis to those in children with newly diagnosed tuberculosis. Of the 1397 children without tuberculosis, 253 (26.4%) reported a cough during the preceding 3 months. Prolonged symptoms were not uncommon, and 66 (6.9%) reported coughing of more than 3 weeks duration. No significant differences in symptom prevalence or symptom duration existed between age groups (<2 years, 2–4 years, 5–9 years, 10–14 years) or between those with and without TST proof of tuberculosis infection. Significantly more children with newly diagnosed tuberculosis reported weight loss in the preceding 3 months (OR = 4.5, 95% CI 1.5 to 12.3). None of the other symptoms showed statistically significant differences between the two groups, even when the reported duration of symptoms was taken into account. When children less than 5 years of age were analysed separately, to correct for possible selection bias in the older children, the results were similar.

The presence of more than one symptom increased the odds for tuberculosis, but this reached significance only with the inclusion of weight loss. The combination of cough and weight loss was most significant (OR = 5.4, 95% CI 1.7 to 16.9). Nine (50%) children with newly diagnosed tuberculosis reported no symptoms at all, compared to 910 (65.2%) of those without tuberculosis (OR = 1.9, 95% CI 0.7 to 4.7).

DISCUSSION

This community based survey documents the prevalence of symptoms associated with pulmonary tuberculosis in the average child living in a high burden community. The results demonstrate that within this setting, these symptoms are common in the general paediatric population. Only weight loss, alone or in combination with other symptoms, was significantly more common in children with newly diagnosed tuberculosis. Combinations of symptoms (symptom complexes) were more likely to be associated with tuberculosis than individual symptoms, but the small number of tuberculosis cases limited the analysis. This is an inevitable limitation of community based surveys that document relatively rare events.¹⁸ The detection of 18 new tuberculosis cases during this survey demonstrate a high prevalence of undiagnosed childhood tuberculosis within this community (1272/100 000 of the child population).

The observation that nearly a third of children coughed in the preceding 3 months may be explained by the fact that both viral and bacterial infections are common in this community, especially during the winter months (May to September). It is also common practice for children to share a bed with their parents, which may increase parental awareness of night time symptoms such as cough, fever, or sweating. Similar to previous observations, some (exactly 50%) of the children with newly diagnosed tuberculosis reported no symptoms.^19–,²¹ Although significant differences were found between the symptoms reported in children with tuberculosis and those without, these differences are of limited diagnostic value and only relevant from an epidemiological perspective. This is illustrated by the fact that the symptom complex with the highest significance, cough and weight loss, had a positive predictive value of only 5%.

The study had several limitations; of most importance is the fact that CXR were performed on a selective subgroup of children. Only children under 5 years of age were radiologically screened following household exposure or a positive TST and as a result older children with asymptomatic primary infection may not have been diagnosed. Although this could have biased the results, the statistical findings in children less than 5 years of age and in the group as a whole were similar when comparing symptoms in those with and without tuberculosis. This potential bias seems unavoidable as both the WHO and NTCP guidelines recommend that a CXR should only be performed in older children if they are symptomatic. It may even be argued that it is unethical to perform a CXR in an asymptomatic child older than 5 years of age, as the risk of developing disease following primary infection is less than 1%.¹⁰ The group diagnosed with tuberculosis during this survey represents a true reflection of children who would have been diagnosed with tuberculosis in a “real life” setting, where resources permit active contact tracing and adequate investigation.

The majority of children diagnosed with tuberculosis had signs indicative of recent primary infection only (positive TST and elements of the Ghon complex on CXR). The natural history of tuberculosis in children illustrates that following primary pulmonary infection, transient visibility of the Ghon complex is common and does not necessarily indicate disease.^10,22,23 The fact that current, internationally accepted definitions of disease often reflect recent primary infection, especially in children who are diagnosed after active contact tracing, does hamper the interpretation. Therefore, it is important to point out that, although the study indicates that symptoms have limited diagnostic value, their value for differentiating active disease from recent primary infection requires further investigation.

Questionnaire driven surveys are inherently limited by recall bias and subjectivity. Bias was reduced by the use of standardised questionnaires and the fact that the diagnostic tests (TST and CXR) were performed after completion of the questionnaire. Reported weight loss reflected a subjective impression and objective weight measurements were not taken into account, as these were not routinely done in all children. In addition, the questions used may have been too imprecise, as the pre-chemotherapy literature identifies important symptom characteristics, associated with progressive tuberculosis, that were not elucidated in the present study. These characteristics include persistent, non-remitting symptoms of recent onset.¹⁰ Careful symptom characterisation may be essential to improve their diagnostic value.

Nearly a third (5, 27.8%) of children with newly diagnosed tuberculosis reported previous household contact with a sputum smear positive source case. This is in agreement with a study from the same community that employed restriction fragment length polymorphism (RFLP) analysis, providing proof that 34% of children acquired disease following infection from a source case within the household.²⁴ In high burden settings, most infection, particularly in children older than 2 years of age, is contracted outside the household within the community.^9,13,24

In conclusion, this study illustrates the importance of community based research. It describes symptomatology within the general community and highlights the danger of extrapolating findings from hospital based studies to the community level. It is essential to know the prevalence of symptoms within the general community in order to evaluate the value of any particular symptom based diagnostic approach. Comparing the symptoms reported in children without tuberculosis to those in children with newly diagnosed tuberculosis demonstrates why current symptom based scoring systems perform poorly in clinical practice. Most importantly, the study emphasises the need for improved symptom characterisation and accurate outcome definitions which will adequately differentiate M tuberculosis infection from active disease.