Background: Tuberculosis control programs place an almost exclusive emphasis on adults with sputum smear-positive tuberculosis, because they are most infectious. However, children contribute a significant proportion of the global tuberculosis caseload and experience considerable tuberculosis-related morbidity and mortality, but few children in endemic areas have access to antituberculosis treatment. The diagnostic difficulty experienced in endemic areas with limited resources has been identified as a major factor contributing to poor treatment access. In general, there is a sense of scepticism regarding the potential value of symptom-based diagnostic approaches, because current clinical diagnostic approaches are often poorly validated. The natural history of childhood tuberculosis demonstrates that symptoms may offer good diagnostic value if they are well defined and if appropriate risk stratification is applied. This study aimed to determine the value of well-defined symptoms to diagnose childhood pulmonary tuberculosis in a tuberculosis-endemic area.

Methods: A prospective, community-based study was conducted in Cape Town, South Africa. Specific well-defined symptoms were documented in all children < 13 years of age reporting a persistent, nonremitting cough of > 2 weeks' duration; study participants were thoroughly evaluated for tuberculosis. In addition, all of the children who received antituberculosis treatment during the study period were reviewed by the investigator, irrespective of study inclusion. This concurrent disease surveillance provided a comprehensive overview of all of the childhood tuberculosis cases, allowing accurate assessment of the possible disadvantages associated with this symptom-based diagnostic approach. In the absence of an acceptable gold standard test, optimal case definition is an important consideration. Children were categorized as "bacteriologically confirmed tuberculosis," "radiologically certain tuberculosis," "probable tuberculosis," or "not tuberculosis." Bacteriologically confirmed tuberculosis was defined as the presence of acid-fast bacilli on sputum microscopy and/or Mycobacterium tuberculosis cultured from a respiratory specimen. Radiologically certain tuberculosis was defined as agreement between both independent experts that the chest radiograph indicated certain tuberculosis in the absence of bacteriologic confirmation. Probable tuberculosis was defined as the presence of suggestive radiologic signs and good clinical response to antituberculosis treatment in the absence of bacteriologic confirmation or radiologic certainty. Good clinical response was defined as complete symptom resolution and weight gain of > or = 10% of body weight at diagnosis, within 3 months of starting antituberculosis treatment. Not tuberculosis was defined as spontaneous symptom resolution or no response to antituberculosis therapy in the absence of bacteriologic confirmation or radiologic signs suggestive of tuberculosis. Pulmonary tuberculosis was defined as a symptomatic child with: (1) bacteriologically confirmed tuberculosis, (2) radiologically confirmed tuberculosis, or (3) probable tuberculosis (as defined), excluding isolated pleural effusion.

Results: In total, 1024 children were referred for evaluation. Resolving symptoms were reported in 596 children (58.2%); 428 (41.8%) children with persistent, nonremitting symptoms at evaluation were investigated for tuberculosis. Pulmonary tuberculosis was diagnosed in 197 children; 96 were categorized as bacteriologically confirmed tuberculosis, 75 as radiologically certain tuberculosis, and 26 as probable tuberculosis. Combining a persistent nonremitting cough of > 2 weeks' duration, documented failure to thrive (in the preceding 3 months), and fatigue provided reasonable diagnostic accuracy in HIV-uninfected children (sensitivity: 62.6%; specificity: 89.8%; positive predictive value: 83.6%); the performance was better in the low-risk group (> or = 3 years; sensitivity: 82.3%; specificity: 90.2%; positive predictive value: 82.3%) than in the high-risk group (< 3 years; sensitivity: 51.8%; specificity: 92.5%; positive predictive value: 90.1%). In children with an uncertain diagnosis at presentation, clinical follow-up was a valuable diagnostic tool that further improved diagnostic accuracy, particularly in the low-risk group. Symptom-based approaches offered little diagnostic value in HIV-infected children. Three (15%) of the 20 HIV-infected children diagnosed with pulmonary tuberculosis failed to report symptoms of sufficient duration to warrant study inclusion, whereas 25% reported persistent, nonremitting symptoms in the absence of tuberculosis. In addition, the tuberculin skin test was positive in < 20% of HIV-infected children diagnosed with pulmonary tuberculosis.

Discussion: The combined presence of 3 well-defined symptoms at presentation (persistent, nonremitting cough of > 2 weeks' duration; objective weight loss [documented failure to thrive] during the preceding 3 months; and reported fatigue) provided good diagnostic accuracy in HIV-uninfected children > or = 3 years of age, with clinical follow-up providing additional value. The approach performed less well in children < 3 years. However, the presence of a persistent, nonremitting cough together with documented failure to thrive still provided a fairly accurate diagnosis (sensitivity: 68.3%; specificity: 80.1%; positive predictive value: 82.1%), illustrating the importance of regular weight monitoring in young children. Clinical follow-up also offered additional diagnostic value, but caution is required, because very young children have an increased risk of rapid disease progression. The approach performed poorly in HIV-infected children. Recent household contact with an adult index case seemed to provide more diagnostic value than a positive tuberculin skin test, but novel T-cell-based assays may offer the only real improvement in sensitivity to diagnose M. tuberculosis infection in HIV-infected children. The variable diagnostic value offered by this symptom-based diagnostic approach illustrates the importance of risk stratification, as demonstrated by the fact that 11 (91.7%) of 12 children with severe disease manifestations who failed to meet the entry criteria were < 3 years of age or HIV infected. Particular emphasis should be placed on the provision of preventive chemotherapy after documented exposure and/or infection in these high-risk children. Study limitations include the small number of HIV-infected children, but on the positive side, the large number of HIV-uninfected children permitted adequate evaluation in this important group. It is often forgotten that HIV-uninfected children constitute the majority of child tuberculosis cases, even in settings where HIV is endemic. This study demonstrates the importance of ascertaining a child's HIV status before symptom-based diagnosis is attempted. Because children were recruited at both the clinic and hospital level, some selection bias may have been introduced; however, the only significant difference between the 2 groups was the proportion of HIV-infected children. Pulmonary tuberculosis was diagnosed with different levels of certainty, but no significant differences were recorded between these groups.

Conclusions: Pulmonary tuberculosis can be diagnosed with a reasonable degree of accuracy in HIV-uninfected children using a simple symptom-based approach. This offers the exciting prospect of improving treatment access for children, particularly in resource-limited settings where current access to antituberculosis treatment is poor.