Reliability of chest radiograph interpretation for pulmonary tuberculosis in the screening of childhood TB contacts and migrant children in the UK

Abstract

To determine interobserver agreement between paediatric radiologists interpreting tuberculosis (TB) screening chest radiographs (CXR) in children in the UK, and the TB detection rate. A retrospective review was undertaken of electronic request, notes, and CXRs (>7 years) in children exposed to an infectious case of TB and new entrants to the UK, at a tertiary children's hospital. Included were those with positive Mantoux/interferon gamma release assay (IGRA), positive culture, or high clinical suspicion of TB. CXR reports were categorised as normal, abnormal without features of TB, or abnormal with features of pulmonary TB. Three paediatric radiologists from a dedicated paediatric radiology department evaluated available CXRs, aware of the TB screening indication, using a published CXR reporting tool and blinded to the initial CXR report and to each other. A majority decision was collated, and considered lymphadenopathy and miliary nodules as diagnostic of primary TB. Interobserver agreement was calculated using Cohen's kappa. One hundred and forty-eight children underwent TB screening with a CXR. One hundred and twenty-five had available CXR reports and case notes, which indicated 20/125 (16%) had CXR features of TB. One hundred and twenty-one of the 125 had CXRs available to for panel review. Twenty of these 121 (17%) yielded a majority decision of pulmonary TB. Inter-reader agreement was moderate in all aspects (kappa 0.4-0.6). The high percentage of pulmonary TB on CXR (16% original reports; 17% by panel review) suggests that it is worthwhile investigating childhood TB contacts; however, the routine use and recommendation for CXR is questionable because of only moderate interpretation reliability (kappa 0.5), even by experts.