Evaluation of the utility of specific CXR features for diagnosis of pulmonary tuberculosis in young children using multiple readers

• Main Author: Ho-Yee, Ruschka
• Other Authors: Andronikou, Savvas
• Format: Dissertation
• Language: English
• Published: University of Cape Town 2016
• Subjects: Diagnostic Radiology
• Online Access: http://hdl.handle.net/11427/16594

Includes bibliographical references === INTRODUCTION: The diagnosis of childhood pulmonary tuberculosis (TB) can be notoriously difficult. The chest X-ray (CXR) is a significant diagnostic resource in the detection of PTB in children. However, non-specific radiological features combined with variable inter-observer assessment s contribute to diagnostic uncertainty. The CXR would be of most value when used specifically to evaluate those features of childhood TB that it shows best and where expert observers agree, namely those signs indicating lymphadenopathy. AIM: To identify simple and reliable CXR features of primary TB in children by determining signs and anatomical sites of best observer agreement. METHOD: This is a retrospective descriptive study within a clinical trial performed by the South African TB Vaccine Initiative (SATVI). Healthy BCG-vaccinated newborn infants in a high TB prevalence rural area in Worcester, near Cape Town, South Africa, were followed for a minimum of two years for possible incident al pulmonary TB. Three independent, blinded, expert paediatric radiologists reported the resultant CXR images using a standardised data collection tick sheet, on which the specific anatomical sites and signs of pathology consistent with pulmonary TB were recorded. The first 200 original data collection tick sheets were sampled and recorded in a pre-compiled data spreadsheet for our study. The sampled data were t hen analysed using kappa statistics. RESULTS: The overall combined agreement for airway compression (by presumed lymphadenopathy) was 0.5%. Five % of the CXR's had soft tissue densities reflecting lymphadenopathy on the frontal view and 5% on the lateral view. The most common site reflecting lymphadenopathy through airway narrowing or displacement was the left main bronchus. The hilar region (kappa 0.27) on the frontal CXR and behind bronchus intermedius (kappa 0.18) on the lateral were the most common sites of soft tissue densities reflecting lymphadenopathy. There were no positive findings for cavitation or pleural effusion. The overall decisions reflecting PTB (lymphadenopathy or miliary) by each individual reader were 27.6% by Reader 1, 8.5% by Reader 2 and 24.6 % by Reader 3. Abnormal findings not specific for PTB were found in 3.5 % by Reader 1, 10.5% by Reader 2 and 3.5% by Reader 3.68. 3 % of the radiographs were reported as normal by Reader 1, 81.9% by Reader 2 and 66.8 % by Reader 3. Only 5% of the radiographs were found to be unreadable by one reader. The overall agreement of all three readers on PTB was 14.6 % and for normal CXR 49.2%. CONCLUSIONS: The fair degree of agreement amongst expert readers suggests that the CXR alone is not a reliable tool for detecting pulmonary TB and should be utilised in conjunction with the clinical features and/or skin tests and blood results. Soft tissue masses rather than airway compression are a more reliable sign for lymphadenopathy, with the most agreed upon sites on the frontal projection for soft tissue mass detection being the right hilar region, followed by the left hilum. Unfortunately, this study could not confirm the usefulness of the CXR in subcategorising PTB into severe and non-severe groups due to the absence of any positive features for severe PTB in the selected sample. The use of prescribed tick-sheets with specified features for detecting lymphadenopathy did not have the expected impact of promoting interobserver consensus of CXR findings in children in terms of detection of TB. The absence of a credible reference standard for lymphadenopathy remains a significant limitation.