Objective: To refine and evaluate a recently published radiological disease severity score for the prediction of month 2 and end of treatment outcomes in pulmonary tuberculosis (TB). Radiological extent of disease has been linked to early and late outcomes of anti-tuberculosis treatment, but no validated tools are available to quantify this parameter.
Design: We enrolled 449 adult, human immunodeficiency virus negative participants with smear- or culture-proven TB from three TB biomarker studies in Cape Town, South Africa. Full-size posteroanterior baseline chest X-rays (CXRs) were evaluated by two clinicians after standardising the published scoring method and the predictive ability assessed for month 2 and final treatment outcomes.
Results: Baseline CXR scores were significantly different in the favourable and unfavourable outcome groups; however, the predictive ability for outcomes at all time points was poor (ROC area under curve ⩿0.68). Inter-reader reliability was high (r = 0.86, P < 0.001), but agreement in cavity identification was modest.
Conclusion: Standardised application of a CXR score derived from the presence of cavities and overall extent of parenchymal disease in active TB showed good inter- and intrareader reliability. Scores differed significantly in treatment outcome groups, but did not allow accurate outcome prediction.