Derivation, validation and comparative performance of a simplified chest X-ray score for assessing the severity and outcome of pulmonary tuberculosis

Abstract

To derive and validate against the Ralph et al. score, a simplified chest X-ray (CXR) score (SCS) for predicting the outcome of smear-positive pulmonary tuberculosis (SPPTB) among patients with high prevalence of human immunodeficiency virus (HIV) infection. CXR for 636 SPPTB patients (377 men, median age 31 years) from the Yaounde Jamot Hospital (Cameroon) were examined for the presence of tuberculosis lesions, and estimation of the proportion of lungs affected. SPPTB's evolution was based on sputum smear examination after 2 months of treatments. Logistic regressions were used to derive the SCS with internal validation via bootstrap resampling. Receiver operating characteristic curves (AUC) analyses were used to validate, determine of the optimal cut-offs and compare models' performance. HIV co-infection was present in 191 (30%) patients. Sputum smear non-conversion was found in 45 (7.1) patients after 2 months of treatment. The SCS was expressed as proportion of lung affected plus 23 if cavitation is present. The AUC (95% confidence interval) was 0.718 (0.640-0.795), with little variation in bootstrap validation, and no sizable difference from the Ralph et al. model [0.714 (0.633-0.7860), P = 0.717 for the difference). Both models had good calibration (Hosmer-Lemeshow P ≥ 0.183). The optimal cut-off for predicting non-conversion of the sputum at 2 months was 52 for the SCS (sensitivity/specificity 66.7%/62.8%) and 74.5 (62.2%/74.3%) with Ralph et al. score. The outcome of SPPTB among patients with high prevalence of HIV can acceptably be predicted by simple scores comprising baseline CXR variables, including the Ralph et al. score and the SCS from this study.