A Diagnostic Predictive Model of Bronchoscopy with Radial Endobronchial Ultrasound for Peripheral Pulmonary Lesions

Abstract

Background: Several factors have been reported to affect the diagnostic yield of bronchoscopy with radial endobronchial ultrasound (R-EBUS) for peripheral pulmonary lesions (PPLs). However, it is difficult to accurately predict the diagnostic potential of bronchoscopy for each PPL in advance. Objectives: Our objective was to establish a predictive model to evaluate the diagnostic yield before the procedure. Method: We retrospectively analysed consecutive patients who underwent diagnostic bronchoscopy with R-EBUS between April 2012 and October 2015. We assessed the factors that were predictive of successful bronchoscopic diagnosis of PPLs with R-EBUS using a multivariable logistic regression model. The accuracy of the predictive model was evaluated using the receiver operator characteristic area under the curve (ROC AUC). Internal validation was analysed using 10-fold stratified cross-validation. Results: We analysed a total of 1,634 lesions; the median lesion size was 25.0 mm. Of these, 1,138 lesions (69.6%) were successfully diagnosed. In the predictive logistic model, significant factors affecting the diagnostic yield were lesion size, lesion structure, bronchus sign, and visible on chest X-ray. The predictive model consisted of seven factors: lesion size, lesion lobe, lesion location from the hilum, lesion structure, bronchus sign, visibility on chest X-ray, and background lung. The ROC AUC of the predictive model was 0.742 (95% confidence interval: 0.715–0.769). Internal validation using 10-fold stratified cross-validation revealed a mean ROC AUC of 0.734. Conclusions: The predictive model using the seven factors revealed a good performance in estimating the diagnostic yield.