Recurrent Hemoptysis After Bronchial Artery Embolization: Prediction Using a Nomogram and Artificial Neural Network Model.

Abstract

OBJECTIVE. The purpose of this study was to develop an effective nomogram and artificial neural network (ANN) model for predicting recurrent hemoptysis after bronchial artery embolization (BAE). MATERIALS AND METHODS. The institutional ethics review boards of the two participating hospitals approved this study. Patients with hemoptysis who were treated with BAE were allocated to either the training cohort (Hospital A) or the validation cohort (Hospital B). The predictors of recurrent hemoptysis were identified by univariable and multivariable analyses in the training cohort. A nomogram and ANN model were then developed, and the accuracy was validated by the Harrell C statistic and ROC curves in both the training and validation cohorts. RESULTS. A total of 242 patients (training cohort, 141; validation cohort, 101) were enrolled in this study. The univariable and multivariable analyses revealed that age of 60 years old or older (hazard ratio [HR], 3.921; 95% CI, 1.267-12.127; p = 0.018), lung cancer (HR, 18.057; 95% CI, 4.124-79.068; p < 0.001), bronchial-pulmonary shunts (HR, 11.981; 95% CI, 2.593-55.356; p = 0.001), and nonbronchial systemic artery involvement (HR, 4.194; 95% CI, 1.596-11.024; p = 0.004) were predictors of recurrent hemoptysis. The developed nomogram and ANN model had high accuracy, with a Harrell C statistic of 0.849 (95% CI, 0.778-0.919) internally (for the training cohort) and 0.799 (95% CI, 0.701-0.897) externally (for the validation cohort). The optimal cutoff value of the recurrent hemoptysis risk was 0.16. CONCLUSION. The nomogram and ANN model could effectively predict the risk for recurrent hemoptysis after BAE. Further interventions should be considered for patients with a high suspicion of risk (> 0.16) according to the nomogram.