3:36 PM Abstract No. 368 Prediction for the recurrence of hemoptysis patients treated with bronchial artery embolization: a novel nomogram and artificial neural network

Abstract

To identify the predictors for recurrence of hemoptysis (RH) patients treated with bronchial artery embolization (BAE), and develop an effective predictive nomogram and artificial nerual network (ANN). The retrospective study included hemoptysis patients treated with first BAE in two institutions, which were allocated to primary cohort and validation cohort. The predictors concerning on RH were identified by univariate and multivariate analyses in primary cohort, and then a predictive nomogram and ANN were developed, with the accuracy evaluated by Harrell’s c statistic and receiver operating characteristic curves in both primary and validation cohort. A total of 242 patients (primary cohort: n = 114; validation cohort: n = 101) were enrolled in this study. After univariate and multivariate analyses, age≥60 (hazard ratio [HR], 3.921; 95% confidence interval [CI], 1.267-12.127; P = 0.018), lung cancer (LC; HR, 18.057; 95% CI, 4.124-79.068; P <0.001), bronchial-pulmonary shunts (BPS; HR, 11.981; 95% CI, 2.593-55.356; P = 0.001) and nonbronchial systemic artery (NBSA; HR, 4.194; 95% CI, 1.596-11.024; P = 0.004) were identified as predictors for RH. The predictive nomogram and ANN were developed and revealed high accuracy, with Harrell’s c statistic of 0.85 internally and 0.80 externally (primary cohort and validation cohort, seperately). Age≥60, LC, BPS and NBSA were identified as predictors for RH. The nomogram and ANN were effective to predict and replensih the early prediction for RH after first BAE.