Nomogram Analysis and Internal Validation to Predict the Risk of Cystobiliary Communication in Patients Undergoing Hydatid Liver Cyst Surgery

Abstract

Abstract Purpose Biliary leakage caused by cystobiliary communication (CBC) is a common clinical concern. This study sought to identify predictors of CBC in hepatic cystic echinococcosis (HCE) patients undergoing hydatid liver cyst surgery and establish nomograms to predict CBC. Methods A predictive model was established in a training cohort of 310 HCE patients diagnosed between January 2013 and May 2017. Upon revision of the records of clinical parameters and imaging features of these patients, the lasso regression model was used to optimize feature selection for the CBC risk model. Combined with feature selection, a CBC nomogram was developed with multivariable logistic regression. C-index and calibration plots were used to analyze and evaluate the discrimination and calibration. The net benefit and predictive accuracy of the nomogram were performed via decision curve analysis (DCA) and receiver operating characteristic (ROC) curve. An independent validation cohort of 132 patients recruited from June 2017 to May 2019 was used to evaluate the practicability of the nomogram. Results Predictors contained four features, namely alkaline phosphatase (ALP), glutamyl transpeptidase (GGT), cyst size and cyst location. The C-index of the nomogram is 0.791 (95% CI, 0.736–0.845), while the C-index verified by bootstrap is 0.746, indicating high prediction accuracy. The area under the curve (AUC) of the CBC in training was 0.766. ROC curve analysis demonstrated high sensitivity and specificity. Decision curve analysis confirmed the CBC nomogram was clinically useful when the intervention was determined at the non-adherence possibility threshold of 8%. Conclusion The nomogram developed using the ALP, GGT, cyst size and cyst location could be used to facilitate the CBC risk prediction in HCE patients.