基于决策树模型的内镜逆行胰胆管造影术后胆总管结石复发的研究

A decision tree model-based investigation of the recurrence of calculus of common bile duct following endoscopic retrograde cholangiopancreatography

摘要:

目的 通过决策树模型，探讨内镜逆行胰胆管造影术（ERCP）后胆总管结石复发的危险因素及风险。方法 回顾性分析该院2016年1月－2020年12月因胆总管结石行ERCP取石术的500例患者的临床资料，分别建立决策树模型和Logistic回归模型，并绘制受试者操作特征曲线（ROC curve），用曲线下面积（AUC）、敏感度和特异度对两种模型的预测效果进行评价，筛选影响ERCP后胆总管结石复发的独立危险因素，并使用模型预测患者复发风险。结果 决策树分析结果显示，既往有胆道取石史（P = 0.000）、胆总管直径 > 1.60 cm（P = 0.000）和既往有胆囊切除史（P = 0.004）为ERCP后胆总管结石复发的高危危险因素；多因素Logistic回归分析显示，既往有胆囊切除史（P = 0.003）、既往有胆道取石史（P = 0.000）、胆总管直径 > 1.60 cm（P = 0.000）、结石数量 ≥ 2枚（P = 0.001）和胆道支架（P = 0.038）为影响ERCP后胆总管结石复发的独立危险因素，白蛋白 ≥ 40.5 g/L（P = 0.026）为保护性因素。决策树模型与多因素Logistic回归模型相比，总体准确率为92.2%和93.3%，AUC为0.890（95%CI：0.839～0.940）和0.926（95%CI：0.887～0.964），敏感度为87.0%和85.3%，特异度为81.0%和89.2%。结论 既往有胆道取石史、胆总管直径 > 1.60 cm和既往有胆囊切除史是影响结石复发的高危危险因素。决策树模型简单易行，可以较好地预测结石患者的复发风险，可用于筛选ERCP后胆总管结石复发的高危人群，并针对性预防。

Abstract:

Objective The decision tree model was utilized to investigate the risk factors and recurrence probability of calculus of common bile duct following endoscopic retrograde cholangiopancreatography (ERCP).Methods Clinical data of 500 patients who underwent ERCP with calculus of common bile duct from January 2016 to December 2020 were retrospectively analyzed. Decision tree and Logistic regression models were separately established, and the receiver operator characteristic curve (ROC curve) were established, area under the curve (AUC), sensitivity, and specificity were used to evaluate the predictive performance of both models. Independent risk factors of recurrence for calculus of common bile duct after ERCP were identified, and the models were utilized to predict the risk of recurrence in patients.Results The decision tree analysis revealed that the history of choledocholithotomy (P = 0.000), common bile duct diameter > 1.60 cm (P = 0.000), and history of cholecystectomy (P = 0.004) were significant risk factors for calculus of common bile duct recurrence after ERCP. Furthermore, Logistic regression analysis identified independent risk factors for calculus of common bile duct recurrence after ERCP, including the history of cholecystectomy (P = 0.003), history of choledocholithotomy (P = 0.000), common bile duct diameter > 1.60 cm (P = 0.000), number of stones ≥ 2 (P = 0.001), and biliary stent (P = 0.038), albumin level ≥ 40.5 g/L (P = 0.026) as a protective factor. Compared to the Logistic regression model (93.3%), the decision tree model had an overall accuracy of 92.2%, an AUC of 0.890 (95%CI: 0.839 ~ 0.940) and 0.926 (95%CI: 0.887 ~ 0.964), sensitivity of 87.0% and 85.3%, and specificity of 81.0% and 89.2%.Conclusion The history of choledocholithotomy, common bile duct diameter > 1.60 cm, and cholecystectomy significantly contribute to the risk of stone recurrence. The decision tree model offers a simple and user-friendly approach that enhances the prediction accuracy of stone patients' recurrence risk. It can serve as a valuable tool for screening and targeted prevention strategies aimed at high-risk groups susceptible to calculus of common bile duct recurrence after ERCP.