内镜逆行胰胆管造影术后胰腺炎风险预测模型的构建及验证

Construction and validation of a risk predictive model for post endoscopic retrograde cholangiopancreatography pancreatitis

摘要:

目的 探讨患者行内镜逆行胰胆管造影术（ERCP）后，发生术后胰腺炎的危险因素，建立定量风险预测模型，并进行外部验证。方法 回顾性分析该院行ERCP的患者的临床资料。其中，603例于2017年1月－2021年1月行ERCP的患者为建模组，205例于2021年3月－2022年3月行ERCP的患者为验证组，根据亚特兰大标准，建模组45例ERCP后发生胰腺炎，验证组23例ERCP后发生胰腺炎。比较建模组胰腺炎患者与无胰腺炎患者的临床资料和生化指标，采用多因素Logistic回归模型，分析ERCP后发生胰腺炎的危险因素，建立风险预测模型，并进行外部验证。结果 单因素分析发现，两组患者年龄、胃切除术史、胆总管结石、乳头孔类型、胰腺导丝通道、括约肌切开术、血清总胆红素和白蛋白比较，差异均有统计学意义（P < 0.05）。多因素Logistic回归分析显示，胃切除术史（OR＾ = 6.417，95%CI：1.900～21.675，P = 0.000）、胆总管结石（OR＾ = 3.442，95%CI：1.496～7.917，P = 0.000）、结节型乳头孔（OR＾ = 2.447，95%CI：1.072～5.585，P = 0.018）、胰腺导丝通道（OR＾ = 3.673，95%CI：1.609～8.383，P = 0.000）、括约肌切开术（OR＾ = 1.758，95%CI：1.140～2.711，P = 0.004）、总胆红素升高（OR＾ = 1.415，95%CI：1.084～1.847，P = 0.008）和白蛋白降低（OR＾ = 1.239，95%CI：1.016～1.510，P = 0.010），是ERCP后发生胰腺炎的独立危险因素。建立风险预测模型Y = -1.023 + 1.859×（胃切除术史）+ 1.236×（胆总管结石）+ 0.895×（结节型乳头孔）+ 1.301×（胰腺导丝通道）+ 0.564×（括约肌切开术）+ 0.347×（总胆红素升高）+ 0.214×（白蛋白降低）。受试者操作特征曲线（ROC curve）显示，在建模组与验证组中，模型预测胰腺炎的曲线下面积（AUC）分别为0.895和0.864。对模型中各个变量进行赋值后，分为低风险（0～5分）、中风险（5～10分）和高风险（ ≥ 10分），建模组与验证组中，高风险患者胰腺炎实际发生率明显高于低风险患者，高风险患者胰腺炎实际发生率明显高于中风险患者，差异均有统计学意义（P < 0.05）。结论 胃切除术史、胆总管结石、结节型乳头孔、胰腺导丝通道、括约肌切开术、总胆红素升高和白蛋白降低，是ERCP后发生胰腺炎的独立危险因素。笔者建立的定量风险预测模型，对胰腺炎的预测效能好，有重要的临床应用价值。

Abstract:

Objective To investigate the risk factors of postoperative pancreatitis after endoscopic retrograde cholangiopancreatography (ERCP), establish a quantitative risk prediction model, and conduct external validation.Methods The clinical data of patients with ERCP were analyzed. Among them, 603 ERCP patients From January 2017 to January 2021 were selected as the modeling group, and 205 ERCP patients from March 2021 to March 2022 were selected as the validation group. Pancreatitis was diagnosed according to Atlanta standards. There were 45 cases in the modeling group and 23 cases in the validation group developed pancreatitis after ERCP. Compare the clinical data and biochemical indicators of patients with and without pancreatitis in the modeling group, and screen for risk factors of pancreatitis through multivariate Logistic regression analysis. Then, establish a risk prediction model and validate it.Results Univariate analysis showed that there were statistically significant differences in age, history of gastectomy, calculus of common bile duct, papillary foramen nodule type, pancreatic wire channel, sphincterotomy, serum total bilirubin and albumin between the two groups (P < 0.05). Multivariate Logistic regression analysis showed that the history of gastrectomy (OR＾ = 6.417, 95%CI: 1.900 ~ 21.675, P = 0.000), calculus of common bile duct (OR＾ = 3.442, 95%CI: 1.496 ~ 7.917, P = 0.000), papillary foramen nodule type (OR＾ = 2.447, 95%CI: 1.072 ~ 5.585, P = 0.018), pancreatic wire channel (OR＾ = 3.673, 95%CI: 1.609 ~ 8.383, P = 0.000), sphincterotomy (OR＾ = 1.758, 95%CI: 1.140 ~ 2.711, P = 0.004), elevated total bilirubin (OR＾ = 1.415, 95%CI: 1.084 ~ 1.847, P = 0.008) and decreased albumin (OR＾ = 1.239, 95%CI: 1.016 ~ 1.510, P = 0.010) were independent risk factors for post ERCP pancreatitis. Establish a risk prediction model Y = -1.023 + 1.859×(history of gastrectomy) + 1.236×(calculus of common bile duct) + 0.895×(papillary foramen nodule type) + 1.301×(pancreatic wire channel) + 0.564× (sphincterotomy) + 0.347×(elevated total bilirubin) + 0.214×(decreased albumin). The receiver operator characteristic curve (ROC curve) showed that the area under the curve (AUC) of the model predicting pancreatitis in the modeling and validation groups were 0.895 and 0.864, respectively. After assigning values to each variable in the model, it was divided into low risk (0 ~ 5 points), medium risk (5 ~ 10 points), and high risk ( ≥ 10 points). The actual incidence of high-risk pancreatitis in the modeling and validation groups was significantly higher than that of low-risk patients, and the actual incidence of high-risk pancreatitis was significantly higher than that of medium risk patients, with statistical significance (P < 0.05).Conclusion The history of gastrectomy, calculus of common bile duct, papillary foramen nodule type, pancreatic wire channel, sphincterotomy, elevated total bilirubin and decreased albumin were independent risk factors for post ERCP pancreatitis. We have developed a quantitative risk prediction model with good predictive efficacy for pancreatitis, which has important clinical application value.