构建胰十二指肠切除术后胰瘘预测模型及其应用价值

Construction and application value of prediction model of pancreatic fistula after pancreaticoduodenectomy

目的构建胰十二指肠切除术后胰瘘预测模型并探讨其应用价值。方法采用回顾性病例对照研究方法。收集2015年1月至2018年9月哈尔滨医科大学附属第一医院收治的285例行胰十二指肠切除术治疗壶腹周围疾病患者的临床病理资料;男183例,女102例;年龄为(56±14)岁,年龄范围为12~84岁。285例患者通过计算机产生随机数方法以3∶1比例分为训练集214例和测试集71例。训练集用于构建预测模型,测试集用于验证预测模型效能。观察指标:(1)术后胰瘘发生情况。(2)胰十二指肠切除术后胰瘘预测模型构建。(3)胰十二指肠切除术后胰瘘预测模型验证。正态分布的计量资料以±s表示,组间比较采用t检验。偏态分布的计量资料以M(范围)表示,组间比较采用Mann-Whitney秩和检验。计数资料以绝对数或百分比表示,组间比较采用χ2检验。单因素和多因素分析采用Logistic回归模型。绘制受试者工作特征曲线,并计算曲线下面积(AUC)分析预测模型准确性。结果(1)术后胰瘘发生情况:训练集214例患者中,45例术后发生胰瘘,其中B级瘘39例、C级瘘6例。(2)胰十二指肠切除术后胰瘘预测模型构建。单因素分析结果显示:体质量指数(BMI)、X线计算机体层摄影术检查主胰管直径、术中探查主胰管直径、胰腺质地、术后首日腹腔积液淀粉酶是术后胰瘘发生的相关因素(χ2=32.450,15.789,19.577,4.559,Z=-7.962,P<0.05)。多因素分析结果显示:BMI>25 kg/m2、术中探查主胰管直径<3 mm和术后首日腹腔积液淀粉酶>2651 U/L是术后胰瘘发生的独立危险因素(优势比=0.148,4.286,0.086,95%可信区间为0.058~0.376,1.736~10.580,0.032~0.231,P<0.05)。根据多因素分析结果,构建胰十二指肠切除术后胰瘘预测模型:胰瘘预测值=Exp[0.452-1.914(BMI)+1.455(术中探查主胰管直径)-2.451(术后首日腹腔积液淀粉酶)]/1+Exp[0.452-1.914(BMI)+1.455(术中探查主胰管直径)-2.451(术后首日腹腔积液淀粉酶)]。该模型AUC为0.888(95%可信区间为0.832~0.943,P<0.05)。(3)胰十二指肠切除术后胰瘘预测模型验证:在测试集中,胰十二指肠切除术后胰瘘预测模型AUC为0.868(95%可信区间为0.780~0.957,P<0.05)。训练集与测试集AUC比较,差异无统计学意义(Z=0.514,P>0.05)。结论BMI>25 kg/m2,术中探查主胰管直径<3 mm,术后首日腹腔积液淀粉酶>2615 U/L是术后胰瘘的独立危险因素。建立术后胰瘘预测模型可有效预测胰十二指肠切除术后胰瘘风险。

Objective To construct a prediction model of pancreatic fistula after pancreaticoduodenectomy and explore its application value.Methods The retrospective case-control study was conducted.The clinicopathological data of 285 patients with periampullary diseases who underwent pancreaticoduodenectomy in the the First Affiliated Hospital of Harbin Medical University from January 2015 to September 2018 were collected.There were 183 males and 102 females,aged(56±14)years,with a range from 12 to 84 years.According to the random numbers showed in the computer,patients were randomly divided into training dataset consisting of 214 patients and validation dataset consisting of 71 patients,with a ratio of 3∶1.The training dataset was used to construct prediction model,and the validation dataset was used to evaluate performance of prediction model.Observation indicators:(1)incidence of postoperative pancreatic fistula;(2)construction of prediction model of pancreatic fistula after pancreaticoduodenectomy;(3)validation of prediction model of pancreatic fistula after pancreaticoduodenectomy.Measurement data with normal distribution were represented as Mean±SD,and comparison between groups was analyzed by the t test.Measurement data with skewed distribution were represented as M(range),and comparison between groups was analyzed using the Mann-Whitney rank sum test.Count data were described as absolute numbers or percentages,and comparison between groups was conducted using the chi-square test.Univariate analysis and multivariate analysis were conducted using the Logistic regression model.The accuracy of prediction model was analyzed by drawing receiver operating characteristic curve and calculating area under curve(AUC).Results(1)Incidence of postoperative pancreatic fistula:of 214 patients in the training dataset,45 patients had postoperative pancreatic fistula,including 39 of grade B and 6 of grade C,respectively.(2)Construction of prediction model of pancreatic fistula after pancreaticoduodenectomy.Results of univariate analysis showed that body mass index(BMI),diameter of the main pancreatic duct on computed tomography(CT)scan,diameter of the main pancreatic duct by intraoperative exploration,pancreas texture,and level of amylase in ascites at the postoperative first day were related factors for pancreatic fistula after pancreaticoduodenectomy(χ2=32.450,15.789,19.577,4.559,Z=-7.962,P<0.05).Results of multivariate analysis showed that BMI>25 kg/m2,diameter of the main pancreatic duct by intraoperative exploration<3 mm and level of amylase in ascites at the postoperative first day>2651U/L were independent risk factors for pancreatic fistula after pancreaticoduodenectomy(odds ratio=0.148,4.286,0.086,95%confidence interval:0.058-0.376,1.736-10.580,0.032-0.231,P<0.05).Based on results of multivariate analysis,a prediction model of pancreatic fistula after pancreaticoduodenectomy was built:the predicted value of pancreatic fistula=Exp[0.452-1.914(BMI)+1.455(diameter of the main pancreatic duct by intraoperative exploration)-2.451(level of amylase in ascites at the postoperative first day)]/1+Exp[0.452-1.914(BMI)+1.455(diameter of the main pancreatic duct by intraoperative exploration)-2.451(level of amylase in ascites at the postoperative first day)].The model had the AUC of 0.888(95%confidence interval:0.832-0.943,P<0.05).(3)Validation of prediction model of pancreatic fistula after pancreaticoduodenectomy:in the validation dataset,the prediction model of pancreatic fistula after pancreaticoduodenectomy had the AUC of 0.868(95%confidence interval:0.780-0.957,P<0.05).There was no significant difference in the AUC between the training dataset and validation dataset(Z=0.514,P>0.05).Conclusions BMI>25 kg/m2,diameter of the main pancreatic duct by intraoperative exploration<3 mm and level of amylase in ascites at the postoperative first day>2651 U/L are independent risk factors for pancreatic fistula after pancreaticoduodenectomy.Construction of a prediction model of pancreatic fistula after pancreaticoduo-denectomy can effectively predict the risks of postoperative pancreatic fistula.