原发性肝癌患者并发自发性腹膜炎临床诊断模型的构建和应用

Construction and application of clinical diagnostic model for spontaneous peritonitis in patients with primary liver cancer

目的:研究原发性肝癌患者并发自发性腹膜炎的影响因素,并建立和评价列线图预测模型。方法:选择于2012年01月至2022年12月期间在昆明市第三人民医院首次住院的1538例原发性肝癌患者作为研究对象,并按照是否合并自发性腹膜炎分为感染组和对照组,首先利用单因素和Lasso回归分析筛选变量,然后利用多因素logistic回归分析原发性肝癌患者发生自发性腹膜炎的影响因素并建立列线图预测模型,最后利用校准曲线、受试者工作特征(receiver operating characteristic,ROC)曲线、临床决策曲线(clinical decision curve analysis,DCA)、临床影响曲线(clinical impact curve,CIC)对列线图预测模型进行评价。结果:纳入的1538例原发性肝癌患者中有365例合并自发性腹膜炎为感染组,1173例未合并自发性腹膜炎为对照组。单因素分析结果显示,感染组在饮酒、吸烟、CNLC分期、Child-Pugh分级、门静脉癌栓、腹水量、WBC、Hb、LYM、NE、TBil、GGT、AST、ALT、ALB、GLO、Cr、UA、AFP、CEA、PIVKA-Ⅱ、PCT、SAA、IL-6、CRP、IL-4、CD3、CD4、CD8等方面与对照组的差异具有统计学意义(P<0.05)。将这些变量纳入Lasso回归进行变量的筛选,筛选出的变量有6个,分别是Child-Pugh分级、门静脉癌栓、腹水量、WBC、ALB、CRP。多因素logistic回归分析结果显示,原发性肝癌患者并发自发性腹膜炎的影响因素为Child-Pugh分级、门静脉癌栓、腹水量、WBC、CRP,利用这5个变量构建列线图预测模型后对预测模型进行评价,绘制的ROC曲线的曲线下面积(area under curve,AUC)为0.850,灵敏度为0.789,特异度为0.744,绘制的校准曲线与理想曲线基本吻合,平均绝对误差为0.015。结论:原发性肝癌患者并发自发性腹膜炎的影响因素为Child-Pugh分级、门静脉癌栓、腹水量、WBC、CRP,利用这五个变量构建列线图预测模型,经评价发现这五个变量预测自发性腹膜炎的准确性较高,预测值与实际值具有较高的一致性,该模型具有一定的临床实用性。

Objective:To study the influencing factors of spontaneous peritonitis in patients with primary liver cancer,and to establish and evaluate a nomogram prediction model.Methods:A total of 1538 primary liver cancer patients who were first hospitalized in Kunming Third People's Hospital from January 2012 to December 2022 were selected as the study subjects.They were divided into infection group and control group according to whether they were complicated with spontaneous peritonitis.Firstly,univariate and Lasso regression analysis were used to screen variables,and then multivariate logistic regression was used to analyze the influencing factors of spontaneous peritonitis in patients with primary liver cancer and establish a nomogram prediction model.Finally,the calibration curve,receiver operating characteristic(ROC)curve,clinical decision curve analysis(DCA)and clinical impact curve(CIC)were used to evaluate the nomogram prediction model.Results:Among 1538 patients with primary liver cancer,365 patients with spontaneous peritonitis were enrolled as the infection group,and 1173 patients without spontaneous peritonitis were enrolled as the control group.Univariate analysis showed that,in the infection group,drinking,smoking,CNLC stage,Child-Pugh classification,portal vein tumor thrombus,ascites,WBC,Hb,LYM,NE,TBil,GGT,AST,ALT,ALB,GLO,Cr,UA,AFP,CEA,PIVKA-Ⅱ,PCT,SAA,IL-6,CRP,IL-4,CD3,CD4,CD8 and other aspects compared with control group were statistically significant(P<0.05).These variables were included in Lasso regression for variable screening,and 6 variables were selected,which were Child-Pugh classification,portal vein tumor thrombus,ascites,WBC,ALB and CRP.The results of multivariate logistic regression analysis showed that the influencing factors of spontaneous peritonitis in patients with primary liver cancer were Child-Pugh classification,portal vein tumor thrombus,ascites,WBC and CRP.The nomogram prediction model was constructed and evaluated by using these five variables.The area under curve(AUC)of ROC curve was 0.850.The sensitivity was 0.789,and the specificity was 0.744.The calibration curve was basically consistent with the ideal curve,and the average absolute error was 0.015.Conclusion:The influencing factors of spontaneous peritonitis in patients with primary liver cancer are Child-Pugh classification,portal vein tumor thrombus,ascites,WBC and CRP.The nomogram prediction model is constructed using these five variables,and the evaluation found that the accuracy of these five variables in predicting spontaneous peritonitis is high,and the predicted value is highly consistent with the actual value.This model has certain clinical practicability.