恶性胆道梗阻金属支架再狭窄的预测模型构建与验证

Construction and validation of a predictive model for metal stent restenosis in malignant biliary obstruction

背景与目的:金属支架和^(125)I粒子腔内照射是恶性胆道梗阻非手术治疗的首选方法,疗效确切,能改善患者生存预后,但金属支架再狭窄发生率较高,导致其发生的因素尚不明确,且缺乏可靠的预测模型。因此,本研究探讨恶性胆道梗阻金属支架再狭窄的影响因素,建立其预测模型并验证性能。方法:选取2019年1月—2022年3月河北省邢台市人民医院收治的110例初次接受金属支架和^(125)I粒子腔内照射恶性胆道梗阻患者,根据术后12个月内金属支架是否再狭窄分为再狭窄组与无再狭窄组。采用单因素、LASSO回归初步筛选金属支架再狭窄的特征变量,Logistic回归进一步分析金属支架再狭窄的相关影响因素,并以R语言绘制列线图预测模型,分别采用一致性指数(C指数)、Hosmer-Lemeshow拟合优度、受试者工作特征曲线下面积(AUC)、临床影响曲线分别评价所构建恶性胆道梗阻金属支架再狭窄预测模型的价值;另选同一中心不同时期的50例初次接受金属支架和^(125)I粒子腔内照射恶性胆道梗阻患者作为外部验证数据集,采用κ检验比较列线图预测再狭窄与临床实际的符合率。结果:110例患者中,术后12个月内共发生再狭窄58例。单因素分析,糖尿病、总胆红素(TBIL)、糖类抗原19-9(CA19-9)、术后胆系感染、胆结石、白蛋白、射频消融、光动力治疗可能与再狭窄有关(均P<0.05)。Logistic回归分析,TBIL、CA19-9、术后胆系感染、胆结石、射频消融、光动力治疗是恶性胆道梗阻金属支架再狭窄的独立相关因素(均P<0.05);基于Logistic回归绘制列线图预测模型显示,其C指数为0.838,Hosmer-Lemeshow拟合优度检验显示,模型预测值与实际观测值之间的差异无统计学意义(χ^(2)=2.796,P=0.803);所构建的恶性胆道梗阻金属支架再狭窄的列线图预测模型的AUC为0.838(95%CI=0.762~0.913),绘制临床影响曲线显示,在各个阈概率下,被所构建的恶性胆道梗阻金属支架再狭窄的列线图预测模型划分为高风险的人数与实际情况的符合度较高;采用列线图预测模型对外部验证数据集进行预测显示,列线图预测再狭窄的发生率与实际情况符合率为94.00%,κ值为0.880。结论:TBIL、CA19-9、术后胆系感染、胆结石、射频消融、光动力治疗是恶性胆道梗阻金属支架再狭窄的独立相关因素,基于以上因素的预测模型具有良好的预测能力,能为临床早期识别高风险人群提供一定的参考。

Background and Aims:Metal stents and intraluminal irradiation with^(125)I seeds are the preferred nonsurgical treatments for malignant biliary obstruction,with proven efficacy in improving patient survival outcomes.However,the incidence of restenosis in metal stents remains high,and the factors contributing to this are not well understood,with no reliable predictive models currently available.Therefore,this study investigated the factors influencing metal stent restenosis in malignant biliary obstruction,developed a predictive model,and validated its performance.Methods:A total of 110 patients with malignant biliary obstruction who received metal stents and^(125)I seed intraluminal irradiation for the first time between January 2019 and March 2022 were selected.Patients were divided into restenosis and non-restenosis groups based on whether restenosis occurred within 12 months after operation.Univariate analysis and LASSO regression were initially used to screen for characteristic variables associated with stent restenosis,followed by Logistic regression to further analyze the related influencing factors.A nomogram predictive model was constructed using R language,and its value was assessed using the concordance index(C-index),Hosmer-Lemeshow goodness-of-fit test,the area under the receiver operating characteristic curve(AUC),and clinical impact curves.An additional 50 patients with malignant biliary obstruction from the same center,who received metal stents and^(125)I seed intraluminal irradiation for the first time during a different period,were selected as an external validation dataset,and theκstatistic was used to compare the concordance rate between the nomogram prediction of restenosis and clinical reality.Results:Of the 110 patients,58 cases of restenosis occurred within 12 months after operation.Univariate analysis showed that diabetes,total bilirubin(TBIL),carbohydrate antigen 19-9(CA19-9),postoperative biliary infection,gallstones,albumin level,radiofrequency ablation,and photodynamic therapy were associated with restenosis(all P<0.05).Logistic regression analysis identified TBIL,CA19-9,postoperative biliary infection,gallstones,radiofrequency ablation,and photodynamic therapy as independent factors associated with metal stent restenosis in malignant biliary obstruction(all P<0.05).The nomogram predictive model based on Logistic regression had a C-index of 0.838,and the Hosmer-Lemeshow goodness-of-fit test indicated no significant difference between the predicted and observed values(χ^(2)=2.796,P=0.803).The AUC of the constructed nomogram for predicting metal stent restenosis in malignant biliary obstruction was 0.838(95%CI=0.762-0.913),and the clinical impact curve showed a high concordance between the predicted high-risk group and actual outcomes across various threshold probabilities.Using the nomogram predictive model to predict the external validation dataset showed a concordance rate of 94.00%between the predicted restenosis rate and the actual one,with aκvalue of 0.880.Conclusion:TBIL,CA19-9,postoperative biliary infection,gallstones,radiofrequency ablation,and photodynamic therapy are independent factors associated with metal stent restenosis in malignant biliary obstruction.The predictive model based on these factors demonstrates good predictive ability and may provide a reference for early clinical identification of high-risk patients.