进展期胃癌脉管侵犯术前列线图预测模型的建立和验证

Establishment and validation of a predictive nomogram model for advanced gastric cancer with lymphovascular invasion

目的探讨进展期胃癌脉管侵犯(LVI)的术前预测因素,建立对应的列线图预测模型并进行内部验证。方法选取经胃肠外科手术切除的进展期胃癌246例。根据术后病理诊断将患者分为LVI阳性组(95例)和LVI阴性组(151例)。收集患者年龄、性别、肿瘤分化、肿瘤大小、肿瘤部位、Borrmann分型、Lauren分型、cT分期、cN分期、系统性免疫炎症指数(SII),并比较2组间上述基线及临床特征的差异,将2组间差异有统计学意义的变量进行多因素Logistic回归,并进一步建立可视化列线图预测模型,运用Bootstrap法对模型预测效能进行内部验证。结果2组间肿瘤大小、Borrmann分型、肿瘤分化、Lauren分型、cT分期、cN分期及SII差异均有统计学意义(P<0.05)。多因素Logistic回归分析结果显示肿瘤大小(OR=2.184,95%CI:1.224~3.898)、Borrmann分型(OR=2.517,95%CI:1.294~4.896)、cT分期(OR=1.860,95%CI:1.045~3.308)、cN分期(OR=1.816,95%CI:1.004~3.285)及SII(OR=1.001,95%CI:1.000~1.002)是进展期胃癌LVI的独立预测因素,依据多因素分析结果建立进展期胃癌LVI的术前列线图预测模型。经内部验证,列线图模型受试者工作特征(ROC)曲线的曲线下面积(AUC)为0.735,分别高于肿瘤大小(0.599)、Borrmann分型(0.564)、cT分期(0.604)、cN分期(0.582)和SII(0.615)。校准曲线显示模型预测LVI的概率与实际发生的概率具有较好一致性。Hosmer-Lemeshow检验显示拟合优度良好(χ^(2)=4.387,P=0.821)。结论建立的列线图预测模型有助于术前预测进展期胃癌LVI发生的概率,可为临床个体化治疗提供指导。

Objective To explore the preoperative predictors of lymphovascular invasion(LVI)in patients with advanced gastric cancer,and establish the corresponding nomogram prediction model and conduct internal validation.Methods A total of 246 cases of advanced gastric cancer who underwent surgical resection in the Department of Gastrointestinal Surgery of Hengshui People′s Hospital from January 2018 to December 2021 were selected.Patients were divided into the LVI positive group and the LVI negative group according to postoperative pathological diagnosis.The age,gender,tumor differentiation,tumor size,tumor site,Borrmann classification,Lauren′s classification,cT stage,cN stage and systemic immune-inflammation index(SII)of patients were collected and compared between the two groups.The predictors that were statistically different between the two groups were subjected to multivariate Logistic regression and further developed into a visual prediction model.Bootstrap method was applied for internal validation of the prediction efficiency of the model.Results The differences of tumor size,Borrmann classification,tumor differentiation,Lauren classification,cT staging,cN staging and SII were statistically significant between the two groups(P<0.05).Multivariate Logistic regression analysis showed that tumor size(OR=2.184,95%CI:1.224-3.898),Borrmann classification(OR=2.517,95%CI:1.294-4.896),cT staging(OR=1.860,95%CI:1.045-3.308),cN staging(OR=1.816,95%CI:1.004-3.285)and SII(OR=1.001,95%CI:1.000-1.002)were independent predictors of LVI in advanced gastric cancer.A preoperative nomogram prediction model for advanced gastric cancer LVI was developed based on results of multivariate analysis.By internal validation,the area under curve(AUC)value of the subject operating characteristic(ROC)curve of the nomogram was 0.735,which was higher than that of tumor size(0.599),Borrmann staging(0.564),cT staging(0.604),cN staging(0.582)and SII(0.615),respectively.The calibration curve showed that the probability of predicted LVI by the nomogram was in a good agreement with the probability of actual LVI occurrence.The Hosmer-Lemeshow test showed good model fit(χ2=4.387,P=0.821).Conclusion The established nomogram prediction model can help to predict the probability of LVI in advanced gastric cancer preoperatively,which can provide a guideline for clinical individualized treatment.