目的

探究食管癌术后吻合口瘘的危险因素，基于危险因素建立风险预测列线图模型，并通过临床决策曲线分析（DCA）评估该模型的预测效能和临床应用价值。

方法

回顾性选取2020 年1 月至2024 年9 月南通市肿瘤医院收治食管癌手术患者342 例，根据术后是否并发吻合口瘘，划分为发生组（n=36）与未发生组（n=306）。比较两组患者基线资料、围手术期相关指标、血清学因子水平差异，运用Lasso 与多因素Logistic 回归筛选食管癌术后吻合口瘘风险因素，构建风险预测列线图模型，并通过受试者工作特征曲线（ROC）验证其预测准确性。通过Bootstrap 法与DCA决策曲线对其进行内部验证，验证列线图模型的稳定性与收益。

结果

发生组与非发生组中合并糖尿病、吸烟史、吻合方式、低密度脂蛋白胆固醇（LDL-C）、白蛋白比较，差异有统计学意义（均P＜0.05）。Lasso 回归显示，糖尿病、吻合方式、白蛋白均可纳入食管癌术后吻合口瘘风险预测模型。二元Logistic 回归显示，有糖尿病、手工吻合方式、低白蛋白水平是影响食管癌术后吻合口瘘的独立危险因素（均P＜0.001）。ROC 结果显示，合并糖尿病、吻合方式、白蛋白、列线图模型预测食管癌术后吻合口瘘曲线下面积分别为0.713、0.726、0.787、0.904；灵敏度分别为0.583、0.583、0.735、0.889；特异度分别为0.843、0.869、0.750、0.788。通过Bootstrap 法（B=1000）行内部验证，C 指数为0.807，表明食管癌术后吻合口瘘预警模型有良好稳定性。决策分析表明，该模型均能带来净收益，明显优于无效策略。

结论

合并糖尿病、手工吻合、低白蛋白为食管癌术后吻合口瘘的独立风险因素，据此建立的风险预测模型效能佳、稳定，且具净收益，优于无策略对照。

Objective

To explore the risk factors of postoperative anastomotic fistula of esophageal cancer, establish a risk prediction nomogram model based on the risk factors, and evaluate the prediction efficiency and clinical application value of this model through clinical decision curve analysis(DCA).

Methods

A total of 342 patients undergoing esophageal cancer surgery in Nantong Cancer Hospital from January 2020 to September 2024 were retrospectively selected and divided into occurrence group (n=36) and non-occurrence group (n=306) according to whether postoperative anastomotic fistula was complicated.The differences in baseline data, perioperative relevant indicators and serological factors between the two groups were compared.Lasso and multivariate Logistic regression were used to screen the risk factors of postoperative anastomotic fistula for esophageal cancer, the risk prediction nomogram model was constructed, and the prediction accuracy was verified by receiver operating characteristic (ROC) curve.Bootstrap method and DCA decision curve were used for internal verification to verify the stability and benefits of the nomogram model.

Results

There were statistically significant differences in diabetes mellitus, smoking history, anastomosis mode, low density lipoprotein cholesterol (LDL-C) and albumin between the occurrence group and the non-occurrence group (all P<0.05).Lasso regression showed that diabetes mellitus, anastomotic mode and albumin could be included in the risk prediction model of anastomotic fistula after esophageal cancer surgery.Binary Logistic regression showed that diabetes mellitus, manual anastomosis and low albumin level were independent risk factors for anastomotic fistula after esophageal cancer surgery (all P<0.001).ROC results showed that the area under the curve predicted by diabetes, anastomotic mode, albumin and nomogram models was 0.713, 0.726, 0.787 and 0.904,respectively.The sensitivity was 0.583, 0.583, 0.735 and 0.889, respectively.The specificity was 0.843, 0.869,0.750 and 0.788, respectively.Internal verification was performed by Bootstrap method (B=1000), and C-indexwas 0.807, indicating that the early warning model of anastomotic fistula after esophageal cancer surgery had good stability.Decision analysis showed that the model can bring net benefits, which was obviously better than the ineffective strategy.

Conclusion

Diabetes mellitus, manual anastomosis and low albumin are independent risk factors for anastomotic fistula after esophageal cancer surgery.The established risk prediction model is effective, stable and has net benefit, which is better than the control without strategy.