基于决策树算法的胃癌患者腹腔镜术后发生肺动脉栓塞的风险预测模型的建立

Establishment of a Risk Prediction Model for Pulmonary Embolism in Gastric Cancer Patients after Laparoscopic Surgery Based on Decision Tree Algorithm

目的:对胃癌患者在腹腔镜手术后发生肺动脉栓塞(PE)的影响因素进行分析,并建立决策树模型。方法:回顾性收集我院于2018年1月—2022年6月行腹腔镜手术的胃癌患者118例为研究对象,依据胃癌患者在腹腔镜手术后是否发生肺动脉栓塞,分为肺动脉栓塞组与非肺动脉栓塞组,分析胃癌患者腹腔镜术后发生肺动脉栓塞的危险因素,采用SPSS Modeler软件构建胃癌患者腹腔镜术后发生肺动脉栓塞的决策树模型,并分析其预测效能。结果:单因素分析结果显示,年龄、围术期是否输注红细胞、下肢DVT、手术时间、术后第1天D-二聚体水平等资料皆存在统计学差异(P<0.05);多因素Logistic回归分析结果显示,年龄>60岁、围术期输注红细胞、下肢深静脉血栓形成(DVT)、手术时间≥3h、术后第1天D-二聚体水平是胃癌患者腹腔镜术后发生肺动脉栓塞的危险因素;构建了胃癌患者腹腔镜术后发生肺动脉栓塞的决策树模型,决策树模型共5层,13个节点,模型选择了年龄>60岁、围术期输注红细胞、下肢DVT、手术时间≥3h、术后第1天D-二聚体水平5个临床特征作为模型的节点,其中重要性排名第一的预测因子是术后第1天D-二聚体水平。胃癌患者腹腔镜术后发生肺动脉栓塞的决策树模型AUC是0.859(95%CI:0.783~0.917),胃癌患者腹腔镜术后发生肺动脉栓塞的Logistic回归模型的AUC是0.763(95%CI:0.676~0.836),决策树模型与Logistic回归模型的delong检验结果为Z=2.350,P=0.019。结论:年龄>60岁、围术期输注红细胞、下肢DVT、手术时间≥3h、术后第1天D-二聚体水平是胃癌患者腹腔镜术后发生肺动脉栓塞的危险因素。本研究构建的胃癌患者腹腔镜术后发生肺动脉栓塞的决策树模型预测效能较为优异,有利于早期精准鉴别胃癌患者腹腔镜术后PE的发生。

Objective:Analyzing the influencing factors of pulmonary embolism(PE)in gastric cancer patients after laparoscopic surgery and establishing a decision tree model.Methods:A retrospective study was conducted on 118 gastric cancer patients who underwent laparoscopic surgery at our hospital,from January 2018 to June 2022.Based on whether pulmonary embolism occurred after laparoscopic surgery in gastric cancer patients,they were divided into the pulmonary embolism group and the non-pulmonary embolism group.The risk factors for pulmonary embolism in gastric cancer patients after laparoscopic surgery were analyzed.The decision tree model for predicting the occurrence of pulmonary embolism in gastric cancer patients after laparoscopic surgery was constructed using SPSS Modeler software,and its predictive performance was analyzed.Results:Univariate analysis results showed that age,perioperative red blood cell transfusion,lower extremity deep vein thrombosis(DVT),surgery duration,and D-dimer levels on the first postoperative day had statistically significant differences(P<0.05)in relation to the occurrence of pulmonary embolism after laparoscopic surgery in gastric cancer patients.Multivariate Logistic regression analysis results showed that age>60 years,perioperative red blood cell transfusion,lower extremity DVT,surgery duration≥3 hours,and D-dimer levels on the first postoperative day were risk factors for the occurrence of pulmonary embolism after laparoscopic surgery in gastric cancer patients.A decision tree model for predicting the occurrence of pulmonary embolism after laparoscopic surgery in gastric cancer patients was constructed.The decision tree model had 5 layers,13 nodes,and selected age>60 years,perioperative red blood cell transfusion,lower extremity DVT,surgery duration≥3 hours,and D-dimer levels on the first postoperative day as the clinical features for the model’s nodes.The most important predictor,ranked first in importance,was the D-dimer level on the first postoperative day.The area under the curve(AUC)of the decision tree model for predicting the occurrence of pulmonary embolism after laparoscopic surgery in gastric cancer patients was 0.859(95%CI:0.783~0.917),while the AUC of the Logistic regression model was 0.763(95%CI:0.676~0.836).The Delong test between the decision tree model and the Logistic regression model yielded a Z value of 2.350 and a P value of 0.019,indicating a significant difference between the two models.Conclusion:Age>60 years,perioperative red blood cell transfusion,lower extremity DVT,surgery duration≥3 hours,and D-dimer levels on the first postoperative day were identified as risk factors for the occurrence of pulmonary embolism after laparoscopic surgery in gastric cancer patients.The decision tree model for predicting the occurrence of pulmonary embolism after laparoscopic surgery in gastric cancer patients constructed in this study demonstrated excellent predictive performance,which is beneficial for early and accurate identification of pulmonary embolism after laparoscopic surgery in gastric cancer patients.