基于SMOTE算法的结直肠癌患者术后发生腹腔感染的预警模型构建

doi:10.3969/j.issn.2095-1264.2025.03.10

首页 > 过刊浏览>2025年第15卷第3期 >363-369. DOI:10.3969/j.issn.2095-1264.2025.03.10

基于SMOTE算法的结直肠癌患者术后发生腹腔感染的预警模型构建
DOI:
                        10.3969/j.issn.2095-1264.2025.03.10
                    
CSTR:
                        
                    
作者:
                        
                        
                    
作者单位:四川大学华西医院 结直肠肿瘤中心，四川 成都，610041
作者简介:杨静，女，护师，研究方向为腹腔镜术后感染。
通讯作者:杨婕，女，硕士，副主任护师，研究方向为加速康复外科护理。
中图分类号:R735.3
基金项目:

Construction of early warning model of postoperative abdominal infection in patients with colorectal cancer based on SMOTE algorithm

Author:

Affiliation:

Colorectal Cancer Center, West China Hospital of Sichuan University, Chengdu, 610041, Sichuan, China

Fund Project:

摘要

图/表

访问统计

参考文献

相似文献

引证文献

资源附件

文章评论

摘要:

目的基于SMOTE算法构建结直肠癌患者术后发生腹腔感染的预警模型。方法将328例结直肠癌患者作为观察对象，根据术后是否发生腹腔感染分为感染组（50例）和非感染组（278例），收集和分析受试者资料，用单因素及Logistic回归分析筛选结直肠癌患者术后发生腹腔感染的影响因素，再通过SMOTE算法重建影响因素的原始数据集，构建风险预警模型并验证其预测效能。结果 328例结直肠癌患者中有50例术后发生腹腔感染，发生率为15.24%。合并糖尿病（X₁）、切口长度（X₂）、手术时间（X₃）、出血量（X₄）、联合脏器切除（X₅）、术后发生吻合口瘘（X₆）、术后引流管留置时间（X₇）是结直肠癌患者术后发生腹腔感染的危险因素（P<0.05）。得到原始预警模型P₁=1/［1+e^{-（-5.009+1.190}^X^1+0.737^X^2+0.905^X^3+0.941^X^4+1.028^X^5+1.047^X^6+0.929^X^7）］和基于SMOTE算法的预警模型P₂=1/［1+e^{-（-6.152+1.251}^X^1+0.788^X^2+0.915^X^3+0.972^X^4+1.109^X^5+1.276^X^6+0.958^X^7）］。基于SMOTE算法的预警模型P₂的AUC值［0.833 （95% CI： 0.792~0.854）］高于原始预警模型P₁的AUC值［0.814 （95% CI： 0.765~0.826）］。结论基于SMOTE算法构建的预警模型AUC值高于原始预警模型，其预测性能更优。

Abstract:

Objective To establish the early warning model of postoperative abdominal infection in patients with colorectal cancer based on SMOTE algorithm.Methods 328 patients with colorectal cancer were divided into abdominal infection group (50 cases) and non-abdominal infection group (278 cases) according to whether postoperative abdominal infection occurred. Data of the subjects were collected and analyzed, and the influencing factors of postoperative abdominal infection in patients with colorectal cancer were screened by univariate and Logistic regression analysis. Then the original data set of influencing factors was rebuilt by SMOTE algorithm to get the risk early warning model, and its prediction efficiency was verified via analyzing the area under the receiver operator characteristic curve (AUC).Results Of 328 patients with colorectal cancer, 50 cases developed abdominal infection after operation, with an incidence of 15.24%. Diabetes mellitus (X₁), incision length (X₂), operative time (X₃), bleeding (X₄), combined organ resection (X₅), postoperative anastomotic fistula (X₆), and duration of postoperative drain retention (X₇) were risk factors for postoperative development of abdominal infection in colorectal cancer patients (P<0.05). Get the original early warning model P₁=1/[1+e-^(1.190^X^1+0.737^X^2+0.905^X^3+0.941^X^4+1.028^X^5+1.047^X^6+0.929^X^7-5.009)] and the SMOTE-based early warning model P₂=1/[1+e-^(1.251^X^1+0.788^X^2+0.915^X^3+0.972^X^4+1.109^X^5+1.276^X^6+0.958^X^7-6.152)]. The AUC of the original early warning model P₁ was 0.814 (95% CI: 0.765-0.826), lower than that of SMOTE-based early warning model P₂, which was 0.833 (95% CI: 0.792-0.854).Conclusion The early warning model based on the SMOTE algorithm has a higher AUC than the original prediction model, with a better prediction performance.