Comparison of value of Logistic regression model and XGBoost model in predicting upper limb lymphedema after radical mastectomy
10.3760/cma.j.cn115807-20241031-00339
- VernacularTitle:基于Logistic回归模型和XGBoost模型预测乳腺癌根治术后并发上肢淋巴水肿的价值
- Author:
Yanyan WU
1
;
Yajie ZHANG
1
;
Yan LIU
1
Author Information
1. 济宁医学院附属医院乳腺外科，济宁 272000
- Publication Type:Journal Article
- Keywords:
Logistic regression model;
XGBoost model;
Breast cancer;
Surgery;
Upper limb lymphedema
- From:
Chinese Journal of Endocrine Surgery
2025;19(2):172-176
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To study the value of Logistic regression model and XGBoost model in predicting upper limb lymphedema after radical mastectomy.Methods:A total of 150 patients with breast cancer admitted to the Department of Breast Surgery of the Affiliated Hospital of Jining Medical College from Jan. 2023 to Jan. 2024 were collected as the study objects, all of whom underwent surgical treatment. The subjects were divided into the occurrence group (30 cases) and the non-occurrence group (120 cases) according to whether upper limb lymphedema occurred after operation. The clinical data of the patients in the study were collected, and the risk prediction models of upper limb lymphedema after radical mastectomy were constructed based on Logistic regression analysis and XGBoost method, respectively, and the predictive value of the two models was compared.Results:Among the 150 patients, 30 patients developed upper limb lymphedema after surgery, with an incidence of 20.00% (30/150) . The age, body mass index (BMI) level, clinical stage (stage III) , number of excised lymph nodes ≥ 10, postoperative non functional exercise, postoperative adjuvant radiotherapy, and proportion of hypertension in the incidence group were higher than those in the non incidence group ( t=3.21, 5.21, χ 2=30.68, 16.24, 12.40, 11.11, 8.58, P<0.05) . Logistic regression analysis showed that BMI, clinical stage (stage III) , number of lymph nodes removed ≥10, postoperative non-functional exercise, postoperative adjuvant radiotherapy and hypertension were all risk factors for upper limb lymphedema after radical mastectomy ( P<0.05) . ROC curve showed that the area under the curve (AUC) of multivariate Logistic regression and XGBoost model predicting upper limb lymphedema after radical mastectomy was 0.886 (95% CI: 0.843-0.941) and 0.874 (95% CI: 0.829-0.935) , respectively. There was no significant difference in AUC between the two models ( Z=1.34, P=0.462) . The Calibration curves of multivariate Logistic regression and XGBoost model were tested by Hosmer-Lemeshow, but there was no significant difference ( χ 2=0.83, P=0.516; χ 2=0.71, P=0.637) . Both models fit well. Conclusions:Multiple Logistic regression and XGBoost model were equally effective in predicting upper limb lymphedema after radical mastectomy. BMI, clinical stage (stage III) , number of lymph nodes removed ≥10, postoperative non-functional exercise, postoperative adjuvant radiotherapy, and hypertension are all important factors affecting upper limb lymphedema after radical mastectomy.
