Construction of a prediction model for the preoperative frailty risk of breast cancer patients based on interpretable machine learning algorithms
ZHANG Qing
ZHANG Ying
DONG Jianli
YU Wenlong
XIONG Yinhuan
ZHAO Jiayue
XU Hongmei
Abstract：Objective:To analyze the influencing factors of preoperative frailty in breast cancer patients and to develop a risk prediction model.Methods:A total of 583 inpatients scheduled for surgical treatment in the breast surgery departments of two Grade A tertiary hospitals in Binzhou city,Shandong province,were selected between August 2024 and January 2025.They were randomly divided into a training set(408 cases)and a validation set(175 cases)in a 7∶3 ratio.Variables were screened using both Lasso regression and the Boruta algorithm.Four machine learning algorithms-support vector machine,decision tree,light gradient boosting machine,and extreme gradient boosting-were employed to develop prediction models.Model performance was compared based on accuracy,precision,sensitivity,specificity,F1-score,and the area under the receiver operating characteristic(ROC)curve(AUC).And the optimal model was interpreted using the SHAP method.Results:The incidence of preoperative frailty in breast cancer patients was 26.93%.Compared to the support vector machine,decision tree,and light gradient boosting machine models,the extreme gradient boosting model demonstrated the best performance,with an AUC of 0.909,accuracy of 0.829,precision of 0.646,sensitivity of 0.857,specificity of 0.818,and an F1-score of 0.737.The SHAP bar plot identified the top five influencing factors as age,hemoglobin,albumin,neutrophil percentage,and comorbidities.Conclusions:The extreme gradient boosting model exhibits the best predictive performance and can serve as a reliable tool for healthcare providers to effectively assess and scientifically manage preoperative frailty in breast cancer patients.
Keywords：breast cancerfrailtypreoperativemachine learningprediction modelinfluencing factorscomorbidities
Publication Date：2026-01-10
Online Publishing Date：2026-01-12(First online date of this platform, not the publication date of the document)
Pages：11( 60-70 )