实用临床医药杂志2026，Vol.30Issue(16)：58-64,7.DOI:10.7619/jcmp.20260357
基于机器学习多模态数据构建的乳腺癌新辅助治疗后病理完全缓解预测模型的应用价值
Application value of prediction model for pathological complete response after neoadjuvant therapy for breast cancer based on multimodal data of machine learning
摘要
Abstract
Objective To construct a predictive model for pathological complete response(pCR)after neoadjuvant therapy in breast cancer based on machine learning-derived multimodal data and ana-lyze its application value.Methods This study included 456 patients with neoadjuvant therapy fol-lowed by surgery after pathological confirmation of breast cancer at Xuzhou Cancer Hospital and the Affiliated Hospital of Xuzhou Medical University from January 2020 to October 2025.Clinical and pathological characteristics were recorded,including imaging findings,serum tumor markers[serum human epidermal growth factor receptor 2(s-HER-2),carcinoembryonic antigen(CEA),carbohy-drate antigen 125(CA125),carbohydrate antigen 153(CA153)],blood routine indicators(white blood cell count,lymphocyte-to-neutrophil ratio,etc.),and immunohistochemical indicators such as the positive expression rates of estrogen receptor(ER),progesterone receptor(PR),prolifera-ting cell nuclear antigen(Ki-67),and human epidermal growth factor receptor 2(HER-2)in histo-pathology.Six machine learning algorithms,including elastic net(EN),random forest(RF),sup-port vector machine(SVM),K-nearest neighbors(KNN),stochastic gradient boosting(SGB),and linear discriminant analysis(LDA),were employed for feature modeling and parameter tuning to evaluate model performance and ultimately select the machine learning model with the best predic-tive performance.Results A total of 456 patients who completed at least two cycles of neoadjuvant therapy for breast cancer were included in this study,among whom 279 patients achieved pCR with a median age of 54 years,and 177 patients did not achieve pCR with a median age of 54 years.No significant difference in age was observed between the pCR and non-pCR groups(P＞0.05).Through comprehensive evaluation using multidimensional indicators,among the six machine learn-ing models(EN,RF,SGB,LDA,SVM,and KNN),the RF model demonstrated superior predic-tive performance in the test set(AUC=0.916,optimal cutoff value of 0.664 based on the Youden index,sensitivity of 81.82％,specificity of 91.43％,95％CI,0.853 9 to 0.978 4),exhibited high average performance(mean value of 0.919),achieved the highest median cross-validation AUC(median value of 0.924),and showed reliable generalization,thus being selected as the opti-mal predictive model.Serum tumor markers such as s-HER-2,CA153,CA125 and CEA were sig-nificantly correlated with pCR and demonstrated significant predictive value in most models,serving as key variables for predicting pCR.Conclusion RF model is the optimal model for predicting pCR after neoadjuvant therapy in breast cancer in terms of comprehensive performance.Serum tumor markers such as s-HER-2,CA153,CA125,and CEA are the most important predictors of pCR.关键词
乳腺癌新辅助治疗/病理完全缓解/机器学习/血清肿瘤标志物/预测模型/随机森林模型/血清人表皮生长因子受体2Key words
neoadjuvant therapy for breast cancer/pathological complete response/machine learning/serum tumor markers/predictive model/random forest model/serum human epidermal growth factor receptor 2分类
医药卫生引用本文复制引用
赵利,曹生亚,王宇新,朱玉芝,张敬,丁爽..基于机器学习多模态数据构建的乳腺癌新辅助治疗后病理完全缓解预测模型的应用价值[J].实用临床医药杂志,2026,30(16):58-64,7.基金项目
徐州市卫生健康委基金资助项目(XWKYHT20240046) （XWKYHT20240046）
