乳腺癌新辅助化疗血清肿瘤指标差异的随机森林预测模型

Objective To evaluate the predictive value of serum tumor marker dynamics before and after neoadjuvant chemotherapy(NAC) and other factors for pathologic complete response(pCR) in breast cancer, and to develop a random forest prediction model. Methods We retrospectively analyzed clinical data from 84 breast cancer patients undergoing NAC. Patients were grouped into pCR(n=50) and Non-pCR(n=34) based on postoperative pathology. The impact of factors including serum marker changes was assessed using Cox regression and random forest models. Results Batch univariate analysis revealed that lymph node stage(N stage), estrogen receptor status(ER status), progesterone receptor status(PR status), human epidermal growth factor receptor 2 status(HER2 status), completion of 6 cycles of neoadjuvant chemotherapy, anthracycline-containing regimen, platinum-containing regimen,and ΔCA125/ΔFerritin, suggesting predictive potential. Multivariate Cox analysis identified N stage and high ΔCA153 as independent risk factors for shorter DFS, while HER2 positivity was protective. Prediction by the constructed random forest model showed that the changes in serum tumor markers including CA125, carbohydrate antigen 15-3(CA153), ferritin, tumor specific growth factor(TSGF), carbohydrate antigen 72-4(CA724), vascular endothelial growth factor(VEGF) and cytokeratin 19 fragment(CYFRA21) before and after chemotherapy played a core role in the prediction. The area under the curve(AUC) of the model was 0.92 in the training set and 0.875 in the test set. Conclusion This study founal that tumor biological characteristics such as N stage, ER, PR, and HER2 status, as well as chemotherapy regimens and differences in serum tumor markers before and after treatment, are significantly associated with the achievement of pCR after NAC in breast cancer patients.