A clinicopathological-MRI model for predicting pathological complete response after neoadjuvant therapy in breast cancer: a single-center retrospective study with transcriptomic pathway exploration

Background Accurate pretreatment prediction of pathological complete response (pCR) after neoadjuvant therapy may help guide individualized management in breast cancer. This study aimed to develop and internally validate a practical pCR prediction model based on routinely available clinicopathological and MRI-report features and to explore pCR-related biological pathways using a public transcriptomic dataset. Methods This retrospective study included 201 patients with breast cancer who received neoadjuvant chemotherapy followed by surgery at the General Hospital of Northern Theater Command from April 2021 to May 2026. Patients were divided into pCR and non-pCR groups according to postoperative pathology. Clinicopathological and MRI-report features were compared between groups. Univariate and multivariable logistic regression analyses were performed to identify independent predictors of pCR, and three models were constructed: a clinicopathological model, an MRI-report model, and a clinicopathological-MRI combined model. Model performance was evaluated using receiver operating characteristic curves, DeLong tests, calibration analysis, decision curve analysis, and 1000-times bootstrap internal validation. Hallmark gene set enrichment analysis (GSEA) was additionally performed using GSE194040 (I-SPY2-990) to explore pCR-related pathway differences. Results Among the 201 patients, 60 achieved pCR and 141 had non-pCR. Multivariable logistic regression identified HER2-positive status, Ki-67 index, and MRI maximum tumor diameter as independent predictors of pCR. The AUC values of the clinicopathological model, MRI-report model, and combined model were 0.880, 0.703, and 0.910, respectively. The combined model significantly outperformed both the clinicopathological model and the MRI-report model. Calibration analysis and decision curve analysis suggested good agreement and favorable clinical net benefit for the combined model. Bootstrap internal validation further supported model stability, with a bootstrap AUC of 0.913 (95% CI: 0.864–0.953). Exploratory GSEA indicated that differences between pCR and non-pCR may involve TGF-beta signaling, TNF-alpha/NF-kappaB signaling, estrogen response, KRAS signaling, and coagulation pathways, with heterogeneity across HR/HER2 subgroups. Conclusions A clinicopathological-MRI combined model incorporating HER2 status, Ki-67 index, and MRI maximum tumor diameter showed favorable performance for predicting pCR after neoadjuvant therapy in breast cancer. Public transcriptomic analysis provided exploratory molecular context for treatment-response heterogeneity. External validation in larger multicenter cohorts is warranted.