[Construction and validation of a nomogram prediction model for brain metastasis in breast cancer]

Objective: To construct a prediction model integrating molecular markers, imaging features, and clinicopathological characteristics for brain metastasis (BM) of breast cancer and provide a quantitative tool for clinical precise stratification of breast cancer. Methods: A total of 346 breast cancer patients admitted to the Affiliated Tumor Hospital of Xinjiang Medical University from January 2018 to December 2023 were enrolled, including 214 cases in the BM group and 132 cases in the control group. The patients were divided into a training set (242 cases) and an internal validation set (104 cases) at a ratio of 7∶3. A total of 1,483 female breast cancer patients from the Surveillance, Epidemiology, and End Results Program (SEER) database (2017-2021) were used as the external validation set, including 530 cases in the BM group and 953 cases in the control group. Least Absolute Shrinkage and Selection Operator (LASSO) regression was used to screen key features, and a nomogram prediction model for breast cancer BM was constructed based on the results of multivariate logistic regression analysis of factors influencing breast cancer BM. The receiver operating characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA) were used to evaluate the model performance, and internal and external validations of the model were conducted. Results: LASSO regression based on the training set screened out 11 key features, namely, primary side, primary quadrant, number of lesions, histological grade, differentiation degree, Ki-67 index, N stage, clinical stage, lymphovascular invasion, aspect ratio, and lesion margin morphology. Multivariate logistic regression analysis showed that the number of lesions (multiple lesions: OR=7.49, 95% CI: 3.59-15.64), histological grade (grade 2: OR=9.02, 95% CI: 4.87-16.70; grade 3: OR=12.57, 95% CI: 6.60-23.94), differentiation degree (moderate differentiation: OR=0.18, 95% CI: 0.06-0.53; well differentiation: OR=0.02, 95% CI: 0.01-0.06), Ki-67 index (15%-29%: OR=0.28, 95% CI: 0.11-0.74; ≥30%: OR=4.01, 95% CI: 2.45-6.58), clinical stage (stageⅢ: OR=3.53, 95% CI: 1.76-7.07; stage Ⅳ: OR=35.43, 95% CI: 13.97-89.91), and lymphovascular invasion (presence of lymphovascular invasion: OR=12.42, 95% CI: 6.84-22.53) were independent influencing factors for breast cancer BM. The nomogram prediction model for breast cancer BM constructed based on the results of multivariate logistic regression analysis had areas under the curve (AUCs) for predicting breast cancer BM of 0.869 (95% CI: 0.826-0.912), 0.857 (95% CI: 0.817-0.897), and 0.868 (95% CI: 0.849-0.887) in the training set, internal validation set, and external validation set, respectively. All calibration curves were close to the reference line, and when the threshold probability was >0.2, the net benefit was significantly higher than that of the "treat all" or "treat none" strategy. Conclusions: A nomogram prediction model for breast cancer BM integrating multi-dimensional information (molecular subtypes, imaging features, and clinicopathological characteristics) was successfully constructed. This model has good ability to distinguish breast cancer patients with BM from thosewithout, with favorable calibration and clinical applicability, and provides a quantitative tool for clinical assessment of the risk of breast cancer BM.