A visualized model for identifying optimal candidates for aggressive locoregional surgical treatment in patients with bone metastases from breast cancer.

Abstract

The impact of surgical resection of primary (PTR) on the survival of breast cancer (BC) patients with bone metastasis (BM) has been preliminarily investigated, but it remains unclear which patients are suitable for this procedure. Finally, this study aims to develop a predictive model to screen BC patients with BM who would benefit from local surgery. BC patients with BM were identified using the Surveillance, Epidemiology, and End Results (SEER) database (2010 and 2015), and 39 patients were obtained for external validation from an Asian medical center. According to the status of local surgery, patients were divided into Surgery and Non-surgery groups. Propensity score matching (PSM) analysis was performed to reduce selection bias. Kaplan-Meier (K-M) survival and Cox regression analyses were conducted before and after PSM to study the survival difference between the two groups. The survival outcome and treatment modality were also investigated in patients with different metastatic patterns. The logistic regression analyses were utilized to determine significant surgery-benefit-related predictors, develop a screening nomogram and its online version, and quantify the beneficial probability of local surgery for BC patients with BM. Receiver operating characteristic (ROC) curves, the area under the curves (AUC), and calibration curves were plotted to evaluate the predictive performance and calibration of this model, whereas decision curve analysis (DCA) was used to assess its clinical usefulness. This study included 5,625 eligible patients, of whom 2,133 (37.92%) received surgical resection of primary lesions. K-M survival analysis and Cox regression analysis demonstrated that local surgery was independently associated with better survival. Surgery provided significant survival benefits in most subgroups and metastatic patterns. After PSM, patients who received surgery had a longer survival time (OS: 46 months vs. 32 months, p < 0.001; CSS: 50 months vs. 34 months, p < 0.001). Logistic regression analysis determined six significant surgery-benefit-related variables: T stage, radiotherapy, race, liver metastasis, brain metastasis, and breast subtype. These factors were combined to establish the nomogram and a web probability calculator (https://sunshine1.shinyapps.io/DynNomapp/), with an AUC of 0.673 in the training cohort and an AUC of 0.640 in the validation cohort. The calibration curves exhibited excellent agreement. DCA indicated that the nomogram was clinically useful. Based on this model, surgery patients were assigned into two subsets: estimated sur-non-benefit and estimated sur-benefit. Patients in the estimated sur-benefit subset were associated with longer survival (median OS: 64 months vs. 33 months, P < 0.001). Besides, there was no difference in survival between the estimated sur-non-benefit subset and the non-surgery group. Our study further confirmed the significance of local surgery in BC patients with BM and proposed a novel tool to identify optimal surgical candidates.