Establishment and verification of a nomogram to predict tumor-specific mortality risk in triple-negative breast cancer: a competing risk model based on the SEER cohort study.

Background: Triple-negative breast cancer (TNBC) is the subtype of breast cancer with the worst prognosis, and traditional survival analysis methods are biased when predicting mortality. To predict the risk of death in patients with triple-negative breast cancer more precisely, a competing risk model was developed. Methods: The clinicopathological data of the TNBC patients from 2010 to 2015 were collected from the Surveillance, Epidemiology, and End Results (SEER) database. The data were assigned into a training set and testing set at a ratio of 7:3 in a randomized pattern. Univariate and multivariate competing risk models were applied to find the independent prognostic factors. A prediction nomogram for cancer-specific mortality (CSM) risk was constructed. The accuracy and discrimination of the nomogram were assessed using receiver operating characteristic (ROC) area under the curve (AUC), concordance index (C-index), and a calibration curve using the training and testing sets, respectively. Results: A total of 28,430 TNBC patients were randomly grouped into the training set (n=19,900) and the testing set (n=8,530). The median time for follow-up was 59 [1-107] months. A total of 7,014 (24.67%) patients died, among whom 4,801 (68.45%) died from breast cancer and 2,213 (31.55%) due to non-breast cancer events. The independent risk factors were primary site of tumor, grade, tumor-node-metastasis (TNM) stage, T stage, approach of surgery, chemotherapy, axillary lymph node metastases, brain metastases, and liver metastases. The prediction nomogram was constructed by using the aforementioned variables. The 1-, 3-, and 5-year AUC of CSM were predicted to be 0.856, 0.81, and 0.782, respectively, in the training set, and 0.856, 0.81, and 0.782 in the testing set, respectively. The C-index of the nomogram was 0.801 and 0.799 in the training and testing sets, respectively. As confirmed by the validation and training calibration curves, the nomogram was consistent with the results. Conclusions: We used competing risk models to identify risk factors for CSM and constructed a CSM risk prediction nomogram for TNBC patients, that may be utilized to predict CSM risk in TNBC patients clinically and assist in the creation of individualised clinical treatment options.