Establishment and validation of an ultrasound-based nomogram with risk stratification for short disease-free survival in breast cancer.

ABSTRACT

PURPOSE: This retrospective study aimed to develop and validate an Ultrasound (US)-based nomogram to predict short disease-free survival (short-DFS, less than 120 months DFS) in breast cancer (BC). METHODS: Nomogram was established based on a training data of 311 BC patients by multivariable logistic regression, and were assessed by discrimination, calibration, and clinical usefulness. Risk stratification was performed by X-tile. An independent testing data of 200 patients with BC was used for external validation. RESULTS: Nine predictors including three US features and six clinical parameters were screened into the nomogram by Lasso (log λ = -3.594) in training data. Better performance was obtained in the training data (C-index 0.942) and testing data (C-index 0.914). Calibration analysis indicated optimal agreement between nomogram predictions and actual observations (p = 0.67). Decision curve analysis showed a great clinical benefit (Youden index 0.634). Three risk levels are low-risk (<184.0), moderate-risk (184.0-345.3) and high-risk (>345.3). Our nomograms had larger area under the receiver operating characteristic (ROC) curves compared with Magee Equation and Nottingham Prognostic models (0.942 vs. 0.824, 0.790). CONCLUSION: The US-based nomogram and the practical score system facilitate individualized prediction of short-DFS to optimize clinical decisions and improve prognosis in patients with BC.