Abstract:

Background and purpose: Breast cancer is a major global public health problem. Bone is the most common site of distant metastasis of breast cancer, accounting for about 70% of all metastatic cases. Bone metastasis of breast cancer can cause a series of complications, including severe pain, pathological fracture, hypercalcemia, spinal cord compression, etc., which bring great inconvenience to patients' physical activities and affect their quality of life. Metastatic recurrence is the leading cause of death in breast cancer patients. Therefore, there is an urgent need to build a diagnostic model of bone metastasis in breast cancer to identify patients with a high risk of bone metastasis. The aim of this study was to develop a predictive model based on machine learning to predict the probability of breast cancer developing bone metastasis. Methods: Data of breast cancer patients diagnosed between 2010 and 2015 were extracted from The Surveillance, Epidemiology, and End Results (SEER) database. The variables were screened by least absolute shrinkage and selection operator (LASSO) regression, univariate and multivariate logistic regression analysis, and statistically significant risk factors were included to build a prediction model. In this study, nine machine learning algorithms, including decision tree, elastic network, K-nearest neighbor, lightweight gradient elevator, logistic regression, neural network, random forest, support vector machine and limit gradient lifting, were used to adjust the model hyperparameters through random search and 5x cross-validation to build a breast cancer bone metastasis prediction model. The area under the receiver operating characteristic (ROC) curve, calibration curve and decision curve were used to evaluate the model, the optimal model was obtained, and the importance of variables was analyzed based on the optimal model. Finally, a network calculator for predicting the risk of bone metastasis of breast cancer was established using the optimal model. Results: The study included 10 106 patients with breast cancer, 7 073 patients in the training set, and 3 033 patients in the validation set. We found that 4 494 (63.5%) patients in the training set and 1 927 (63.5%) patients in the validation set developed bone metastases, respectively. Race, pathologic grade, estrogen receptor (ER) status, progesterone receptor (PR) status, human epidermal growth factor receptor 2 (HER2) status, N stage, lung metastasis, radiotherapy, chemotherapy and surgery were independent predictors of bone metastasis. The training set and verification set were used to verify the model, and the limit gradient lifting algorithm was superior to other machine learning algorithms by integrating the evaluation indexes such as the area under the ROC curve, calibration curve and decision curve. Finally, we used limit gradient algorithm to build network calculator for prediction of breast cancer bone metastases (https://bcbm.shinyapps.io/DynNomapp/). Conclusion: This study developed a predictive model based on machine learning to predict the probability of bone metastases in breast cancer patients, hoping to help clinicians make more rational treatment decisions.

Key words: Breast cancer, Bone metastasis, Prediction model, Machine learning, Network calculator