Prediction of Breast Cancer Metastasis Via Decision Tree Modeling

Abstract

Background: Nowadays, breast cancer (BC) metastasis is a nightmare for women and one of the main challenges among researchers worldwide. Unlike traditional statistical methods that are not able to handle and take into account the complexity of effects and existence of interactions among predictor variables, the decision trees can overcome these problems. This study aimed to predict and identify the main prognostic factors of BC metastasis status (binary response) using decision tree modeling.Methods: This retrospective cohort study was conducted on 375 patients with BC who had registered with the Comprehensive Cancer Control Center from 1998 to 2013. Some demographic features related to the conditions of the Person’s disease and the type of treatment received were recorded. We applied a tree-based approach using the Gini index as the homogeneity criterion to explore the factors affecting metastasis occurrence in BC patients.Results: The mean (SD) age of BC patients with and without metastasis was 55.7 (12.4) and 43.1 (7.2) years, respectively (P<0.001). The rate of metastasis was 33.3%. The five most important risk factors for metastasis of tumor proposed by tree diagram were age at diagnosis, grade of tumor, type of surgery, number of deliveries, and axillary surgery. The prediction accuracy of the proposed model was 84.3%, and its sensitivity and specificity were 66.4% and 93.2%, respectively.Conclusion: Age at diagnosis was the most important factor for predicting breast cancer metastasis, so that breast cancer patients aged over 54 were at high risk of metastasis.

Keywords

• breast neoplasms
• neoplasm metastasis
• decision trees
• machine learning
• risk factors