A Hybrid Approach for the Prediction of Breast Cancer Recurrence and Metastasis

Abstract

Today Breast Cancer has become one of the most common reasons for mortalities due to cancer in the women worldwide. Even with the progress in the latest medical technology starting from the diagnostic methods to the prognosis, managing breast cancer is still a challenge due to the increase in the risk of the relapse of cancerous cells and its spread in the body. The objective of the research is to propose a hybrid model by implementing feature subset selection using Chi-square test and leave-one-out techniques and then applying various statistical predictive and ensemble learning techniques to predict the chances of relapse and metastasis in those patients who had the breast cancer as their primary cause. The SEER database covering years 2017-2021 was used as the dataset in this research that included extensive data on patient profile statistics, tumor attributes and its features, treatment therapies and clinical methods and progression status of metastasis, with a total of 79 features. In the pre-processing step, the dataset was reduced to 24 attributes with 105,404 instances, which were the crucial risk factors that could lead the cancerous cells to re-occur at the primary site or spread in the body. In-depth analysis was done on a sample size of 5387 records where the patients died due to breast cancer. Statistical analysis showed that around 30% of the patients died having age greater than 85 years and above and HR+/HER2- was found to be the most common breast cancer subtype. Many of the patients were found with a single tumor (2,143 cases). Multiple statistical and predictive models including decision tree, random forest, support vector machine, logistic regression, AdaBoost and XGBoost were trained and evaluated to predict recurrence and metastasis. This work highlights the need for developing comprehensive and usable hybrid predictive models to support clinical medical decisions. The primary contribution of this work lies in the fact that there is no research on this specific pipeline configuration to address Breast Cancer Recurrence and Metastasis in Lungs having been applied together in this way to a SEER cohort of this recency and size (105,404 records, 2017–2021), and implementing these machine learning models into a unified and clinically robust and reproducible prediction framework specifically designed for breast cancer recurrence and lung metastasis prediction using the SEER database