A Model to Predict Breast Cancer Survivability Using Logistic Regression

Background: Breast cancer is the most common type of cancer amongst women worldwide. Considering its high incidence, effective detection and prognosis of this type of cancer may have a significant effect on reducing expenditures. In this study, we propose a model to predict the ۶۰-month survivability in patients with breast cancer and investigate the effects of each feature on the obtained model. Methods: We base this model on the information gathered by the Breast Disease Research Center, Shiraz University of Medical Sciences, Shiraz, Iran from ۵۶۷۳ patients with breast cancer. The goal of this study was to predict breast cancer survivability at early diagnosis, so the features used in the research are among those considered affordable, specifically at the initial steps of diagnosis. After preprocessing all of the cases and features, we constructed this model based on ۱۹۳۰ cases and ۱۶ of their associated features using logistic regression method. The model then was evaluated with ۱۰-fold cross validation. Results: Based on all subsets of the ۱۶ features, we evaluated numerous models. We selected a model that achieved the best sensitivity and specificity, and used fewer features as the best model. We considered this model for further analysis, which is consisted of following features: age at the time of diagnosis, type of invasion, HER۲, size of the tumor, in situ component, lymph node involvement ratio, progesterone receptor status, and the total number of dissected lymph nodes. The best model obtained overall accuracy, specificity and sensitivity of ۷۲.۴۹%, ۷۲.۸۳%, and ۷۱.۸۵%, respectively. Conclusion: The performance of model is quite satisfactory due to the fact that we only used features, which could be obtained at the initial steps of diagnosis. Even though, the effect of patient’s age is controversial, we concluded that ageing would decrease the ۶۰-month survivability. Our model indicated that having all type of invasions (i.e. vascular, lymphatic, etc.) would result in poorer chance of survival compared to other features effect.

Keywords:

Statistical learning , Data mining , Breast cancer , Logistic regression , Survivability