A Two-Step Machine Learning Model for Stage-Specific Disease Survivability Prediction

While traditional medical informatics focus primarily on disease classification problems, the disease survivability prediction for patients suffering from multi-stage conditions (e.g., congestive cardiac disorders, cancer types, diabetes, chronic kideny disorder, and so forth) surprisingly remains as an overlooked research topic. In this paper, we address this topic, and among the numerous multi-stage chronic diseases, we select the breast cancer use-case due to the importance of breast cancer patients survivability analysis and prediction for healthcare providers to make informed decisions on recommended treatment pathways for different patients. Then, we combine two main strategies in solving the breast cancer survivability prediction problem using Machine Learning techniques. In the first strategy, we model the survivability prediction task as a two-step problem, namely 1) a classification problem to predict whether or not a patient survives for five years, and 2) a regression problem to forecast the number of remaining months for those who are predicted to not survive for five years. The second strategy is to develop stage-specific models, where each model is trained on instances belonging to a certain cancer stage, instead of using all stages together, in order to predict survivability of patients from the same stage. We investigate the impact of adapting these strategies along with applying different balancing techniques over the model performance using the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) dataset. The obtained results demonstrate that the proposed methods prove effective in both survivability classification and regression.