Survival-Based Treatment Planning Using Stage-Specific Machine Learning Models

The significance of prognostic survivability in determining optimal treatment strategies for critical illnesses is widely acknowledged. However, there has been a lack of emphasis on the advancement of treatment planning models based on survival outcomes within clinical decision support systems. The research presented in this paper proposes an innovative framework for the planning of treatment strategies based on survival outcomes in the context of multi-stage diseases, with the aim of effectively tackling this issue. Our proposed system aims to predict a comprehensive list of treatment combinations for cancer patients, specifically focusing on their expected survival outcomes. The proposed solution aims to enhance the decision-making process of medical professionals by providing them with comprehensive and comprehensible treatment recommendations. To conduct survivability classification and regression analysis for patients with identical cancer stages, a two-step approach is employed. This involves the development of stage-specific Machine Learning models using breast cancer data that includes treatment information. Based on a real dataset on cancer patients, we aim to investigate the performance of the models under different balancing strategies. Our contribution in this work is the formulation of a treatment planning inference system, which focuses on prognostic considerations. This system utilizes patient data and estimates the survivability associated with each treatment plan in order to predict the recommended course of action. This facilitates the integration of the developed survival prediction models into the process of treatment planning. Ultimately, the system generates visual representations that illustrate the comparative significance of different features, as well as the decision-making process employed by the model in order to yield easily comprehensible outcomes for a specific patient. The study presents experimental findings that illustrate the efficacy of our proposed framework in the domains of treatment planning and survival estimation.