The critical effects of self-management strategies on predicting cancer survivors' future quality of life and health status using machine learning techniques.

Despite the significance of enhancing the quality of life (QoL) and overall health status (including physical, mental, social, and spiritual well-being) among individuals who have survived cancer, the existing prediction model for QoL and health status lacks sufficient interpretation. Our primary objectives were to develop and validate simple prediction models for QoL and secondary health statuses. Additionally, we aimed to interpret these prediction models using explainable artificial intelligence (XAI) methods, including extracting important features and creating dependence plots. Lastly, we sought to predict and interpret individual outcomes, visualizing the results using the XAI technique known as SHapley Additive explanation (SHAP). In this prospective cohort study, conducted through a web-based survey, we established prediction models for QoL and health statuses, comparing their performance with ensemble methods, including decision trees, random forest, gradient boosting, eXtreme Gradient Boost (XGBoost), and LightGBM. Following the model comparison, we selected the XGBoost model for further analysis. We identified crucial features associated with QoL and each health status separately and leveraged SHAP to extract individual prediction results from the XGBoost model. After preprocessing the data and selecting the appropriate model, our final dataset consisted of 256 cancer survivors with 42 predictive features. Repeated stratified K-fold validation demonstrated high performance of the XGBoost predictive model for QoL. Similarly, the XGBoost predictive model exhibited good performance for each health status, including mental, social, and spiritual well-being. The important features identified in these predictive models varied based on the specific health outcomes. This study represents the first endeavor to develop and validate predictive models for QoL and health status among cancer survivors while also providing interpretations of these models.