Towards a Digital Twin for Critical Care Through Explainable Digital Human Modeling of ICU Patients

Early and reliable mortality prediction in the Intensive Care Unit (ICU) remains a significant challenge due to diverse disease progressions and continuously changing patient conditions. Traditional clinical scoring systems (Acute Physiology and Chronic Health Evaluation (APACHE II), and Simplified Acute Physiology Score II (SAPS II)) rely on predefined rules and static measurements. In addition, many existing data-driven mortality prediction models use only the first 24 h of ICU data and provide limited interpretability, reducing their usefulness in clinical decision-making. To address these challenges, a unified Digital Human (DHM) and Digital Twin (DTw) framework is proposed for a disease specific ICU mortality prediction and early warning system. Within the DHM component, tree-based Machine Learning (ML) models (XGBoost, Random Forest, and LightGBM) were trained and evaluated on preprocessed disease specific ICU data, with model explainability enhanced through SHAP-based feature analysis. The proposed DTw framework provides a retrospective window-by-window digital twin simulation of an individual ICU patient that sequentially updates mortality risk estimates using only data available up to each time point, enabling explainable risk trajectory and early warning decision support.