Causally-informed deep learning towards explainable and generalizable outcome prediction in critical care

The rapid evolution of deep learning is revolutionizing critical care medicine by enabling highly accurate early warning systems. These systems are capable of predicting diverse clinical deteriorations, including acute kidney injury, myocardial infarction, and circulatory failure. While demonstrating strong predictive capabilities, conventional deep learning models face two fundamental limitations in clinical implementation: their opaque decision-making processes and limited generalizability across diverse patient populations and care environments. To address these challenges, we present a causally-informed deep learning framework that integrates causal discovery to jointly identify the causal drivers of clinical outcomes alongside prediction. This approach yields two unique advantages: demonstrating the explicit interpretation of the prediction while exhibiting decent performance when applied to unfamiliar environments. Benefiting from these features, our approach achieves superior accuracy across six different critical deteriorations and also demonstrates improved generalizability across diverse patient groups compared to various baseline algorithms. Besides, we provide explicit causal pathways to serve as references for assistant clinical diagnosis and potential interventions. By incorporating causal reasoning into clinical prediction, our approach can enhance clinical decision support through both reliable forecasting and transparent physiological interpretation, paving the way for more trustworthy AI implementation in critical care.