Abstract
Sepsis is a life-threatening condition that presents substantial challenges to healthcare and pharmacological management due to its high mortality rates and complex patient responses. Accurately predicting patient outcomes is essential for optimizing therapeutic interventions and improving clinical decision-making. This study evaluates the predictive performance of the Cox proportional hazards model and advanced machine learning techniques, such as extreme gradient boosting (XGBoost), gradient boosting machine (GBM), and random survival forests (RSF), in forecasting survival outcomes for sepsis patients. Feature selection methods, including adaptive elastic net (AEN), smoothly clipped absolute deviation (SCAD), minimax concave penalty (MCP), and information gain (IG), were employed to refine model performance by identifying the most relevant clinical features. The results demonstrate that XGBoost consistently outperforms the Cox model, achieving a higher concordance index and demonstrating superior accuracy in handling complex, non-linear clinical interactions. The integration of feature selection further enhanced the machine learning models' predictive capabilities. These findings emphasize the potential of machine learning techniques to improve outcome prediction and guide personalized treatment strategies, offering valuable tools for critical care settings.