Integrative analysis of mortality risk in SFTS using machine learning and genetic approaches

This study presents a novel approach to predicting mortality risk in patients with severe fever with thrombocytopenia syndrome (SFTS) by integrating machine learning models and genetic analysis techniques. Utilizing clinical data from 107 confirmed cases, we developed several predictive models and identified the extreme gradient boosting (XGBoost) algorithm as the most effective, achieving an area under the curve (AUC) of 0.950. Further analysis was performed based on the 15 robust features selected and performance-optimized through the Boruta algorithm, integrating Shapley additive explanations (SHAP) plots to identify key variables such as IL-6, TNF-α, PLR, and AST. While Mendelian randomization did not reveal definitive causal pathways across all examined factors, notable associations were observed in specific metabolic-inflammatory pathways. These findings offer a foundation for enhanced predictive accuracy and mechanistic understanding of SFTS mortality.IMPORTANCESevere fever with thrombocytopenia syndrome (SFTS) is a tick-borne viral disease with high mortality, posing a major threat to public health in East Asia. This study develops a machine learning (ML) model to quickly predict SFTS patients' mortality risk using routine clinical tests, helping doctors stratify risks early. Moreover, by exploring causal pathways between bodily indicators and disease outcomes, it provides new clues for understanding why SFTS becomes severe, which could guide better treatments in the future.