Understanding Concussion Knowledge in Urban Community Settings: Perspectives of School Nurses and Youth Athletic Program Leaders.

Sepsis, a systemic inflammatory response syndrome triggered by infection, is a major cause of multiple organ dysfunction syndrome and septic shock, with both high incidence and mortality rates in intensive care units (ICUs) (). Most previous estimates of sepsis incidence and mortality have relied on hospital administrative databases, excluding patients who were never admitted to the hospital (). In addition to its substantial health burden, septic shock is one of the most costly medical conditions to treat, with an estimated annual healthcare cost of 24 billion USD (). The pathophysiology of sepsis is complex, involving a multidimensional cascade including immune system overactivation, endothelial damage, mitochondrial dysfunction, and coagulation disorders (). The disease progression is highly heterogeneous, with patients potentially deteriorating from a localized infection to multiorgan failure within hours, making early clinical identification and dynamic risk assessment extremely challenging (). Conventional scoring systems, including the Sequential Organ Failure Assessment (SOFA) () and the Acute Physiology and Chronic Health Evaluation II (), are commonly applied to evaluate prognosis and risk factors in clinical practice. However, the limitations of these systems, which depend on static cutoffs and linear modeling, have become increasingly apparent. For instance, the SOFA score is based on snapshot physiological and laboratory measures (such as lactate levels and platelet counts), failing to track ongoing physiological trends. As a result, the SOFA score lacks sufficient sensitivity for real-time surveillance and early warning in sepsis management (). Johnson et al. pointed out the limitations of traditional scoring systems (such as Acute Physiology and Chronic Health Evaluation II) in capturing the nonlinear feedback between inflammatory mediators and organ function in sepsis. They emphasized the advantages of machine learning (ML) methods in addressing complex pathophysiological mechanisms (). Esteva and colleagues emphasized the importance of imaging techniques (such as computed tomography texture analysis) and genomic data in disease risk prediction. They pointed out that traditional scoring systems may lead to missing critical information because they rely primarily on structured data like lab tests and vital signs (). With the advancement of medical informatization, electronic health records and wearable devices have provided a data foundation for the application of ML (,). Research indicates that ML models can leverage time-series physiological signals (such as heart rate variability and blood pressure waveforms) and multimodal inputs (like computed tomography texture features and metabolomic biomarkers). They have been shown to markedly enhance the precision of early sepsis recognition and mortality risk prediction (,). However, when studies prioritize performance too heavily, they may produce overly complex models—such as deep neural networks—that operate as “black boxes.” This lack of interpretability makes it difficult for clinicians to understand the prediction logic, severely limiting real-world clinical adoption (,). Therefore, there is a critical need to construct a sepsis mortality risk prediction model that integrates high predictive accuracy with robust interpretability to address the specific requirements of clinical practice (). According to the Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3), sepsis is now defined as “a life-threatening organ dysfunction resulting from a dysregulated host response to infection” (). The pathophysiology of sepsis is believed to begin with a prolonged hyperinflammatory phase lasting several days, followed by an anti-inflammatory and immunosuppressive state. The inflammation may progress to shock and multiple organ failure, resulting in severe immune dysfunction and endocrine-metabolic disturbances. Patients with severe infections have a high in-hospital mortality rate, and even those who survive often experience poor outcomes (). Against this backdrop, this study aims to develop a mortality risk prediction model for sepsis patients using tree-based ensemble classifiers with post hoc interpretation facilitated by Shapley Additive Explanations (SHAP). The data in this study were obtained from clinical records of sepsis patients admitted to the ICU of the First Affiliated Hospital of Xinjiang Medical University between January 2015 and May 2024. The dataset includes demographic information, vital signs, laboratory test results, and other clinical parameters. Inclusion criteria: patients with complete and accurate personal and clinical information who met the diagnostic criteria. Exclusion criteria: patients with other serious diseases such as severe heart disease, chronic renal failure, or liver cirrhosis; pregnant or lactating women; and individuals with incomplete required data. Specifically, variables contributing li