Abstract Rationale Sepsis is a life-threatening organ dysfunction shaped by both macro-level epidemiological and molecular-level proteomic factors, while the contributions of these factors and underlying molecular mechanisms of sepsis remain poorly understood. Methods In this study, we analyzed data from 472,311 UK Biobank participants, including 49,872 with proteomic data. Sepsis was classified into explicit and implicit types in this population-based study. We examined associations between new-onset sepsis and six modifiable risk factors, estimating corresponding population attributable fractions. Among participants with proteomic data, proteins associated with sepsis subtypes were analyzed using sex-stratified Cox regression models, followed by pathways enrichment analyses. A Light Gradient Boosting Machine with forward feature selection was used to identify and rank proteins on the shared pathways, irrespective of sex or sepsis subtype. Model performance was evaluated by area under the receiver operating characteristic curve (AUC), and SHAP values were used for feature interpretation. Results During a median 14.5-year follow-up, the explicit sepsis incidence rate was 3.95 cases per 1,000 person-years in men and 2.47 cases per 1,000 person-years in women. The corresponding incidence rates of implicit sepsis were 5.97 cases per 1,000 person-years and 5.37 cases per 1,000 person-years. Lower education was the strongest contributor to sepsis, followed by smoking. We identified both shared and sex- or type-specific molecular pathways associated with sepsis. Among these, 12 pathways and 133 proteins that appear independent of sex or sepsis type. Pathway enrichment analysis highlighted the role of neutrophil degranulation. A model combining six proteins (GDF15, EDA2R, WFDC2, TNFRSF10B, IL6, ACTA2) and six modifiable risk factors achieved an AUC of 0.703 for sepsis prediction. Conclusion Our findings underscore the importance of incorporating both macro- and micro-level determinants in evaluating sepsis risk. This framework could guide the development of tailored pharmacological treatments and lifestyle modification strategies, thereby enhancing the effectiveness of clinical management. This abstract is funded by: The study was granted by Noncommunicable Chronic Diseases-National Science and Technology Major Project (2023ZD0506505), Postdoctoral Fellowship Program of CPSF (GZC20250498); National High Level Hospital Clinical Research Funding (2025-PUMCH-A-133)
Cao et al. (Fri,) studied this question.