Identifying early blood glucose trajectories in sepsis linked to distinct long-term outcomes: a K-means clustering study with external validation.

Background: Blood glucose (BG) dysregulation, including hyperglycemia, hypoglycemia and increased glycemic variability (GV), is common in septic patients and potentially associated with poor clinical outcomes. However, the prognostic value of early BG trajectories remains unclear. We intend to investigate the association between the early dynamic trajectory of BG and 1-year mortality among sepsis patients. Methods: This retrospective study comprises a derivation cohort of sepsis patients admitted to the First Affiliated Hospital of Sun Yat-sen University (FAH-SYSU) from January 2018 to December 2023, and an external validation cohort of 10,874 sepsis patients from the Medical Information Mart for Intensive Care (MIMIC) IV database. Distinct clusters were demarcated using K-means clustering based on the BG trajectory within the first 48 hours after ICU admission, while the optimal number of clusters was determined by a consensus of quantitative metrics and the elbow plot. Kaplan-Meier survival curves and multivariable Cox proportional hazards regression models were used to assess the association between these identified clusters and 1-year mortality. Results: Among 3,655 sepsis patients from the FAH-SYSU dataset, we identified 5 distinct clusters of BG trajectories, which were significantly associated with 1-year mortality risk. In the full Cox regression model, patients with "low-stable" and "moderate-stable" trajectories had the lowest 1-year mortality risk ( Conclusion: Early BG trajectory patterns are independently associated with long-term mortality in sepsis patients. Incorporating dynamic BG measurements into clinical practice may improve risk stratification and guide individualized glucose management strategies.