Early Glycemic Variability from Routine Glucose Measurements, Diabetes Status, and In-Hospital Mortality in Sepsis: A 14-Year Retrospective Cohort Study

Background/Objectives: The prognostic meaning of early glycemic variability in sepsis may depend on whether the patient has diabetes. We examined the association of pre-existing diabetes with in-hospital mortality and tested whether diabetes modifies the prognostic association of early glycemic variability estimated from routine glucose measurements. Methods: This study uses a single-center retrospective cohort of 10,832 adults with sepsis identified from discharge documentation at a Romanian tertiary hospital (2011–2024). Early glycemic variability was the coefficient of variation (CV) of glucose within 24 h of the first measurement, computable in 6450 patients (59.5%). Logistic models using diabetes-by-glycemic-parameter interactions were complemented by stratified, landmark and weighted analyses and by standardized absolute risks. Results: In-hospital mortality was 53.6%. Diabetes was associated with lower adjusted odds of death (odds ratio [OR] of 0.822; 95% confidence interval [CI] of 0.751–0.899), corresponding to a standardized mortality of 50.5% vs. 54.8% (risk difference, −4.3 percentage points); the difference was smaller among patients alive 24 h after the first measurement (−2.8 percentage points). Per 0.10 increase in CV-24 h, the OR for death was 1.137 (1.102–1.173) without and 1.034 (0.994–1.076) with diabetes (interaction p < 0.001); diabetes also modified the associations of mean glucose and hyperglycemia. The CV interaction persisted after weighting and among 24 h survivors but was attenuated when deaths within 48–72 h were excluded; CV added little to the discrimination. Conclusions: In this observational cohort, the same degree of early glycemic variability carried different prognostic meaning, mainly for short-term mortality, in patients with and without diabetes. Diabetes was associated with modestly lower observed mortality, which should not be read as a protective effect. These are prognostic associations, not independent effects, and they do not define glucose targets.