Leukocyte glycemic index associated with 30-day mortality in sepsis patients with new-onset atrial fibrillation: insights from the MIMIC-IV database

This study aimed to examine whether the leukocyte glycemic index (LGI), a compact marker linking inflammatory burden with stress-related glucose disturbance, is associated with 30-day mortality among patients with sepsis complicated by new-onset atrial fibrillation (NOAF). This retrospective cohort study used MIMIC-IV data. LGI was derived from the first available white blood cell count and glucose measurement within 24 hours after ICU admission; glucose was converted from mg/dL to mmol/L before multiplication, and LGI was then log-transformed using the natural logarithm. We evaluated LGI with tertile-based Cox models, restricted cubic splines, threshold-effect analyses, exploratory subgroup analyses, and prediction-performance analyses. The cohort included 691 patients, of whom 315 (45.59%) died within 30 days. The median age was 72.39 years, and 418 patients (60.49%) were male. The log(LGI) tertile cut points were 4.284 and 4.861. In the fully adjusted tertile analysis, T3 showed a borderline higher risk than T2 (HR 1.33, 95% CI 1.00-1.77, p=0.052), while T1 did not differ significantly from T2. A single linear log(LGI) term did not carry a clear adjusted association with mortality, whereas spline and threshold models supported a nonlinear association. The estimated inflection point was log(LGI) = 4.403. Below this value, each 1-unit increase in log(LGI) was associated with lower mortality risk (HR 0.76, 95% CI 0.64-0.91), indicating that very low log(LGI) values identified a higher-risk subset. Above the inflection point, each 1-unit increase in log(LGI) was associated with higher mortality risk (HR 1.49, 95% CI 1.20-1.85). Adding spline-transformed log(LGI) to a clinical model modestly improved discrimination (AUC 0.715 to 0.732; DeLong p=0.023). In patients with sepsis and NOAF, LGI showed a nonlinear association with 30-day mortality, with increased risk at both lower and higher ranges, more pronounced at the high extreme. LGI may add a small but measurable layer of prognostic information to routinely available clinical variables, but it should be interpreted as an adjunctive, hypothesis-generating marker rather than a stand-alone risk tool.