Association of lactate-to-albumin ratio with 28-day mortality in elderly sepsis patients: a retrospective MIMIC-IV database analysis

BACKGROUND: The lactate-to-albumin ratio (LAR) is correlated with mortality in critically ill patients; however, its predictive value for sepsis in the elderly remains underexplored. METHODS: This retrospective cohort study used the data from the MIMIC-IV database. The primary outcome was 28-day ICU mortality in elderly patients with sepsis. We used multivariate logistic regression, restricted cubic splines (RCS), subgroup analyses, and machine learning (LightGBM/XGBoost/SVM) to assess the predictive capacity of LAR. RESULTS: Among 5,115 eligible patients, elevated log (LAR) independently predicted mortality, as a continuous variable (adjusted odds ratio [OR] = 3.29; 95% confidence interval [CI]: 2.51-4.31; p < 0.001) and in quartiles (Q4 vs. Q1: adjusted OR = 2.11; 95% CI: 1.72-2.59). The RCS indicated a linear LAR-mortality association (p for nonlinear = 0.12). Subgroup analyses confirmed consistency across strata (interaction, p > 0.05). The LightGBM model achieved optimal discriminative performance (area under the curve = 0.788), with log (LAR) as the second-most important feature after the Simplified Acute Physiology Score II. CONCLUSION: LAR is an independent predictor of 28-day mortality in elderly patients with sepsis, despite its limited standalone use. Our LightGBM-based integrative model demonstrated a robust prognostic performance and may aid in clinical decision-making and pending external validation. CLINICAL TRAIL NUMBER: Not applicable.