Prognostic Models Using the Lactate-to-Albumin Ratio for 28-Day Mortality in ICU Sepsis Patients: A Bayesian Model Averaging Approach

Abstract

Purpose: The lactate-to-albumin ratio (LAR) has recently emerged as a significant predictor of 28-day mortality in sepsis. This study aims to determine the prognostic value of LAR for 28-day mortality and identify prognostic models for sepsis patients. Patients and Methods: A prospective study was conducted on patients admitted to the Intensive Care Unit at Thong Nhat Hospital, Vietnam. The study included patients aged 16 years and older diagnosed with sepsis according to the Sepsis-3 consensus guidelines. The primary outcome was all-cause mortality within 28 days from the time of sepsis diagnosis. Multivariable logistic regression estimated odds ratios (ORs). Bayesian model averaging (BMA) was used to identify candidate models. Discrimination was evaluated by the area under the curve (AUC), and calibration was assessed. Results: This study included 170 participants with a median age of 73 years and a male predominance (54.1%). The overall 28-day mortality rate was 57.6%. The median LAR was 2.2, with a statistically significant difference between the survival and mortality groups (p < 0.001). Multivariate logistic regression analysis revealed that LAR was independently associated with 28-day mortality (OR, 2.74; 95% CI, 1.09-6.86; p = 0.017). The AUC for LAR was 0.81 (95% CI 0.75-0.87; p < 0.001) with a cutoff point of 1.2 (sensitivity 91%, specificity 57%). BMA identified three clinically applicable models: LAR combined with age and respiratory infection (AUC 0.855), LAR combined with respiratory infection (AUC 0.839), and LAR combined with age and respiratory infection and SOFA score (AUC 0.864). Internal validation also represented the stability and reproducibility of these models with AUC 0.848, 0.836, and 0.854, respectively. Conclusion: In this single-center cohort, higher LAR was independently associated with 28-day mortality, and BMA identified simple LAR-based models with good internal discrimination and calibration. These findings require further external validation before routine clinical implementation.