SUN-810 Glucose and Albumin Ratio Biomarkers Improve Sepsis Mortality Prediction: An Exploratory Machine Learning Approach

Abstract Disclosure: J. Fiorica: None. P. Riano: None. L. Forchette: None. J. Gonzalez: None. K. Hamad: None. Hyperglycemia and hypoalbuminemia are linked to mortality in patients with sepsis or septic shock. Hyperglycemia dysregulates immune system, evokes environment favorable to pathogen growth, and restrict blood flow contributing to impaired clearance of infection. Hypoalbuminemia can amplify negative effects of hyperglycemia escalating sepsis severity and mortality risk. While individual and predominantly cross-sectional roles of glucose and albumin have been investigated, synergistic interactions driving host response requires further study. Boosted Tree (BT) modeling was implemented to explore percent variance explained (VE%) in mortality by routinely available blood-based biomarker ratios containing serum glucose or albumin across ICU Day 1, 2, and 3. Under IRB exemption, demographics, laboratory results and administrative data were extracted from EMR in patients at least 18yrs old discharged after at least 48h of ICU censored after 28d between 1/1/2020 to 9/30/2024. Generalized regression with LASSO confirmed putative ratios association (p<.05) with mortality controlling for age, sex, comorbidities, and COVID-19 treatment. BT computed within sepsis severity respectively the area under receiver operating characteristic curve representing ensemble prediction accuracy (R2) and VE% contributed by each ratio. Continuous data summarized with median (IQR) were compared using nonparametric ANOVA. Discrete data summarized as proportions were compared with chi-square test. 1137 patients, with sepsis (n=377) vs. septic shock (n=760), were aged 73 years [63,81], female (43%), and white (87%). Comorbidities included hypertension (69%), neurological disorders (54%), deficiency anemias (52%), heart failure (40%), and diabetes (37%). Top three ratio biomarkers on ICU Day 1 were for (1) sepsis Alkaline Phosphatase/Albumin (VE%=21%), WBC/Glucose (VE%=16%), Lactate/Albumin (VE% = 15%) and (2) septic shock, Alkaline Phosphatase/Albumin (VE%=27%), Lactate/Albumin (VE%=16%), Lactate/Glucose (VE%=11%). Over subsequent two days, Lactate/Albumin became most important predictor on ICU Day 2, while Lactate/Glucose emerged as most important predictor on ICU Day 3. A similar trend was observed in septic shock, with Lactate/Albumin and Albumin/Globulin sharing top spot ICU Day 2, and Albumin/Globulin taking lead on ICU Day 3. These findings evince dynamic interplay between glucose, albumin, and other routinely measured biomarkers may provide valuable insights into pathophysiological mechanisms underlying sepsis and septic shock. Observed temporal changes in predictive importance of different biomarker ratios underscore need for serial monitoring and dynamic assessment of these biomarkers in clinical practice. Further research is warranted to elucidate specific mechanisms by which these biomarker ratios contribute to mortality risk in sepsis and septic shock. Presentation: Sunday, July 13, 2025