Abstract
PURPOSE: Sepsis is a life-threatening condition, and early detection remains a challenge. While bile acids have been implicated in various diseases, their role as biomarkers for sepsis is underexplored. This study aims to identify metabolites associated with sepsis and assess the potential of bile acids for early diagnosis and prognosis of pediatric sepsis. METHODS: We enrolled 100 participants in the discovery phase and 141 participants in the validation phase. Non-targeted liquid chromatography-mass spectrometry (LC-MS) metabolomics analyses were performed to identify differential metabolites between sepsis patients and healthy controls. Targeted quantitative analysis of 12 plasma bile acids (BA) was conducted to assess their concentrations. Machine learning algorithms, including Recursive Feature Elimination (RFE), Least Absolute Shrinkage and Selection Operator (LASSO), Random Forest (RF), Support Vector Machine-RFE (SVM-RFE), and Gradient Boosting Decision Tree (GBDT), were employed to identify key BAs for sepsis diagnosis. RESULTS: Untargeted metabolomics revealed bile acid pathway dysregulation, validated by targeted quantification showing elevated primary bile acids-cholic acid (CA), glycocholic acid (GCA), taurocholic acid (TCA), glycochenodeoxycholic acid (GCDCA), taurochenodeoxycholic acid (TCDCA)-and taurine-conjugated secondary bile acid TDCA (p<0.05). Conversely, secondary bile acids deoxycholic acid (DCA), glycodeoxycholic acid (GDCA), ursodeoxycholic acid (UDCA), glycoursodeoxycholic acid (GUDCA), and lithocholic acid (LCA) were significantly reduced (p<0.05). Univariate logistic regression identified TCA, TDCA, GCA, TCDCA, GCDCA, and TDCA/DCA as risk factors for sepsis, while LCA, DCA, UDCA, GUDCA, and DCA/CA were protective factors. Five machine learning models identified four bile acid indicators-UDCA, GUDCA, GCA, DCA-as key predictors for sepsis diagnosis, with a combined model area under the curve (AUC) of 0.880. Additionally, DCA/CA and GCDCA were important predictors for septic shock, with risks increasing by 4.5% (OR = 1.045, 95% CI: 1.009-1.081) and 30% (OR = 0.700, 95% CI: 0.526-0.932), respectively. LCA was a risk factor for respiratory failure in sepsis, with an OR of 2.154 (95% CI: 1.022-4.540). CONCLUSION: Our results highlight the potential of bile acid profiling as a diagnostic and prognostic biomarker for pediatric sepsis. These findings suggest a path toward early intervention, improving patient outcomes by enabling timely detection and treatment.