Association Between Lactate Trajectories and Clinical Outcomes in Critically Ill Patients with Sepsis-Associated...

Background: Sepsis-associated liver injury (SALI) is a severe complication of sepsis that significantly increases the risk of mortality in intensive care unit (ICU) patients. While lactate is recognized as a prognostic biomarker for sepsis, the predictive potential of dynamic lactate trajectories in patients with SALI has not been investigated. This study aims to elucidate the association between serum lactate trajectories and clinical outcomes. Methods: Latent class trajectories of serum lactate dynamics were modeled in 524 patients with SALI from the MIMIC-IV dataset. Serum lactate levels were measured once daily for four consecutive days after ICU admission. Seven distinct trajectory phenotypes were identified, and their prognostic value was assessed using multivariate Cox regression. Results: The seven lactate trajectory phenotypes identified in SALI patients included the Class 1 low-stable, Class 2 moderate-slow decline, Class 3 low-slow decline, Class 4 moderate-slow decline and mild increase, Class 5 moderate-slow decline and rapid increase, Class 6 high-rapid decline, and Class 7 moderate-persistence groups. Fully adjusted multivariate Cox regression indicated significantly elevated mortality risk in Class 5 (HR 2.81, 95% CI 1.67-4.74) and Class 7 (HR 3.44, 95% CI 2.02-5.85) (all p < 0.001), whereas Classes 1-4 and 6 showed no significant differences. The XGBoost classifier achieved a test set AUROC of 0.7098 (training AUROC = 0.9029), indicating limited generalizability due to overfitting. SHAP analysis was used for exploratory purposes to identify potential predictors, with pH, platelets, and alkaline phosphatase emerging as dominant features. Conclusions: Substantial heterogeneity in lactate trajectories was found among patients with SALI, establishing these dynamic patterns as robust predictors of clinical outcomes.