Dynamic mean arterial pressure trajectories and sepsis-associated acute kidney injury in ICU patients: a retrospective cohort study

Mean arterial pressure (MAP) is a cornerstone of hemodynamic monitoring in sepsis. However, its potential for identifying distinct patient subphenotypes through dynamic early-phase trends remains unexplored. This study aimed to identify MAP trajectory-based subphenotypes in ICU-diagnosed sepsis and evaluate their association with late-onset SA-AKI. This retrospective cohort study included critically ill adult patients with severe sepsis admitted to the intensive care unit (ICU), using data from the MIMIC-IV database (derivation cohort) and the eICU-CRD (validation cohort). We included adult ICU patients with sepsis, excluding those with pre-existing AKI. Group-based trajectory modeling (GBTM) was applied to MAP values during the first 24 h after sepsis onset to define subphenotypes. Associations between these subphenotypes and late-onset SA-AKI (occurring > 24 h after sepsis onset) in ICU were evaluated using Cox proportional hazards models, with robustness assessed via subgroup and stability analyses. Four distinct sepsis subphenotypes were identified based on early MAP trajectories. Compared with Class 1 (stable 70–75 mmHg, 65.3%), Class 3 (rise-and-drop pattern, 9.1%) was associated with the highest risk of SA-AKI (adjusted HR 1.62, 95% CI 1.36–1.93) and the highest 3-day mortality. Subphenotype Class 4 (stable 80–90 mmHg, 23.4%) also showed a significantly higher SA-AKI risk than Class 1. The association for Class 2 (drop-and-rise pattern, 2.2%) was unstable, likely due to limited sample size. The robustness and reproducibility of these subphenotypes and their outcome associations were supported in the external validation cohort. Early MAP trajectories permit the identification of reproducible sepsis subphenotypes with distinct clinical outcomes in ICU-diagnosed sepsis. Class 1 demonstrates the lowest observed risk of SA-AKI. Classification of ICU-diagnosed septic patients according to dynamic MAP trends represents a simple method for describing early clinical heterogeneity.