Abstract: FR-PO0086
Unsupervised Machine Learning in Identification of Septic Shock Phenotypes and Their In-Hospital Outcomes: A Multicenter Cohort Study
Session Information
- AKI: Epidemiology and Clinical Trials
November 07, 2025 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Acute Kidney Injury
- 102 AKI: Clinical, Outcomes, and Trials
Authors
- Lorenzo Capps, Maria Jose, RWJBarnabas Health, Toms River, New Jersey, United States
- Ang, Song Peng, RWJBarnabas Health, Toms River, New Jersey, United States
- Lee, Eunseuk, RWJBarnabas Health, Toms River, New Jersey, United States
- Rajendran, Jackson, RWJBarnabas Health, Toms River, New Jersey, United States
- Chia, Jia Ee, Texas Tech University Health Sciences Center El Paso, El Paso, Texas, United States
- Iglesias, Jose I., RWJBarnabas Health, Toms River, New Jersey, United States
Background
Septic shock management remains standardized despite its pathophysiological heterogeneity. Current risk stratification tools, including lactate and severity scores, may inadequately capture disease diversity, limiting personalized patient care. We aim to define septic shock subphenotypes using admission data and laboratory parameters to evaluate their association with clinical outcomes.
Methods
Retrospective analysis of 10,462 adults with ICD–10–defined septic shock admitted to intensive care units (ICU) between 2014 and 2015. Two-Step Cluster Analysis (log-likelihood distance, Bayesian Information Criterion) identified two phenotypes. Outcomes included mortality, days on mechanical ventilation, vasopressor use, acute kidney injury (AKI), AKI requiring renal replacement therapy (RRT), ICU and hospital length of stay.
Results
Two clusters emerged: Cluster 1 (n=5,355): Higher lactate (2.40 vs. 2.20 mmol/L), creatinine (1.39 vs. 1.20 mg dL), SOFA (7.05±3.85 vs. 6.76±3.87), and neutrophil-to-lymphocyte ratio (11.12 vs. 10.38; all p<0.001). Cluster 2 (n=5,107): Milder initial derangements. Cluster 1 had higher mechanical ventilation use (46.1% vs. 42.2%), stage 3 AKI (17.2% vs. 15.2%), and dialysis (6.6% vs. 5.2%; p≤0.005). However, in-hospital mortality (15.4% vs. 15.8%, p=0.615) and lengths of stay (ICU: 2.18 vs. 2.26 days; hospital: 6.63 vs. 6.80 days; p>0.25) were comparable among clusters.
Conclusion
Two phenotypes were identified: one with pronounced early organ dysfunction (Cluster 1) and one with milder presentations (Cluster 2). Cluster 1 showed a greater proportion of patients with severe kidney injury, including stage 3 AKI (17.2 % vs 15.2 %) and a higher requirement for RRT. Cluster 2 exhibited milder initial laboratory abnormalities but showed equal or higher rates of organ support, suggesting a phenotype of “silent risk” characterized by relative immune suppression or limited physiological reserve. Paradoxically, despite divergent support needs, both clusters exhibited equivalent mortality and resource utilization. These findings highlight limitations of homogenized management, challenging the assumption that early biochemical severity uniformly portends poor prognosis, advocating for phenotype-driven strategies to optimize care in septic shock.