Dynamic trajectories of organ dysfunction in sepsis using the SOFA-2 score and early prediction from multicenter cohorts - PubMed
6 hours ago
- #Machine Learning Prediction
- #SOFA-2 Score
- #Sepsis
- This study uses the SOFA-2 score to track dynamic organ dysfunction trajectories over 14 days in sepsis patients from multicenter cohorts.
- Three distinct trajectory patterns were identified: rapid recovery (36.2%), delayed recovery (45.0%), and an unfavorable persistent/severe dysfunction trajectory (18.8%), with corresponding 28-day mortality rates of 11.3%, 23.7%, and 52.4%.
- An ensemble machine-learning model was developed to predict the unfavorable trajectory early, using the first 72 hours of ICU data, achieving high AUROC scores (0.84-0.88) and providing a median 36-hour lead time for intervention.
- Implementation of a protocolized alert system based on these predictions was associated with a significant reduction in 28-day mortality (from 27.0% to 21.7%), shorter ICU stays, and reduced vasopressor duration.
- The findings suggest that dynamic SOFA-2 trajectory monitoring and early, explainable prediction models can enhance precision care, improve patient outcomes, and optimize resource use in sepsis management.