Sepsis subphenotyping based on organ dysfunction trajectory.

Area of Science:

• Critical Care Medicine
• Computational Biology
• Clinical Informatics

Background:

• Sepsis is a complex syndrome with varying patient trajectories.
• Organ dysfunction is key in sepsis, but trajectory-based subphenotypes are understudied.
• Identifying distinct sepsis subphenotypes is crucial for effective management.

Purpose of the Study:

• To identify distinct subphenotypes of sepsis based on Sequential Organ Failure Assessment (SOFA) score trajectories.
• To analyze patient characteristics and outcomes across identified sepsis subphenotypes.
• To develop predictive models for sepsis subphenotype membership.

Main Methods:

• Utilized dynamic time warping (DTW) and hierarchical agglomerative clustering (HAC) on 72-h SOFA scores from four ICU cohorts.
• Compared baseline characteristics and organ dysfunction patterns between subphenotypes.
• Developed and validated a random forest model to predict subphenotype membership.

Main Results:

• Identified four sepsis subphenotypes: Rapidly Worsening (13.1%), Delayed Worsening (20.5%), Rapidly Improving (41.3%), and Delayed Improving (25.1%).
• Subphenotypes exhibited distinct baseline characteristics, organ dysfunction patterns, and outcomes.
• The Rapidly Worsening group had the highest mortality (28.3%) despite lower initial SOFA scores.
• Predictive model achieved 0.78 accuracy at 6 hours and 0.87 at 24 hours post-ICU admission.

Conclusions:

• Four novel, validated trajectory-based sepsis subphenotypes were identified.
• These subphenotypes have implications for clinical trial design, potentially improving patient stratification and outcome prediction.
• Further research into the pathophysiology of these trajectories may uncover new therapeutic targets.