Multicenter Validation of Clinical Sepsis Phenotypes.

Area of Science:

• Critical Care Medicine
• Data Science in Healthcare
• Clinical Phenotyping

Background:

• Electronic health records have enabled the proposal of four distinct clinical phenotypes for sepsis.
• The generalizability and multisite validation of these sepsis phenotypes remain uncertain.
• Accurate sepsis subtyping is crucial for targeted treatment and improved patient outcomes.

Purpose of the Study:

• To validate the four clinical phenotypes derived from Sepsis Endotyping in Emergency Care (SENECA) data.
• To assess the generalizability of SENECA sepsis phenotypes across multiple international healthcare settings.
• To compare phenotypes derived at individual sites with the original SENECA phenotypes.

Main Methods:

• A multisite retrospective cohort study involving adult patients from university hospitals in Stockholm, Oxford, and Oslo.
• Inclusion criteria mandated body fluid cultures, antibiotic administration, and Sequential Organ Failure Assessment scores ≥2 within 6 hours of admission.
• Consensus clustering with k-means was employed to derive and compare sepsis phenotypes across sites and with SENECA data.

Main Results:

• Analysis included 30,865 (Stockholm), 15,575 (Oxford), and 1,806 (Oslo) patient encounters.
• Low consistency was observed between SENECA phenotypes and site-specific phenotypes (Cohen κ: 0.32-0.40; Adjusted Rand Indices: 0.21-0.27).
• Significant inconsistencies were also found between phenotypes derived from the three independent study sites.

Conclusions:

• The four clinical phenotypes identified in the SENECA data lack generalizability across diverse European healthcare settings.
• Current clustering methods may introduce stochasticity, limiting the reliability of derived sepsis phenotypes.
• Alternative approaches are necessary to explore underlying sepsis heterogeneity and identify robust subgroups.