A multi-center validation study of clinical sepsis phenotypes

Journal article

Importance: Four clinical phenotypes of sepsis based on electronic health data have been proposed. Although promising, their generalizability remains uncertain, and multi-site validation is needed.

Objectives: To validate, using the same methods and inclusion criteria, the four clinical phenotypes derived from Sepsis Endotyping in Emergency Care (SENECA) data.

Methods: A multi-site retrospective study of adult patients admitted to the Emergency Departments in Stockholm (Karolinska, 2011-2023), Oxford (2014-2021) and Oslo (2017-2023) University Hospitals. We included encounters with body fluid culture(s) taken, documented antibiotic administration, and SOFA scores of x2 all within 6 hours of admission. We used consensus clustering with k-means to derive four clinical phenotypes at each site, comparing them with the SENECA-derived phenotypes, as well as with one another.

Results: There were 30,865 patient encounters in Stockholm, 15,575 in Oxford and 1,806 in Oslo. First, we found little consistency between the SENECA clinical phenotypes and each site’s own phenotypes, with Cohen’s kappa 0.32 (Stockholm), 0.37 (Oslo) and 0.40 (Oxford); the respective Adjusted Rand Indices were 0.21 (Stockholm), 0.27 (Oslo), 0.26 (Oxford). Second, there was also little consistency between the phenotypes derived in Stockholm, Oxford, and Oslo.

Conclusions: The four clinical phenotypes of the SENECA data are not generalizable across three independent cohorts.

Authors: Yoon, CH; Sjoholm, D; Wickstrom, K; Danielsen, AS; Prebensen, C

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: American Medical Association
• Journal: JAMA Network Open
• Volume: 9
• Issue: 6
• Article number: e2616134
• Publication date: 2026-06-01
• Acceptance date: 2026-04-13
• DOI: 10.1001/jamanetworkopen.2026.16134
• EISSN: 2574-3805
• Language: English
• Keywords: electronic health data; sepsis; phenotypes; subgroups