BACKGROUND AND AIM: Early recognition of sepsis remains one of the main hurdles towards timely administration of sepsis treatment bundles. Past approaches to identify children with sepsis primarily use scoring systems, which may not reflect phenotypic information adequately. This study aimed to identify distinct clinical symptom profiles of sepsis in children screened for sepsis in the Emergency Department (ED).

METHOD: Children who were screened for sepsis in 12 EDs in Queensland, Australia, between August 2018 and December 2019 were included. We used profile regression, a Bayesian clustering method, to cluster 16 sepsis screening criteria assessed in the Queensland Paediatric Sepsis Pathway, to identify distinct symptom profiles and estimate the risk of sepsis associated with each profile. Sepsis was defined as Senior Medical Officer diagnosis of sepsis combined with administration of antibiotics in the ED.

RESULTS: 523 (15.1%) children had sepsis, out of 3473 screened. We identified seven distinct symptom profile clusters. Two clusters had high sepsis risk (Cluster 1: risk probability (Pr) 0.73, 95% Credible Interval (CI) 0.55, 0.89; Cluster 2: Pr 0.69, 95% CI 0.59, 0.80). Cluster 1 was characterised by severe illness features such as severe respiratory distress and hypotension, while Cluster 2 was primarily distinguished by elevated lactate. Four additional clusters with moderate risk and one large cluster with comparatively lower risk of sepsis were also identified.

CONCLUSIONS: This study identified distinct clinical symptom profiles for paediatric sepsis, providing insight into which combinations of sepsis screening criteria are associated with increased risk for paediatric sepsis. External validation assessing cluster stability is required.

Keywords: Sepsis screening, clinical symptom profiles, Bayesian profile regression, cluster analysis