Design: Systematic review and meta-analysis of prognostic accuracy studies
Population: 15 studies, 16 cohorts, 2,601,038 paediatric encounters with suspected infection
Primary outcome: In-hospital mortality
Effect: Sensitivity 0.77, specificity 0.73, AUC 0.81 vs 0.71 for IPSCC criteria

A meta-analysis pooled 15 studies, covering more than 2.6 million encounters, that tested the Phoenix Sepsis Criteria in children under 18 with suspected infection in hospital. It was published on 5 August.

For in-hospital mortality, Phoenix had a pooled sensitivity of 0.77 and specificity of 0.73, with an AUC of 0.81, against 0.71 for the older International Pediatric Sepsis Consensus Conference criteria. Specificity was much lower in ICU cohorts (0.48) than in emergency department cohorts (0.99).

The criteria were designed to identify children with infection and organ dysfunction who are at high risk of death, not to catch sepsis early. A child who does not meet Phoenix criteria at triage may still be developing sepsis. The authors call for validation in emergency, ward and lower-resource settings, which includes most Indian hospitals.

Use the Phoenix criteria to stratify risk in children with suspected infection, not as an early screening tool.
Do not withhold sepsis treatment from a child who does not yet meet Phoenix criteria.
Keep early warning signs, such as tachycardia, poor perfusion and altered mental state, as the triggers for urgent assessment.
Expect lower specificity when applying the criteria in intensive care.

Why it matters

A new definition used as a screen would miss children before they have organ dysfunction.

The statistics, in plain English

An AUC of 0.81 means the criteria correctly ranked a child who died above one who survived about 81% of the time. The wide specificity interval (0.51 to 0.87) and the ICU–ED difference reflect very different populations between studies.