A widely used electronic sepsis alert detected about a third of paediatric sepsis in the emergency department

Design

Retrospective multicentre diagnostic accuracy cohort

Population

166,943 paediatric ED visits at 8 EDs in New York, 2024–2025; 189 sepsis cases

Primary outcome

Sensitivity and predictive values for Phoenix sepsis within 24 hours

Effect

Sensitivity 36% (29–43%); PPV 3%; sensitivity before onset 30%

This retrospective cohort evaluated the Epic Pediatric Sepsis Early Detection Model, an electronic health record alert widely deployed in US emergency departments, across 166,943 paediatric visits at eight EDs in one New York health system in 2024–25. Sepsis was defined by a Phoenix sepsis score of 2 or more within 24 hours; there were 189 cases.

The alert's sensitivity was 36% (95% CI 29–43%) and its positive predictive value 3%; its negative predictive value exceeded 99%, as expected when sepsis is rare. It fired before sepsis onset in only 30% of cases. Sensitivity was lower in children under five (20%), Black children (22%) and EDs outside tertiary children's hospitals (25%).

An alert that misses two in three children with sepsis and is wrong 97 times in 100 when it fires adds alarm fatigue without catching the children who matter most. Clinical recognition — vital signs against age-specific ranges, perfusion, mental state and parental concern — remains the screen.

• Do not rely on an automated sepsis alert to identify septic children; most cases did not trigger it.
• The alert performed worst in children under five, the group at highest risk.
• Assess every febrile child with age-specific heart rate, respiratory rate, capillary refill and conscious level.
• Take parental concern that a child is 'not right' seriously as a sepsis red flag.
• Before adopting any commercial sepsis model, ask for independent local validation.

Why it matters

A tool many EDs assume is screening for sepsis was missing most of it.

Don't overread it

This evaluated one model in one health system; other tools and settings may perform differently.

The statistics, in plain English

Sensitivity of 36% means 64% of children with sepsis were not flagged. A positive predictive value of 3% means that of every 100 alerts, about 97 were false alarms. A high negative predictive value is unhelpful here because sepsis is rare in any case.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.