Background: Risk stratification tools for paediatric community-acquired pneumonia (CAP) in well-resourced settings are scarce. We prospectively developed models to predict CAP severity within a multinational cohort of paediatric emergency departments (EDs). Our primary objective was to develop a risk prediction model to discriminate between mild CAP and moderate or severe CAP to assist clinicians in determining the need for hospitalisation.

Methods: This prospective cohort study was conducted from Feb 6, 2019, to June 30, 2021, at 73 EDs in 14 countries. Children aged 3 months to <14 years with clinical diagnoses of CAP were included. Children were excluded if they were recently hospitalised or had a chronic complex condition (eg, immunodeficiency). The primary outcome was severity, defined as mild (CAP treated in the outpatient setting or hospitalisation <24 h with no use of oxygen or intravenous fluids during that time), moderate (hospitalisation <24 h with oxygen or fluids, or hospitalisation ≥24 h regardless of interventions but without an outcome qualifying as severe CAP), or severe (chest drainage, intensive care unit admission >24 h, positive-pressure ventilation, septic shock, vasoactive infusions, extracorporeal membrane oxygenation, or death) occurring within 7 days of the ED visit. Models were developed using logistic regression with bootstrap validation.

Findings: Of 2222 children in the overall study population (1103 [49·7%] female, 1119 [50·3%] male; median age 3 years [IQR 1-5]), 1290 (58·1%) had mild CAP, 812 (36·5%) moderate, and 120 (5·4%) severe. Primary analyses were performed in 1901 patients with complete data: 1011 (53·2%) mild, 772 (40·6%) moderate, and 118 (6·2%) severe CAP. Congestion or rhinorrhoea was negatively associated with moderate or severe CAP (adjusted odds ratio 0·59 [95% CI 0·46-0·76]), while abdominal pain (1·52 [1·17-1·97]), refusal to drink (1·57 [1·24-2·00]), antibiotics before ED visit (1·64 [1·29-2·10]), chest retractions (2·86 [2·24-3·65]), respiratory rate above the 95th percentile for age (1·63 [1·29-2·06]), heart rate above the 95th percentile for age (1·64 [1·27-2·12]), and hypoxaemia (oxygen saturation 90-92%, 3·24 [2·46-4·27]; <90%, 13·39 [8·64-20·73]) were positively associated. The model accurately discriminated between mild CAP and moderate or severe CAP (c-statistic 0·82 [95% CI 0·80-0·84]). Similar results were found in those with radiographic CAP, with decreased breath sounds and multifocal opacities on radiography as additional predictors (c-statistic 0·82 [0·80-0·85]).

Interpretation: We developed accurate, pragmatic severity risk prediction models among children with CAP. After future external validation, these models have the potential to provide individualised risk assessments that can be incorporated into clinical judgement in well-resourced health systems to improve management.

Funding: Division of Emergency Medicine at Cincinnati Children's Hospital Medical Center, Division of Emergency Medicine at Ann & Robert H. Lurie Children's Hospital of Chicago, and Department of Emergency Medicine at University of California, Davis.