Paediatric pneumonia: deriving a model to identify severe disease.

BACKGROUND: Community-acquired pneumonia (CAP) is a leading cause of childhood hospitalisation. Limited data exist on factors predicting severe disease with no paediatric-specific predictive tools.

METHODS: Retrospective cohort (2011-2016) of hospitalised CAP cases. We analysed clinical variables collected at hospital presentation against outcomes. Stratified outcomes were mild (hospitalised), moderate (invasive drainage procedure, intensive care) or severe (mechanical ventilation, vasopressors, death).

RESULTS: We report 3330 CAP cases, median age 2.0 years (IQR 1-5 years), with 2950 (88.5%) mild, 305 (9.2%) moderate and 75 (2.3%) severe outcomes. Moderate-severe outcomes were associated with hypoxia (SaO2 <90%; OR 6.6, 95% CI 5.1 to 8.5), increased work of breathing (severe vs normal OR 5.8, 95% CI 4.2 to 8.0), comorbidities (4+ comorbidities vs nil; OR 8.8, 95% CI 5.5 to 14) and being indigenous (OR 4.7, 95% CI 2.6 to 8.4). Febrile children were less likely than afebrile children to have moderate-severe outcomes (OR 0.57 95% CI 0.44 to 0.74). The full model receiver operating characteristic (ROC) area under the curve (AUC) was 0.78. Sensitivity analyses showed similar results with clinical or radiological CAP definitions. We derived a clinical tool to stratify low, intermediate or high likelihood of severe disease (AUC 0.72). High scores (≥5) had nearly eight times higher odds of moderate-severe disease than those with a low (≤1) score (OR 7.7 95% CI 5.6 to 10.5).

CONCLUSIONS: A clinical risk prediction tool is needed for child CAP. We have identified risk factors and derived a simple clinical tool using clinical variables at hospital presentation to determine a child's risk of invasive or intensive care treatment with an ROC AUC comparable with adult pneumonia tools.