Written by Joshua Belfer
Spoon Feed
The Pneumonia Risk Score (PRS) outperformed the CARPE DIEM model in predicting pediatric radiographic pneumonia, but neither model outperformed clinical judgment.
Can we predict pediatric pneumonia without a chest x-ray?
This prospective validation study evaluated the performance of two clinical prediction models – Pneumonia Risk Score (PRS) and CARPE DIEM – in identifying radiographic pneumonia among 202 children (median age 3 years, IQR 1–6 years) in a pediatric ED. Radiographic pneumonia was found in 41% of the cohort. The PRS model – which includes age, fever, triage oxygen saturation, and the presence of wheeze and crackles on auscultation – had a superior AUROC of 0.72 (95%CI 0.64–0.79) compared to CARPE DIEM (AUROC 0.59, 95%CI 0.51–0.67; p < 0.01), which includes age, focal decreased breath sounds, and fever duration. The PRS model also had better sensitivity (65.2%) and specificity (72.7%) than the CARPE DIEM model (sensitivity 56.5%, specificity 60.9%). Recalibration improved CARPE DIEM’s performance, but neither model reached a level of accuracy sufficient for independent use. The study is limited by its single-center design, small sample size, and reliance on radiographic pneumonia as the reference standard, which itself lacks perfect reliability.
How will this change my practice?
There have been plenty of studies looking at appropriate treatment duration for pediatric pneumonia, but this study focuses on the best way to make the initial diagnosis. While prediction models can help guide clinical decision-making, neither of the models studied can replace clinician judgment. That said, there may be utility in using the PRS model as a potential adjunct in borderline cases where the decision to obtain a chest radiograph isn’t clear-cut. This study also reinforces the continued need for clinical gestalt, especially given the limitations of chest radiographs themselves.
Source
External Validation of Two Clinical Prediction Models for Pediatric Pneumonia. Acad Pediatr. 2025 Jan-Feb;25(1):102564. doi: 10.1016/j.acap.2024.08.009. Epub 2024 Aug 17. PMID: 39159892
