Clinical Prediction Models for Pneumonia in Children Presenting to an Emergency Department in a Resource-Limited Setting Using Lung Ultrasound Diagnosis as the Gold Standard

Abstract

Introduction: Clinical prediction rules for pediatric pneumonia often rely on radiographic pneumonia for diagnosis; however, lung ultrasound has higher diagnostic accuracy. Our objective was to derive a clinical prediction model for pneumonia in children under five using lung ultrasound as the criterion standard.

Methods: This was a prospective study of children under five presenting to an emergency department (ED) with respiratory complaints in a resource-limited setting. Clinical findings, chest X-ray, and lung ultrasound results were recorded for each patient. Classification tree models were used to predict pneumonia using lung ultrasound as the criterion standard. Separate models were used without and with inclusion of chest X-ray results.

Results: Of 386 patients enrolled, 125 patients (32.4%) had pneumonia on lung ultrasound. The mean age was 20.8 (SD 15.5) months. Using recursive feature selection, three variables provided the best prediction for pneumonia, namely, crepitations, retractions, and difficulty breathing, demonstrating a sensitivity of 74.2% and specificity of 38.5%. The algorithm including chest X-ray provided a sensitivity of 51.6% and specificity of 87.7%.

Conclusions: Using lung ultrasound as the gold standard, no single clinical finding or combination of clinical findings provided enough accuracy to reliably diagnose pneumonia in children under five years.