A tool to distinguish viral from bacterial pneumonia

Establishing the etiology of community-acquired pneumonia (CAP) in children at admission is challenging. As a result, most children receive antibiotics that do not need. This study aims to build and validate a diagnostic tool combining clinical, analytical and radiographical features to sequentially differentiate viral from bacterial CAP, and among bacterial CAP, typical from atypical bacteria, to improve choice of treatment. Methods Consecutive hospitalized children between 1 month and 16 years of age with CAP were enrolled. An extensive microbiological workup was performed. A score was built with a training set of 70% patients, to first differentiate between viral and bacterial CAP and secondly, typical from atypical bacterial CAP. To select variables, a Ridge model was used. Optimal cut-off points were selected to maximize specificity setting a high sensitivity (80%). Weights of each variable were calculated with a multivariable logistic regression. The score was validated with the rest of the participants. Results In total, 262 (53%) children (median age, 2 years, 52.3% male) had an etiological diagnosis. The step 1 discriminates viral from bacterial CAP. Bacterial CAPs were classified with a sensitivity=97%, a specificity=48%, and a ROC area under the curve (AUC)=0.81. If a CAP was classificated as bacterial, it was assessed with step 2. The step 2 differentiates typical vs. atypical bacterial CAP. Typical bacteria were classified with a sensitivity=100%, a specificity=64%, and AUC=0.90. Conclusion This two-steps tool can facilitate the physician decision to prescribe antibiotics without compromising patient safety.