Differential diagnosis of viral and bacterial community-acquired pneumonia in children using logistic regression

Abstract

Purpose: to develop a logistic regression model for early differential diagnosis of viral and bacterial community-acquired pneumonia in children. Materials and methods. A prospective study of the clinical and laboratory features of community-acquired pneumonia (CAP) was carried out in 130 children. A complex approach using bacteriological, molecular genetic, serological diagnostic methods (material – nasopharyngeal swabs, pleural effusion, blood) was used to establish the etiology of CAP. Two etiological groups of CAP were distinguished: viral (n=76), bacterial (n=44). The binary logistic regression method was used to create a differential diagnosis model. Potential predictors of CAP etiology were anamnestic, clinical, laboratory (complete blood count), instrumental (chest x-ray) data. The quality of the constructed regression model was evaluated on a validation set of 42 children with CAP. Results. A statistically significant (p<0.001) regression model was created, which looks like: y=exp(2.04- 2.87×X1+2.2×X2+0.13×X3+0.12×X4-0.44×X5)/(1+exp(2.04- 2.87×X1+2.2×X2+ +0.13×X3+0.12×X4-0.44×X5)), where “y” is the probability of bacterial CAP in children, X1 – bronchoobstructive syndrome (BOS; no – 0, yes – 1), X2 – age (<4.5 years – 0, ≥4.5 years – 1), X3 – absolute neutrophil count (ANC, *109 cells/l), X4 – relative band count (Band,%), X5 – platelet distribution width (PDW,%). At y≥0.31, bacterial CAP is diagnosed with sensitivity of 81.8% and specificity of 81.6%, at y<0.31 viral CAP is diagnosed. Proposed predictors are widely available in clinical practice, which makes it possible to apply the method in outpatient and inpatient settings. The regression model confirmed the high classificatory ability using cross-validation. Conclusion. The regression model based on a complex of clinical (age, BOS) and laboratory signs (ANC, Band, PDW) has high statistical significance (p<0.001) and excellent diagnostic ability (84.2%) and can be used for early differential diagnosis viral and bacterial pediatric CAP in different health care settings.