Abstract Objective To identify independent predictors of hospital admission among children with pneumonia using a penalized logistic regression model that integrates physiologic measures and viral pathogen data. Methods (Design, Setting, and Participants) This retrospective cohort study included 11,670 pediatric encounters who presented to the emergency department with cough and respiratory distress and had a PCR-confirmed pathogen (using ICD-10 codes) at a regional tertiary pediatric center between 2019 and 2024. The analytic sample comprised approximately 1,500 complete pneumonia cases with available age, vital signs, and multiplex respiratory pathogen testing results. The primary outcome was hospital admission (1 = admitted, 0 = discharged). Predictors included age in years, oxygen saturation (SpO2), respiratory rate per 10 breaths/min (RR10), heart rate per 10 beats/min (HR10), and binary indicators for common respiratory pathogens. A penalized (L2) logistic regression model was applied using stratified 5-fold cross-validation. Model performance was evaluated using out-of-fold predictions, the area under the receiver operating characteristic curve (AUC), and the Brier score. Results The most common pathogens were Rhinovirus/Enterovirus (RV/EV), Respiratory syncytial virus (RSV) and adenovirus. The penalized model demonstrated good discrimination and calibration, with an AUC of 0.76 (95% CI, 0.76-0.82) and a Brier score of 0.13. Lower SpO2 was strongly associated with hospital admission (adjust. odds ratio aOR, 0.41 per 1% increase), as was higher respiratory rate (aOR, 2.15 per 10 breaths/min). Age and heart rate were modest positive predictors. Mycoplasma pneumoniae infection was independently associated with emergency department discharge (aOR, 0.29), whereas most other pathogens, including respiratory syncytial virus (RSV), showed no independent effect after adjustment for physiologic variables. Calibration curves showed close agreement between predicted probabilities and observed admission rates across deciles of risk. Conclusions A penalized logistic regression model incorporating age, vital signs, and viral pathogen indicators predicted hospital admission among children with pneumonia with strong discrimination and acceptable calibration. Physiologic instability, particularly hypoxemia and tachypnea, emerged as the dominant drivers of admission risk, while pathogen type contributed minimally. This model provides a transparent, reproducible framework for data-informed disposition decision support in pediatric respiratory illness. This abstract is funded by: none
Hesen et al. (Fri,) studied this question.