Related Experiment Videos

Severity of illness models for respiratory syncytial virus-associated hospitalization

F W Moler1, S E Ohmit

  • 1Department of Pediatrics and Communicable Diseases, University of Michigan Medical School, and Department of Epidemiology, School of Public Health, University of Michigan, Ann Arbor, Michigan, USA.

Insights

Predictive models for pediatric respiratory syncytial virus (RSV) hospitalizations are feasible. These models accurately identify children at risk for prolonged stays, aiding clinical management.

Area of Science:

  • Pediatric Infectious Diseases
  • Clinical Epidemiology
  • Biostatistics

Background:

  • Respiratory syncytial virus (RSV) is a common cause of pediatric hospitalization.
  • Predicting prolonged hospital stays in young children with RSV is clinically significant.
  • Existing severity of illness models may not be optimized for RSV in this age group.

Purpose of the Study:

  • To assess the feasibility of developing multivariate severity of illness models for pediatric patients hospitalized with RSV.
  • To identify key variables measurable on the first hospital day that predict prolonged hospitalization.
  • To evaluate the predictive performance (discrimination and calibration) of such models.

Main Methods:

  • Retrospective cohort study of 802 hospitalized children aged 2 years or younger with community-acquired RSV.
  • Multivariate logistic regression analysis using nine variables assessed on day 1 of hospitalization.
  • Receiver operator characteristic (ROC) curve analysis for model discrimination.
  • Goodness-of-fit testing for model calibration.

Main Results:

  • 182 (23%) of patients experienced prolonged hospitalization (≥7 days).
  • Multivariate logistic regression identified significant predictors of prolonged hospitalization (p < 0.0001).
  • The model demonstrated excellent discrimination (Area Under the Curve = 0.894) and calibration (p = 0.216).

Conclusions:

  • Multivariate severity of illness models for RSV-associated hospitalizations are feasible and possess excellent predictive properties.
  • The developed models show strong classification, discrimination, and calibration.
  • Further research is needed to validate generalizability across different centers and epidemics.

Related Concept Videos