Severity of illness models for respiratory syncytial virus-associated hospitalization.

Abstract

The objective of this investigation was to examine the feasibility of multivariate severity of illness models for pediatric patients hospitalized with respiratory syncytial virus (RSV) infection. From a preexisting retrospective cohort study database, all infants and children 2 yr of age or younger with community-acquired RSV infection admitted to the University of Michigan's C. S. Mott Children's Hospital during nine epidemics were examined. The study group consisted of 802 hospitalized patients younger than 2 yr of age with community-acquired RSV infection; 182 (23%) patients had prolonged hospital length of stay defined as 7 d or greater. Multivariate logistic regression modeling of nine variables measurable during the first hospital day was strongly associated with prolonged hospitalization (p < 0.0001). Receiver operator characteristic curve analysis resulted in an area under the curve of 0.894, indicating excellent model discrimination. Goodness-of-fit testing indicated excellent model calibration for observed versus predicted outcomes (p = 0.216). We conclude that severity of illness models for RSV-associated hospitalization with excellent predictive properties in terms of classification, discrimination, and calibration are possible. Further study is required to determine if such models are generalizable across multiple centers and epidemics.