Time series model to predict burden of viral respiratory illness on a pediatric intensive care unit

Abstract

Background: In the United States, viral respiratory infections are a leading cause of illness and hospitalization in young children. Caring for children with severe viral respiratory illness can have a substantial impact on resource utilization in the pediatric intensive care unit (PICU). The objective was to build a robust model that captures the periodicity of severe pediatric viral respiratory illness and forecasts the incidence of viral respiratory illness in the PICU.

Methods: This was a retrospective time series analysis in a PICU at a quaternary care children's hospital. Patients were less than 18 years of age with laboratory-confirmed respiratory syncytial virus, influenza, parainfluenza, or adenovirus infection on or during admission from October 2002 to September 2008. Time series modeling techniques were to used to model viral incidence, using maximum likelihood estimation to identify model parameters.

Results: A total of 289 patients were included in the analysis. An autoregressive model of order 10 that included an exogenous variable of community viral incidence from the previous month was able to explain and predict viral incidence in the PICU. A limitation of the study was that it included a single institution.

Conclusions: The identified model, derived from historical data from both a PICU and the local community, produced accurate 1-month and 3-month forecasts of severe viral respiratory illness presentation to the PICU. These results suggest that time series models may be useful tools in forecasting the burden of severe viral respiratory illness at the institutional level, helping institutions make decisions to optimize the distribution of resources.