Testing Models Predicting Severity of Respiratory Syncytial Virus Infection on the PICNIC RSV Database

Abstract

To determine the sensitivity and specificity of published prognostic models to predict morbidity resulting from lower respiratory tract disease caused by respiratory syncytial virus in an independent pediatric population and to assess the accuracy of single risk factors in predicting adverse outcome. All articles obtained from a MEDLINE search that used the terms prognosis or sequelae and respiratory syncytial virus, and from the references of these articles, were reviewed. Studies were included if risk factors and outcomes were defined and if information was available in a database of prospectively enrolled patients with respiratory syncytial virus infections. A probability of adverse outcome was assigned to each patient in the cohort using prognostic models described in the articles. A test was considered positive if the probability of the adverse outcome was 5% or more. Six hundred eighty-nine patients hospitalized with respiratory syncytial virus in seven tertiary care centers across Canada were prospectively enrolled in the Pediatric Investigators Collaborative Network on Infections in Canada database. The sensitivity and specificity of single predictors and of models in predicting severe disease were determined. The sensitivity of single predictors varied from 17% to 46%. A model that used age and oxygen saturation at admission in previously well infants had a sensitivity of 98% and a specificity of 47% when predicting intensive care unit admission. Another model that included age at hospitalization, gestational age, presence of an underlying condition, and respiratory syncytial virus subtype used to predict the outcome of a high severity index had a sensitivity of 77% and a specificity of 76%. When the above model was modified by exclusion of viral subgroup, sensitivity increased to 94%, but specificity decreased to 46%. Previously described prognostic models were generalizable to an independent study population.