Prediction of duration of hospitalization in respiratory syncytial virus infection

Identification of variables that predict duration of RSV-associated hospitalization may be useful in the identification of preventive and therapeutic strategies. A recently published prediction model (Michigan model) for the duration of hospitalization in RSV infection demonstrated good discrimination between children with and without an increased likelihood of a hospital stay >or= 7 days, based on variables such as log weight, congenital heart disease, failure to thrive, premature birth, bronchopulmonary dysplasia, other pulmonary diseases, miscellaneous conditions, early mechanical ventilation, and early ribavirin treatment (receiver operating characteristic (ROC) area, 0.89). Validation of this model is of particular interest for Europe, since the mean duration of hospitalization in The Netherlands is approximately twice that in the USA. The objective of the current study was 1) to validate the Michigan model for RSV hospitalized patients in a large university hospital in The Netherlands, and 2) to develop our own prediction model for a prolonged hospital stay. Data from 177 children younger than 12 months of age admitted with confirmed RSV infection to the Sophia Children's Hospital Rotterdam between 1992-1995, were used for valiation of the Michigan model and derivation of the Rotterdam model. Mean duration of hospitalization for the Rotterdam database was 10.3 (+/-6.3) days, with a median of 9 days; 138 (78%) patients had a hospital stay >or= 7 days. The Michigan model performed poorly when applied to the Rotterdam database, with an ROC area of 0.65 (95% CI, 0.57-0.73). The Rotterdam prediction model (hospital stay >or= 9 days, the median in our database) considered weight and need for oxygen supplementation. The ROC area was 0.65 (95% CI, 0.57-0.73). When using data from patients for the 1995-1996 season, the ROC area was 0.52 (95% CI, 0.34-0.72). The Michigan and the Rotterdam models failed to identify a considerable number of patients who had a prolonged hospital stay, with a low false-positive rate. We conclude that neither the Michigan, nor the Rotterdam model reliably predicted the duration of hospitalization based on demographic and clinical variables.