Predicting prolonged length of stay in hospitalized children with respiratory syncytial virus.

BACKGROUND

Respiratory syncytial virus (RSV) is the most common cause of lower respiratory tract infections in children. This study aimed to predict the prolonged length of stay in children admitted to hospital with RSV.

METHODS

Children aged <2 years with RSV in the National Inpatient Sample (NIS) were included in the analyses. The primary outcome was prolonged length of stay (≥90th percentile). Logistic regression models were developed using data from 2016; internal validation was completed using a bootstrapped sample. Data from 2017 were used to validate out-of-sample discrimination and calibration of the models.

RESULTS

The sample included 9589 children; 1054 had prolonged length of stay (≥7 days). Children who were younger, transferred from another hospital, and required intubation during admission had a higher risk of prolonged length of stay. The prediction model included age, transport, intubation, comorbidities, hospital location, and teaching status. The area under the receiver operating characteristic curve was 0.73, demonstrating good predictive ability. The model performed similarly in external validation.

CONCLUSIONS

Variables that predict the prolonged length of stay for RSV include younger age, transport, intubation, comorbidities, hospital location, and teaching status. This can be used to predict children who will have a prolonged length of stay when hospitalized for RSV.

IMPACT

There are no recommended treatments for RSV; medical care involves supportive treatment such as oxygen delivery, hydration, and antipyretics. The clinical course is difficult to predict, partially attributable to the supportive nature of care and the sparsity of evidence-based therapies for this population. A prediction model was developed, demonstrating variables that predict prolonged length of stay in RSV hospitalizations, including age, interhospital transport, intubation, comorbidities, hospital location, and teaching status. The model was developed with a sample size of 9589 that is representative of all hospitalizations in the United States.