133 - validation of a Global Respiratory Severity Score in Infants with Primary RSV Infection

Background: We recently developed a global respiratory severity score (GRSS) as a research tool from (n=139) infants with primary respiratory syncytial virus (RSV) infection enrolled prospectively in the Assessing Predictors of Infant RSV Effects and Severity (AsPIRES) study. The objective of the present study was to validate our original findings that the GRSS correlates well with clinical outcomes including hospitalization and length of stay (LOS) utilizing an independent cohort.

Methods: Clinical and demographic data on infants with primary RSV infection were abstracted from the electronic medical record. The GRSS was calculated by applying the original training data formula to the new data set. We compared the mean GRSS between the hospitalized and non-hospitalized group with Welch two sample t-test, and correlated it with hospitalization and LOS using Pearson’s correlation test.

Results: A total of 184 (98 hospitalized and 86 non-hospitalized) subjects were enrolled. The hospitalized and non-hospitalized infants were different in general appearance, the percentage with rales, retractions, lethargy, respiratory rate and oxygen saturation.  The hospitalized group had a significantly (t=9.334, p< 0.0001) higher GRSS (4.20±2.10) than the non-hospitalized group (1.76±1.41). Using GRSS ≤3.5 as the classification criterion, we correctly predicted the hospitalization status of 131 (71.2%) subjects. The area under the ROC curve of the GRSS as a classifier of hospitalization is AUC=0.8265 (p< 0.0001).   Pearson correlation between the GRSS and LOS is  (p< 0.0001).Area under the ROC curve of the GRSS
Boxplot demonstrating the difference in GRSS between hospitalized and non hospitalized infants


Conclusion: Infants hospitalized with RSV have a significantly higher GRSS than non-hospitalized subjects, and the GRSS is strongly associated with LOS. GRSS can also serve as an acceptable predictor of hospitalization. We are currently re-training the GRSS model with both datasets and believe it will lead to an improved predictive power of hospitalization.