A variety of postnatal therapies have been and will be evaluated for preven
tion or treatment of neonatal chronic lung disease (CLD). A simple method f
or early selection of the highest risk infants would optimize intervention
trials. Our study compared a clinical scoring system for predicting neonata
l CLD (defined at 36 weeks postconceptional age) with previous regression m
odels developed by Sinkin et al. (Sinkin model) [Pediatrics 1990:86:728-736
] and Ryan et at. (Ryan model) [Eur J Pediatr 1996;668-671] in two distinct
populations. A respiratory failure score (RFS) was prospectively developed
for infants at <32 weeks of gestation admitted to the Wilford Hall Medical
Center from January 1990-December 1992. Logistic regression modeling ident
ified three independent predictors of CLD: gestation, birth weight, and RFS
. Applying a modified RFS score (to include gestation and birth weight), th
e RFS, Sinkin, and Ryan models were compared among high-risk infants admitt
ed to Wilford Hall from January 1993-December 1995, and to Crawford Long Ho
spital (Atlanta, GA) from January 1993-December 1994. Predictive values, se
nsitivity, specificity, and receiver operating characteristic (ROC) curves
were determined for the primary outcome variable: OLD at 36 weeks of correc
ted gestation.
Of 248 infants at <32 weeks admitted to Wilford Hall, 220 survived >7 days.
Thirty of 31 (97%) infants diagnosed with CLD were less than or equal to 2
9 weeks or less than or equal to 1,000 g at birth. Despite important demogr
aphic and treatment differences between the study populations, similar ROC
curves were found for each scoring method when individually evaluated among
the three study groups. The RFS method at 72 h demonstrated the greatest a
rea under the ROC curve for prediction of neonatal CLD in the groups as a w
hole. Application of the RFS method for early prediction of neonatal CLD at
age 72 h should improve patient selection for early prevention trials. Ped
iatr Pulmonol, 1999; 27:388-394. Published 1999 Wiley-Liss, Inc.dagger