Early prediction of neonatal chronic lung disease: a comparison of three scoring methods

Abstract

A variety of postnatal therapies have been and will be evaluated for prevention or treatment of neonatal chronic lung disease (CLD). A simple method for early selection of the highest risk infants would optimize intervention trials. Our study compared a clinical scoring system for predicting neonatal CLD (defined at 36 weeks postconceptional age) with previous regression models developed by Sinkin et al. (Sinkin model) [Pediatrics 1990;86:728-736] and Ryan et al. (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 identified three independent predictors of CLD: gestation, birth weight, and RFS. Applying a modified RFS score (to include gestation and birth weight), the RFS, Sinkin, and Ryan models were compared among high-risk infants admitted to Wilford Hall from January 1993-December 1995, and to Crawford Long Hospital (Atlanta, GA) from January 1993-December 1994. Predictive values, sensitivity, specificity, and receiver operating characteristic (ROC) curves were determined for the primary outcome variable: CLD at 36 weeks of corrected gestation. Of 248 infants at <32 weeks admitted to Wilford Hall, 220 survived >7 days. Thirty of 31 (97%) infants diagnosed with CLD were <29 weeks or < or =1,000 g at birth. Despite important demographic 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 area under the ROC curve for prediction of neonatal CLD in the groups as a whole. Application of the RFS method for early prediction of neonatal CLD at age 72 h should improve patient selection for early prevention trials.