Prediction of Neurologic Morbidity in Extremely Low Birth Weight Infants

Abstract

OBJECTIVE: (1) Identify major determinants of adverse neurodevelopmental outcome in extremely low birth weight (ELBW) infants. (2) Compare neural networks and regression analysis in the prediction of major handicaps and Bayley scores (MDI and PDI) in individual ELBW neonates followed to 18 months.

DESIGN: Retrospective cohort study of regional tertiary care NICU database. A database with 21 selected variables was divided into training (n=144) and test sets (n=74). The training set was used to train a neural network and develop regression equations to predict outcomes in the test set.

RESULTS: Determinants (descending order of contribution to variance): Major handicap: intraventricular hemorrhage (IVH) grade, necrotizing enterocolitis≥stage II, black race, and no chorioamnionitis; low MDI: IVH grade, plurality, bronchopulmonary dysplasia (BPD), lower maternal grade, and no chorioamnionitis; low PDI: IVH grade, BPD, periventricular leukomalacia, lower maternal grade, and no chorioamnionitis. Regression techniques and neural networks were comparable and had relatively low sensitivity and correlation with adverse outcomes.

CONCLUSIONS: Much of the variance in ELBW neurologic outcome cannot be explained by either regression analysis or neural network approaches.