The ability to predict the occurrence of neonatal periventricular-intraventricular hemorrhage (PVH-IVH) would be useful in the design of clinical trials to prevent its occurrence. Therefore, data were collected from 463 consecutive infants less than or equal to 1500 g birth weight delivered between March 1, 1982 and February 28, 1985. This large population made it feasible to divide the infants into two groups, using one group to develop a model predictive of ICH and the second group to test the validity of the model. Infants were randomly grouped by sex, race, gestational age, birth weight, month of birth, mortality, and incidence and grade of worst PVH-IVH. In Group A (n = 232), respiratory distress syndrome, ventilator therapy, PaCO2 greater than or equal to 60 mmHg, PO2 less than or equal to 40 mmHg greater than or equal to 2 h, lower 1- and 5-min Apgar scores, lower pediatric estimation of gestational age, and pneumothorax were significantly associated with PVH-IVH by univariant analyses (chi 2, P less than 0.03). Multivariant discriminant analysis performed on Group A revealed that pneumothorax, cesarean section, PaCO2, and ventilator therapy were most predictive of PVH-IVH, but sensitivity was 55% and specificity 78%. Applying the model to Group B, sensitivity decreased to 21% while specificity rose to 93%. Logistic regression, which takes into account non-normally distributed variables, did not improve predictability.(ABSTRACT TRUNCATED AT 250 WORDS)