We analyzed data from a cohort of all ELBW infants (401–1000 g) who were born between 1/1/1998 and 1/1/2006 and admitted to the NICU at Cincinnati Children’s Hospital Medical Center, University Hospital, or Good Samaritan Hospital in Cincinnati, Ohio. Data was initially collected on these infants as part of the NICHD Neonatal Research Network Generic Database, and the infants were prospectively followed to 18–22 months corrected age as part of the Neonatal Research Network Follow Up study. For the current retrospective analysis, institutional review board (IRB) approval was obtained from each hospital. Infants were included if they were coded as having at least one abnormal head ultrasound scan in the Generic Database. Abnormal head ultrasound was defined as germinal matrix hemorrhage, intraventricular hemorrhage, ventricular dilation, cystic area in the parenchyma, cystic periventricular leukomalacia, and/or porencephalic cyst. The protocol for head ultrasound screening at the time of the study included a first head ultrasound at 7–10 days of life and a ssecond head ultrasound after 28 days of life. The most severe head ultrasound finding was used to categorize the infants. Infants with lethal congenital malformations, chromosomal abnormalities, history of meningitis, and periventricular leukomalacia were excluded.
After the infants meeting the above criteria were identified by their NICHD NRN number, the data was re-identified to find the infants’ names, dates of birth, and medical record numbers. Using this information, all of the head ultrasound reports for each infant were obtained. The reports were reviewed and data on laterality and highest grade of IVH was entered into a new computerized database. This data was then matched with the infant’s data from the Generic Database on potential confounders and from the Follow-up Database on the infant’s neurodevelopmental outcome at 18–22 months. Information extracted from the Follow-up Database included Bayley Scales of Infant Development Second Edition Mental Development Index (BSID II MDI), Psychomotor Development Index (PDI), presence of cerebral palsy, blindness, and hearing impairment.
The primary outcome was neurodevelopmental impairment at 18–22 months corrected age. Neurodevelopmental impairment was defined as the presence of any of the following: cerebral palsy, MDI < 70, PDI < 70, blindness, or hearing impairment. Secondary outcomes were the BSID II MDI score and BSID II PDI score. The key predictor variables were IVH grade and laterality of IVH (unilateral versus bilateral). Potential confounder variables were gender, race, birth weight, presence of bronchopulmonary dysplasia (BPD), postnatal steroids, early or late culture-positive sepsis, and necrotizing enterocolitis (NEC) requiring surgery. These variables have been reported to be associated with poor neurodevelopmental outcome in prior studies.
A logistic regression model was developed using the key independent variables (IVH grade and laterality) and the potential confounder variables as covariates to predict the binary outcome of combined neurodevelopmental impairment. Firth’s penalized maximum likelihood estimation was used to reduce bias in the parameter estimates and to address issues pertaining to cells having zero frequency. The Hosmer-Lemeshow goodness-of-fit test was used to evaluate overall model fit. Regardless of statistical significance, laterality (unilateral vs. bilateral IVH) remained in the models, as it was directly related to the study hypothesis. Interaction between laterality and IVH grade was tested in the model to determine if the relationship between IVH grade and neurodevelopmental outcome was modified by whether an infant had unilateral or bilateral IVH. The interaction term was not statistically significant in any of the models, and therefore this term did not remain in the final models. General linear regression models were developed using the key independent variables and the potential confounder variables to estimate the independent relationship between IVH grade with MDI and PDI scores. These models provided mean MDI and PDI scores (using least square means) with 95% confidence intervals (CI), adjusted for the confounders included in the model. A backward elimination strategy was used with p > 0.1 for exit criteria. Regression diagnostics were used to examine outliers and normality of residuals. Statistical analysis was performed using SAS (Version 9.2, SAS Institute Inc, Cary, NC).