727 - Enhancing Prediction of Neurodevelopmental Outcomes in Extremely Premature Infants

Publication Number: 2709.727

Rachel A. Fazenbaker, University of Rochester, Golisano Children's Hospital, Rochester, NY, United States; Joseph M. Bliss, University of Rochester Medical Center, Rochester, NY, United States; Andrew M. Dylag, University of Rochester, Golisano Children’s Hospital, Rochester, NY, United States; Hongyue Wang, University of Rochester School of Medicine and Dentistry, Rochester, NY, United States

Background: Models predicting neurodevelopmental impairment (NDI) in preterm infants are limited because they are primarily focused on the presence or absence of morbidities (eg. necrotizing enterocolitis, bronchopulmonary dysplasia, sepsis) but do not account for the severity of morbidities.

Objective: To develop a prediction model for death and NDI at specific timepoints during the birth hospitalization utilizing a combination of diagnoses, laboratory, ventilatory, and vital sign data, comparing results against the Neonatal Research Network (NRN) NDI calculator.

Design/Methods: We performed a retrospective study of infants < 28 weeks gestation at a single-center level IV NICU. Laboratory, ventilatory, vital sign, diagnoses, medications, and feeding data were abstracted from the electronic medical record based on their established or hypothesized links to NDI. Data were collected at four timepoints: postnatal day (PND) 0, PND7, PND28, and 36 weeks corrected gestational age (CGA). The primary outcome was a composite of death before discharge or NDI at 22-26 months corrected age. Secondary analyses stratified NDI into mild, moderate, and severe. Prediction models were generated using multivariate stepwise logistic regression, and the area under the receiver operating characteristic curve (AUROC) was compared against the NRN NDI estimator using the DeLong Test.

Results: Out of the starting cohort (n=431, mean gestational age (GA) 25 weeks, mean birth weight 735g), clinical and primary outcome data were available for 181 infants at PND0, and 167 infants at PND7. For PND0, GA and chorioamnionitis were negatively associated with Death/NDI, while lactate and pCO2 were positively associated (Table 1). This novel "revised" PND0 model significantly improved outcome prediction compared with the NRN calculator (AUROC 0.755 vs 0.713, p=0.046, DeLong Test). Higher blood glucose and lactate were associated with worse NDI severity (Table 1). At PND7, lower levels of respiratory support and more days on conventional mechanical ventilation (CMV) were associated with improved NDI/Death (Table 2), but the "revised" model had similar performance and fewer variables. NDI severity, however, was significantly worse with number of days on high-frequency ventilation (HFV) and showed a strong trend with the lowest pH (Table 2). Analyses at PND28 and 36 weeks are underway.

Conclusion(s): Utilizing more granular clinical data early in the neonatal hospitalization can improve the prediction of NDI/Death and NDI severity in extremely premature infants. These advances may improve counseling of families and support shared decision making.

PND0 Models
AUC=area under the curve, CI=confidence interval

PND7 Models
AUC=area under the curve, CI=confidence interval, CPAP=continuous positive airway pressure, IVH=intraventricular hemorrhage, CMV=conventional mechanical ventilation, HFV=high-frequency ventilation
*Not statistically significant, but incorporation in model improved Nagelkerke R2 by 19%