These neonatal seizure prediction models were developed by the Helbig lab and the TRiG team at the Children’s Hospital of Philadelphia. We used data from standardized EEG templates reporting on 1117 neonates, 150 with HIE, undergoing continuous EEG monitoring.
The models used for these predictions are distributed random forest models generated using the H2O.ai platform in RStudio. The models were stratified and weighted in order to proportionally distribute points to “non-subsequent seizure” (x0.5) and “subsequent seizure” (x4.0) instances. The model was optimized towards maximum AUCPR and cross-validation (k=10) was implemented within each model. Further details about the models can be found in our paper, McKee et al., 2022. Model performance metrics on our hold-out test datasets are displayed below.