Background/Aims EEG is the gold standard for the identification of neonatal seizures as the vast majority of electrographic seizures do not have a clinical correlate. Both under and over diagnosis of seizures is common in the neonatal intensive care unit (NICU). Computer assisted methods of interpreting the EEG have the potential to improve the accuracy of seizure detection. The aim of this study was to determine the clinical utility of our current neonatal seizure detection algorithm (NSDA).
Methods Multi-channel video-EEG recordings of 70 term neonates admitted to the NICU were analysed: 35 babies with seizure (mixed aetiologies) and 35 babies without seizure. The EEGs were annotated by an experienced neurophysiologist. The performance of the NSDA was assessed using time and event based metrics. An additional, clinically relevant, performance metric (based on the number of neonates correctly administered an anti-epileptic drug (AED) as early as possible after electrographic seizure onset) was calculated.
Results The sensitivity and specificity of the NSDA were 83% and 97% respectively when comparing to the experts annotation. The seizure detection rate and false alarm rate were 80% and 0.7/hr respectively. Thirty-four percent of neonates with seizures received an AED within the defined optimal timeframe, while 20% of neonates without seizure received an AED. These results were improved to 71% and 11%, respectively, by supplementing decision making with the output of the NSDA.
Conclusion Current NSDA performance, while not perfect, would greatly improve the efficacy of seizure detection and optimal AED administration in the NICU.