Early Prediction of Infants with Seizures in Hypoxic Ischaemic Encephalopathy
Abstract number :
V.022
Submission category :
3. Neurophysiology / 3B. ICU EEG
Year :
2021
Submission ID :
1826538
Source :
www.aesnet.org
Presentation date :
12/9/2021 12:00:00 PM
Published date :
Nov 22, 2021, 06:55 AM
Authors :
Andreea Pavel, MD - University College Cork and INFANT Research Centre, Ireland; John O’Toole - INFANT Centre, University College Cork, Ireland; Jacopo Proietti - INFANT Centre, University College Cork, Ireland; Janet Rennie - Institute for Women’s Health, University College London, London, UK; Linda de Vries - Utrecht Brain Center, University Medical Center Utrecht, Utrecht University, The Netherlands; Mats Blennow - Department of Neonatal Medicine, Karolinska University Hospital and Division of Paediatrics, Department CLINTEC, Karolinska Institutet, Stockholm, Sweden; Adrienne Foran - Rotunda Hospital, Dublin, Ireland; Olga Kapellou - Homerton University Hospital NHS Foundation Trust, London, UK; Vicki Livingstone - INFANT Centre, University College Cork, Ireland; Ronit Pressler - Department of Clinical Neurophysiology, Great Ormond Street Hospital for Children NHS Trust, London, UK; Divyen Shah - Royal London Hospital and Queen Mary University of London, London, UK; William Marnane - INFANT Centre, University College Cork, Ireland; Eugene Dempsey - INFANT Centre, University College Cork, Ireland; Deirdre Murray - INFANT Centre, University College Cork, Ireland; Geraldine Boylan - INFANT Centre, University College Cork, Ireland
Rationale: Hypoxic-ischaemic encephalopathy (HIE) is the leading cause of seizures in full term newborns. An infant’s condition at birth does not always correlate well with encephalopathy severity or the risk for seizures, which presents a real challenge for early, appropriate management. Conventional electroencephalographic (EEG) monitoring is being used increasingly in neonatal care settings worldwide and may support the early identification of infants at highest risk of seizure. Our aim was to assess if early clinical and EEG features could accurately predict later development of seizures in infants with HIE.
Methods: This is a secondary analysis of data from two multicentre cohort studies of infants with HIE and early EEG monitoring. Clinical parameters within 6 hours of birth and EEG recordings within 12 hours of birth and before seizure onset were used to predict infants who later develop seizures. Clinical parameters used: gestational age, birth weight, mode of delivery, suspected fetal distress, intrapartum complications, Apgar scores at 1 and 5 minutes, and assisted ventilation at 10 minutes. The earliest hour of EEG recording was visually assed by two experts in neonatal electroencephalography (blinded to the seizure status of each infant) and the presence of the following features was annotated: discontinuity, low voltage to isoelectric, asymmetry/asynchrony, sleep-wake cycles (qualitative analysis). Same EEG epochs was used to extract power, discontinuity, spectral distribution and inter-hemispheric connectivity features (quantitative analysis). Machine-learning models (using random forest and radiant boosting algorithms) to predict infants with seizures were developed separately using clinical parameters as well as qualitative and quantitative EEG parameters. We tested the performance of each model individually and combined.
Results: The analysis included 164 infants with HIE: 109 infants without electrographic seizures and 55 infants with seizures. The prediction values for each model were: clinical model Area Under the Curve (AUC) 0.687, sensitivity 63.6%, specificity 72.5%, positive predictive value (PPV) 53.8%, negative predictive value (NPV) 79.8%; qualitative EEG model AUC 0.713, sensitivity 67.3%, specificity 79.8%, PPV 62.7%, NPV 82.9%; quantitative EEG model AUC 0.757, sensitivity 67.3%, specificity 80.7%, PPV 63.8%, NPV 83.0%. The predictive ability of the combined model for clinical plus qualitative EEG had an AUC 0.758, sensitivity 78.2%, specificity 78.0%, PPV 64.2%, NPV 87.6%; clinical plus quantitative EEG had an AUC 0.770, sensitivity 81.8%, specificity 73.4%, PPV 60.8%, NPV 88.9%.
Conclusions: We have shown that seizure prediction models incorporating EEG parameters in infants with HIE have a higher predictive ability than clinical parameters alone. Importantly, we have also shown that objective quantitative EEG analysis is as predictive as an expert EEG interpretation (qualitative analysis). This may prove very useful for the development of automated seizure prediction algorithms for infants with HIE.
Funding: Please list any funding that was received in support of this abstract.: Strategic Translational Award and an Innovator Award from the Wellcome Trust (098983 & 209325) and Science Foundation Ireland (12/RC/2272).
Neurophysiology