Accounting for Death in the Prediction of Post-intracerebral Hemorrhage Epilepsy Using Quantitative Electroencephalogram and Neuroimaging
Justin Wheelock1, Yilun Chen1, Jacob Garetti1, Haoqi Sun2, Jin Jing3, Sahar Zafar2, Aaron Struck4, Lawrence Hirsch1, Adithya Sivaraju1, Emily Gilmore1, M. Brandon Westover3, Jennifer Kim1
1Department of Neurology, Yale University, 2Department of Neurology, Massachusetts General Hospital, 3Department of Neurology, Beth Israel Deaconess Medical Center, 4Department of Neurology, University of Wisconsin Hospital and Clinics
Objective:
Determine whether incorporating quantitative EEG and neuroimaging features can simultaneously account for risk of epilepsy and death following intracerebral hemorrhage (ICH).
Background:
Hemorrhagic stroke has the highest rates, of all subtypes, of post-stroke epilepsy and mortality. As stroke severity increases, both the risk of epilepsy and mortality increases, yet few studies have disambiguated factors that may uniquely increase risk for each outcome.
Design/Methods:
We identified a retrospective cohort of adult ICH patients admitted to a tertiary care center between 2014-2023 who received continuous EEG monitoring. Of 238 included patients, 51 developed PICHE (unprovoked seizure >7 days post-ICH). We extracted total EEG power, rhythmicity, and epileptiform abnormalities (EA; Jing 2023). We obtained automated intraparenchymal (IPH), extra-axial (EAH), intraventricular (IVH) hemorrhage, and edema volumes from stability computed tomography (CT) scans (Monteiro 2020). We conducted proportional cause-specific hazard modeling to identify predictors of PICHE and death and forward selected features in a random survival forest (RSF). We trained and tested the RSF under a competing risk framework using five-fold cross-validation and evaluated model performance using area under the receiver operating curve (AUROC).
Results:
Features associated with PICHE included early seizures, cortical involvement of ICH, total EA, lateralized period discharges, lateralized rhythmic delta activity, power asymmetry, rhythmicity asymmetry, IPH volume, and edema volume. Features associated with death included age, midline shift, ICH score, generalized rhythmic delta activity, total power, total rhythmicity, and IPH, EAH, edema, and IVH volume. The final model including clinical features, quantitative EEG and neuroimaging predicted PICHE (AUC=0.744 [95CI: 0.581, 0.894]) and death (AUC=0.753 [95CI: 0.664, 0.922]).
Conclusions:
A competing risk model including clinical features, quantitative EEG, and neuroimaging predicts PICHE and death. Further work is needed to validate in a multicenter cohort. This model could be utilized to guide anti-epileptogenic treatment trial enrollment and patient counseling.
Disclaimer: Abstracts were not reviewed by Neurology® and do not reflect the views of Neurology® editors or staff.