Additive Benefit of Neuroimaging and Electroencephalography in Predicting Post-ischemic-stroke Epilepsy
Abstract number :
2.066
Submission category :
3. Neurophysiology / 3G. Computational Analysis & Modeling of EEG
Year :
2022
Submission ID :
2204512
Source :
www.aesnet.org
Presentation date :
12/4/2022 12:00:00 PM
Published date :
Nov 22, 2022, 05:25 AM
Authors :
Yilun Chen, MS – Yale University; Alexandria Soto, BS – Yale School of Medicine; Tejaswi Sudhakar, MBS – Yale School of Medicine; Adeel Zubair, MD – Yale School of Medicine; Lucas Loman, Currently Undergraduate Student at Yale University – Yale School of Medicine; Adithya Sivaraju, MD – Yale School of Medicine; Emily Gilmore, MD – Yale School of Medicine; Nils Petersen, MD/PhD – Yale School of Medicine; Lawrence Hirsch, MD – Yale School of Medicine; Hal Blumenfeld, MD/PhD – Yale School of Medicine; Sahar Zafar, MD – Massachusetts General Hospital; Aaron Struck, MD – University of Wisconsin Hospitals; Nishant Mishra, MD/PhD – Yale School of Medicine; Kevin Sheth, MD – Yale School of Medicine; Michael Westover, MD/PhD – Massachusetts General Hospital; Jennifer Kim, MD/PhD – Yale School of Medicine
This abstract has been invited to present during the Neurophysiology platform session.
Rationale: Post-Ischemic-Stroke Epilepsy (PISE) is a serious complication of ischemic stroke. Admission National Institutes of Health Stroke Scale (NIHSS) and early seizures are established predictors of PISE. Whether neuroimaging and electroencephalography (EEG) characterization provide additional predictive value is unknown. We aimed to evaluate if neuroimaging and EEG information improved PISE prediction.
Methods: We retrospectively reviewed medical records of ischemic stroke patients at Yale New Haven Hospital who were ≥ 18 years of age, had post-stroke neuroimaging, EEG data, and either ≥ 1 year seizure-free follow-up (i.e., non-PISE) or follow-up documenting any unprovoked seizures 7-365 days post-stroke (i.e., PISE). Patients with preceding seizure history or recent (< 5 years) brain injury were excluded. We recorded covariates at admission including age, sex, initial NIHSS, pre-morbid modified Rankin Scale (mRS), and clinical seizure at presentation. Neuroimaging variables included infarct volume and hemorrhagic conversion. EEG features included epileptiform abnormality incidence/burden (EA, i.e., electrographic seizures, sporadic/periodic epileptiform discharges, lateralized rhythmic delta activity), suppression, global and asymmetric rhythmicity and power spectra. We also evaluated NIHSS and mRS scores at discharge. We trained random forest models using leave-one-out cross validation technique to predict PISE. Prediction models were evaluated by Area Under the receiver operating characteristic Curve (AUC).
Results: We included 81 ischemic stroke patients (PISE n [%], 18 [22%]; age median [IQR], 65 [52-75]; female n [%], 40 [49%]; Table 1, Figure 1A). PISE and non-PISE patients had comparable initial NIHSS, admission clinical seizure incidence, and pre-morbid mRS (Table 1). Compared to non-PISE patients, PISE patients had larger infarct volume (median [IQR], 128 [43-196] vs. 16 [4-41], p=0.001), higher EA incidence (67% vs. 29%, p=0.007) and burden (i.e., % of recording with any EA subtype presence, median [IQR], 4.2 [1.7-10.1] vs 1.1 [0.3-5], p=0.012), and greater hemispheric asymmetry of rhythmicity and power spectra (Table 1). Predictive models using admission variables showed poor ability to predict PISE (AUC [95% CI], 0.50 [0.32-0.68]; Figure 1B). However, models combining admission variables with either neuroimaging (AUC [95% CI], 0.69 [0.55-0.83]) or EEG features (AUC [95% CI], 0.68 [0.54-0.82]) significantly outperformed those using admission variables alone (Figure 1C). Although the model based on discharge NIHSS and mRS showed good performance (AUC [95% CI], 0.70 [0.56-0.86]; Figure 1C), it did not outperform neuroimaging or EEG based models. When five of the most important variables (discharge NIHSS, rhythmic delta asymmetry, theta power asymmetry, infarct volume, and EA abnormality) were included, the AUC improved to 0.81 (95% CI, 0.67-0.94; Figure 1D).
Conclusions: In ischemic stroke patients with similar admission characteristics, quantitative neuroimaging and EEG features may be valuable biomarkers for predicting PISE. These findings warrant further validation in a larger prospective cohort.
Funding: National Institute of Neurological Disorders and Stroke
Neurophysiology