Stroke is one of the most common causes of adult-onset seizures, which may occur early (≤7 days) or late (>7 days) after stroke.¹ Post-stroke epilepsy (PSE) is defined as any unprovoked seizure after the acute phase (>7 days) of stroke.¹ Approximately 4% of stroke patients experience PSE in the first year and 8% in the five years after stroke.¹ While the precise epileptogenic mechanism is not fully understood, research has indicated that prompt intervention to interrupt the epileptogenic process may be beneficial for high-risk patients.¹ However, the current evidence does not support the routine use of anti-seizure medications for primary seizure prevention following an acute stroke.¹ Risk stratification models are thus necessary to identify patients at risk of PSE, such as the SeLECT score, which is calculated based on stroke severity, large-artery atherosclerosis, early seizure (≤7 days), cortical involvement and middle cerebral artery (MCA) territory.¹ Due to the increasing availability and accessibility of neuroimaging techniques, including magnetic resonance imaging (MRI) and computed tomography (CT), researchers from the Queen Mary Hospital, Hong Kong, sought to identify new imaging parameters that can improve the prediction of PSE.¹ At the AAN 2024 Annual Meeting, Dr. Leung, Chun-Yin William from the hospital shared with us the identification of H-SeLECT, the novel MRI-based risk scoring system proposed to increase the predictive value in detecting PSE.¹

In this single-center, observational study, patients in the University of Hong Kong (HKU) Stroke Registry aged ≥18 years and admitted for first-ever acute stroke between 2018 and 2020 were recruited.¹ Patients were excluded if they had a transient ischemic attack or hemorrhagic stroke, previous history of stroke or epilepsy, any concurrent use of anti-seizure medications for seizure prevention or other diseases, other intracranial structural abnormalities, or no identifiable infarct on neuroimaging after reperfusion therapy.¹ The final cohort for analysis consisted of 852 patients, of which 100% received a brain CT scan, while only 70.1% had an MRI scan.¹ Seizures occurred in 50 (5.9%) of patients, of which 13 (1.5%) patients had an early seizure (≤7 days after insult) and 37 (4.3%) had a late seizure (>7 days after insult).¹

Univariate analysis with competing risks analysis was performed, with all-cause mortality considered as a competing risk event.¹ Most predictors in the SeLECT score were found to be significantly associated with higher PSE risk, including severe stroke (National Institute of Health Stroke Scale [NIHSS] ≥11; subdistribution hazard ratio [sHR]=3.79; p=0.001), MCA territory involvement (sHR=4.26; p=0.002), cortical involvement (sHR=6.18; p<0.001) and early seizures ≤7 days after stroke (sHR=29.86; p<0.001).¹ Additionally, new predictors of PSE were identified which consisted of the presence of hemorrhagic transformation regardless of grade (sHR=6.99; p<0.001) and large infarct size (≥5cm or ≥1/3 MCA territory; sHR=10.55; p<0.001).¹ Sex- and age-adjusted analyses also found similar results.¹ Multivariate analysis further showed that only hemorrhagic transformation (all grades) (sHR=4.72; 95% CI: 2.36-9.41; p<0.001) and early seizures ≤7 days after stroke (sHR=27.60; 95% CI: 10.61-71.78; p<0.001) were independent risk factors of PSE.¹

Given that hemorrhagic transformation was independently associated with PSE, the researchers decided to combine it with the SeLECT score to further improve its predictive value.¹ Similar to the original methodology used to derive the SeLECT score, the weighting for hemorrhagic transformation was calculated by dividing its sHR (6.99) by the median of the lowest 3 values (3.43) in the univariate analysis and rounded to the nearest integer.¹ The presence of any hemorrhagic transformation would add 2 points in the H-SeLECT scoring system.¹ H-SeLECT showed improved predictive performance for PSE with a significant increase in the area-under-curve (AUC) of the receiver operating characteristic (ROC) curve compared to SeLECT (0.813 vs. 0.774; Z=2.67; p=0.008).¹ The best cutoff of H-SeLECT was found to be 4.¹

In summary, the H-SeLECT risk score utilizes objectively measured, well-defined and routinely available clinical predictors to enhance the prediction of PSE.¹ Given that having early seizures after stroke is also an independent risk factor of PSE, close monitoring for seizure activity during the acute phase of stroke is warranted.¹ Further exploration of a multi-step risk stratification approach incorporating electroencephalogram (EEG), serum and radiological biomarkers, as well as a personalized risk-based strategy to guide primary or secondary seizure prophylaxis following stroke may also be beneficial.¹