Abstract Details

Title Multimodal Prognostic Model to Predict Post-Stroke Epilepsy

Topic Epilepsy/Clinical Neurophysiology (EEG)

Presentation(s) P7 - Poster Session 7 (5:00 PM-6:00 PM)

Poster/Presentation
Number 9-004

Objective

To determine if adding acute EEG findings to the SeLECT score variables improves post-stroke epilepsy (PSE) predictive models.

Background

Stroke is the leading cause of epilepsy in adults over 60.

The current model for predicting PSE, the SeLECT score, includes 5 predictors: Stroke Severity (NIHSS), Large-artery atherosclerosis etiology, Early seizure, Cortical involvement, and MCA Territory. Previous work from Cleveland Clinic shows that after adjusting for SeLECT variables, acute EEG findings (electrographic seizures or epileptiform abnormalities) remain independent predictors of PSE development. However, no existing predictive model includes EEG findings along with clinical/neuroimaging data.

Design/Methods

In this single-center, retrospective chart review, we examined four models: “SeLECT”, “Refit SeLECT”, “Full” (SeLECT plus EEG variables), and “Reduced” (using backward variable selection). For each model, we compared discrimination using the concordance index (C-index) and examined model calibration graphically. Using the reduced model, we created a new score.

Results

375 patients were included and 66 (18%) had PSE. Patients with PSE had worse NIHSS, higher SeLECT score, and higher likelihood of early seizure, clinical status, electrographic seizure, electrographic status, lateralized periodic discharges (LPDs), and sharp waves.

In both the Full and Reduced models, only “early seizure” and “LPDs” were statistically significant predictors. The Reduced model therefore included the SeLECT variables plus LPDs. The c-index for the SeLECT, Refit, Full, and Reduced models were 0.669, 0.679, 0.694, and 0.698, respectively. Calibration was good for all models but best for the refit SeLECT and reduced models.

Conclusions

When compared to the SeLECT model, the C-indexes for the Full and Reduced models were slightly improved. The only statistically significant variables in these models were “early seizure” and “LPDs”. While this data supports the addition of LPDs to PSE models, our study is underpowered, and a larger, multicenter dataset is needed to confirm the role of acute epileptiform findings in PSE prognostic models.

Authors/Disclosures
Vijaya R. Dasari (Cleveland Clinic Neurology) PRESENTER	Miss Dasari has nothing to disclose.
Nicholas Thompson	The institution of Nicholas Thompson has received research support from EMD Serono.
Vineet Punia, MD (Cleveland Clinic)	Dr. Punia has nothing to disclose.

��ɫ��

��ɫ��