Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

In Epilepsy & behavior : E&B

BACKGROUND : The goal of this cohort study was to estimate the predictors for ischemic stroke in patients with epilepsy in a large database containing data from general practitioners in Germany using machine learning methods.

METHODS : This retrospective cohort study included 11,466 patients aged ≥ 60 years with an initial diagnosis of epilepsy in 1182 general practices in Germany between January 2010 and December 2018 from the IQVIA Disease Analyzer database. The Sub-Population Optimization and Modeling Solutions (SOMS) tool was used to identify subgroups at a higher risk of stroke than the overall population with epilepsy based on 37 different variables.

RESULTS : A total of seven variables were considered important. Four co-diagnoses (diabetes, hypertension, heart failure, and alcohol dependence) were by far the strongest predictors with a combined predictive ability of more than 90%, whereby diabetes (41.4%) was the strongest predictor, followed by hypertension (35.0%) and heart failure (11.8%). The predictive importance of male gender was only 1.5%, and age was not recognized as an important predictor. Finally, the prescribed AEDs levetiracetam, with a predictive importance of 5.0%, and valproate, with 2.7%, were found to be weak predictors.

CONCLUSION : The stroke risk in patients with epilepsy was relatively high and could be predicted based on comorbidities such as diabetes mellitus, hypertension, heart failure, and alcohol dependence. Knowing and addressing these factors may help reduce the risk of stroke in patients with epilepsy.

Kostev Karel, Wu Tong, Wang Yue, Chaudhuri Kal, Tanislav Christian


Epilepsy, Machine learning, Prediction analytics, Stroke