A simple five-predictor model (younger age, higher admission temperature, pre-existing hypertension, lower mean corpuscular hemoglobin concentration, lower GCS) predicted coded in-hospital post-stroke seizures in critically ill stroke patients with modest discrimination (AUC 0.702 training, 0.713 validation), and coded PSS was not independently associated with increased in-hospital mortality or length of stay after propensity score matching.
The overall incidence of coded in-hospital post-stroke seizures among 6,586 critically ill stroke patients in the eICU-CRD was 7.5%.
SOFA was the best-performing routine ICU severity score for predicting coded PSS in a head-to-head comparison.
Five independent clinical predictors of coded in-hospital PSS were identified.
The prediction model showed modest discrimination that was similar in training and internal validation sets.
A three-tier bedside risk stratification separated patients into groups with markedly different observed coded PSS rates.
After propensity score matching, coded PSS was not associated with increased in-hospital mortality.
After propensity score matching, coded PSS was not associated with prolonged hospital stay.
The authors state the model is a research tool with limited applicability: it lacks external validation and applies to clinically evident coded convulsive seizures.
This study looked at about 6,600 stroke patients treated in intensive care units across 208 US hospitals to find out who is more likely to have a seizure during their hospital stay. About 7.5% had a seizure recorded in their diagnosis codes. The researchers built a simple prediction tool using five pieces of information: younger age, higher body temperature on admission, a history of high blood pressure, lower mean corpuscular hemoglobin concentration (a blood test value), and a lower Glasgow Coma Scale score (a measure of consciousness). Using these, patients were sorted into low-, intermediate- and high-risk groups, with seizure rates of about 3%, 7% and 16% respectively. The tool's ability to tell patients with and without seizures apart was only moderate (AUC about 0.70–0.71), and it was tested only on a random subset of the same database. It has not been checked in other hospitals or populations. The authors stress that it is a research tool and should not yet guide monitoring or treatment decisions. It also covers only clearly visible, coded convulsive seizures, and may not apply to seizures without obvious convulsions. The study also compared similar patients with and without recorded seizures. After matching, seizures were not linked to a higher risk of dying in the hospital or to a longer hospital stay. This research suggests that seizures recorded in this setting may not independently worsen these outcomes, though this finding comes from a retrospective database and cannot prove cause and effect.
Zhang Y, Yang X, Kang Y, Cui W. (2026). Development and validation of a simple clinical prediction model for post-stroke seizures in critically ill patients: a multicenter retrospective cohort study.. Frontiers in neurology. https://doi.org/10.3389/fneur.2026.1949946