Development and validation of a simple clinical prediction model for post-stroke seizures in critically ill patients: a multicenter retrospective cohort study
- 1Department of Neurosurgery, West China Hospital, Sichuan University
- 2West School of Nursing, Sichuan University
Abstract
Background: Post-stroke seizures (PSS) are common in critically ill stroke patients, but ICU-specific risk prediction tools are lacking, and the independent impact of PSS on mortality remains controversial.
Methods: This retrospective multicenter cohort included 6,586 adult stroke patients from 208 US ICUs in the Collaborative Research Database (eICU-CRD). Patients were randomly divided 7:3 into training and internal validation sets. The primary endpoint was coded in-hospital post-stroke seizures identified via diagnosis codes during the index hospitalization. Candidate predictors were screened using logistic regression and least absolute shrinkage and selection operator (LASSO) regression. Model discrimination, calibration, and clinical utility were, respectively, evaluated using the area under the receiver operating characteristic curve (AUROC), calibration slopes and calibration plots, and decision curve analysis. Propensity score matching (PSM) was used to evaluate the association between PSS and in-hospital mortality and length of stay.
Results: The overall incidence of coded in-hospital PSS was 7.5%. Head-to-head comparison identified SOFA as the best-performing routine ICU severity score for coded PSS prediction, outperforming GCS, OASIS and APACHE II. Multivariate logistic regression analysis identified five independent clinical predictors for coded in-hospital PSS: younger age, elevated admission body temperature, pre-existing hypertension, lower mean corpuscular hemoglobin concentration, and lower Glasgow Coma Scale (GCS) score. The model achieved AUCs of 0.702 in the training set and 0.713 in the validation set, with stable performance across age, gender, and stroke subtype subgroups. A three-tier bedside risk stratification was derived, with observed coded PSS rates of 3.4, 6.8, and 16.2% for low-risk, intermediate-risk, and high-risk groups. After PSM, coded PSS was not associated with increased in-hospital mortality (adjusted OR = 0.84, p = 0.295) or prolonged hospital stay (mean difference = −0.30 days, p = 0.167).
Conclusion: We developed and internally validated a simple prediction model for coded in-hospital PSS in critically ill stroke patients. Coded PSS itself was not independently associated with increased mortality. This model remains a research tool that has not undergone external validation and cannot yet guide real-world clinical monitoring or pharmacological intervention. These findings are applicable to clinically evident, coded convulsive seizures and may not generalize to non-convulsive seizure phenotypes.
Related
- Machine Learning–Based Prediction of Major Postoperative Complications After Noncardiac Surgery Under General Anesthesia Using Routinely Available Electronic Health Record Data: Retrospective Multicenter Development and Validation Study
- Feature-token clinical transformer for 24-h landmark mortality risk stratification in critically ill patients with heart failure: development and multicenter external validation
- Development and Validation of Temperature Trajectory-Based Phenotypes in Pediatric Sepsis: A Multicenter Retrospective Cohort Study, 2012–2018
- Development and internal validation of an exploratory clinical prediction model for MRI-defined white matter injury in moderate-to-late preterm infants: a single-center retrospective study
- Development and external validation of machine learning models for predicting mortality in patients with lung cancer: a multicenter retrospective cohort study
BibTeX
@article{zhang2026development,
title = {{Development and validation of a simple clinical prediction model for post-stroke seizures in critically ill patients: a multicenter retrospective cohort study}},
author = {Yue Zhang and Xiaocui Yang and Yu Kang and Wenyao Cui},
journal = {Frontiers in Neurology},
year = {2026},
month = oct,
doi = {10.3389/fneur.2026.1949946},
url = {https://www.frontiersin.org/articles/10.3389/fneur.2026.1949946/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Development and validation of a simple clinical prediction model for post-stroke seizures in critically ill patients: a multicenter retrospective cohort study
AU - Yue Zhang
AU - Xiaocui Yang
AU - Yu Kang
AU - Wenyao Cui
PY - 2026
DA - 2026/10/13
JO - Frontiers in Neurology
DO - 10.3389/fneur.2026.1949946
UR - https://www.frontiersin.org/articles/10.3389/fneur.2026.1949946/full
AB - Background: Post-stroke seizures (PSS) are common in critically ill stroke patients, but ICU-specific risk prediction tools are lacking, and the independent impact of PSS on mortality remains controversial. Methods: This retrospective multicenter cohort included 6,586 adult stroke patients from 208 US ICUs in the Collaborative Research Database (eICU-CRD). Patients were randomly divided 7:3 into training and internal validation sets. The primary endpoint was coded in-hospital post-stroke seizures identified via diagnosis codes during the index hospitalization. Candidate predictors were screened using logistic regression and least absolute shrinkage and selection operator (LASSO) regression. Model discrimination, calibration, and clinical utility were, respectively, evaluated using the area under the receiver operating characteristic curve (AUROC), calibration slopes and calibration plots, and decision curve analysis. Propensity score matching (PSM) was used to evaluate the association between PSS and in-hospital mortality and length of stay. Results: The overall incidence of coded in-hospital PSS was 7.5%. Head-to-head comparison identified SOFA as the best-performing routine ICU severity score for coded PSS prediction, outperforming GCS, OASIS and APACHE II. Multivariate logistic regression analysis identified five independent clinical predictors for coded in-hospital PSS: younger age, elevated admission body temperature, pre-existing hypertension, lower mean corpuscular hemoglobin concentration, and lower Glasgow Coma Scale (GCS) score. The model achieved AUCs of 0.702 in the training set and 0.713 in the validation set, with stable performance across age, gender, and stroke subtype subgroups. A three-tier bedside risk stratification was derived, with observed coded PSS rates of 3.4, 6.8, and 16.2% for low-risk, intermediate-risk, and high-risk groups. After PSM, coded PSS was not associated with increased in-hospital mortality (adjusted OR = 0.84, p = 0.295) or prolonged hospital stay (mean difference = −0.30 days, p = 0.167). Conclusion: We developed and internally validated a simple prediction model for coded in-hospital PSS in critically ill stroke patients. Coded PSS itself was not independently associated with increased mortality. This model remains a research tool that has not undergone external validation and cannot yet guide real-world clinical monitoring or pharmacological intervention. These findings are applicable to clinically evident, coded convulsive seizures and may not generalize to non-convulsive seizure phenotypes.
SN - 1664-2295
ER -