Software Credentialed Access
2HELPS2B: an EEG-based risk score for seizure probability in hospitalized patients
Aaron F Struck , Berk Ustun , Andres A. Rodriguez Ruiz , Jong Woo Lee , Suzette M. LaRoche , Lawrence J. Hirsch , emily gilmore , Jan Vlachy , Hiba Haider , Cynthia Rudin , M. Brandon Westover
Published: July 9, 2026. Version: 1.0.0
When using this resource, please cite:
(show more options)
Struck, A. F., Ustun, B., Rodriguez Ruiz, A. A., Lee, J. W., LaRoche, S. M., Hirsch, L. J., gilmore, e., Vlachy, J., Haider, H., Rudin, C., & Westover, M. B. (2026). 2HELPS2B: an EEG-based risk score for seizure probability in hospitalized patients (version 1.0.0). Brain Data Science Platform. https://doi.org/10.60508/yzjf-5789.
Abstract
Importance. Continuous EEG detects seizures in hospitalized patients, but monitoring is resource-intensive; a simple risk score could target it.
Objective / Methods. Using 5,742 continuous-EEG records from the Critical Care EEG Monitoring Research Consortium, we identified EEG and clinical predictors of seizures and used a machine-learning method (RiskSLIM) to derive a simple integer point score.
Results. The resulting 2HELPS2B score sums points for: any IIC pattern frequency >2 Hz (2 points), epileptiform discharges, lateralized periodic discharges / lateralized rhythmic delta activity / bilateral independent periodic discharges, prior seizure, and brief (potentially ictal) rhythmic discharges. Seizure risk rose from ~5% (score 0) to >95% (score >=5), with cross-validated AUC ~0.82.
Conclusion. 2HELPS2B stratifies seizure risk on continuous EEG and can guide monitoring duration. This project releases the analysis code and de-identified data.
Code and de-identified CCEMRC data to reproduce the seizure-probability analyses.
Background
Continuous EEG detects nonconvulsive seizures in critically ill patients but is resource-intensive. A simple validated risk score (2HELPS2B) can stratify seizure risk from initial EEG findings and guide monitoring duration.Software Description
MATLAB pipeline (a_step1..a_step4) + bundled glmnet_matlab and the de-identified analysis table CCEMRCDATA.mat (5,742 x 119 coded variables; no PHI). Figures of seizure probability by IIC pattern are included.Technical Implementation
From 5,742 continuous-EEG records (Critical Care EEG Monitoring Research Consortium), IIC-pattern indicators (LPD/GPD/LRDA/GRDA/BiRDs), pattern frequency, prior seizures, and brief rhythmic discharges were related to seizure occurrence using LASSO/logistic modeling with bootstrap AUC, and an interpretable integer score was derived with RiskSLIM.Installation and Requirements
MATLAB R2016+ with the Statistics and Machine Learning Toolbox; add glmnet_matlab to the path. No further installation. See REPRODUCE.md and DATA_SOURCE.md.Usage Notes
Reproduce with a_step2/a_step3 after addpath('glmnet_matlab'); a quick check reproduces the paper's Table-1 seizure proportions from CCEMRCDATA.mat. RiskSLIM (Ustun & Rudin) produces the final integer score.Release Notes
First public release: code + de-identified data + reproduction docs.Ethics
De-identified multicenter consortium data; collected under IRB approval at participating centers.Acknowledgements
Data from the Critical Care EEG Monitoring Research Consortium (CCEMRC).Conflicts of Interest
See the associated publication (JAMA Neurol 2017;74:1419-1424).Access
Access Policy:
Only credentialed users who sign the DUA can access the files.
License (for files):
BDSP Credentialed Health Data License 1.5.0
Data Use Agreement:
BDSP Credentialed Health Data Use Agreement
Required training:
Discovery
Corresponding Author
Files
- be a credentialed user
- sign the data use agreement for the project