Validation of the 2HELPS2B Seizure Risk Score in Acute Brain Injury Patients.

BACKGROUND AND OBJECTIVE: Seizures are common after traumatic brain injury (TBI), aneurysmal subarachnoid hemorrhage (aSAH), subdural hematoma (SDH), and non-traumatic intraparenchymal hemorrhage (IPH)-collectively defined herein as acute brain injury (ABI). Most seizures in ABI are subclinical, meaning that they are only detectable with EEG. A method is required to identify patients at greatest risk of seizures and thereby in need of prolonged continuous EEG monitoring. 2HELPS2B is a simple point system developed to address this need. 2HELPS2B estimates seizure risk for hospitalized patients using five EEG findings and one clinical finding (pre-EEG seizure). The initial 2HELPS2B study did not specifically assess the ABI subpopulation. In this study, we aim to validate the 2HELPS2B score in ABI and determine its relative predictive accuracy compared to a broader set of clinical and electrographic factors. METHODS: We queried the Critical Care EEG Monitoring Research Consortium database for ABI patients age ≥ 18 with > 6 h of continuous EEG monitoring; data were collected between February 2013 and November 2018. The primary outcome was electrographic seizure. Clinical factors considered were age, coma, encephalopathy, ABI subtype, and acute suspected or confirmed pre-EEG clinical seizure. Electrographic factors included 18 EEG findings. Predictive accuracy was assessed using a machine-learning paradigm with area under the receiver operator characteristic (ROC) curve as the primary outcome metric. Three models (clinical factors alone, EEG factors alone, EEG and clinical factors combined) were generated using elastic-net logistic regression. Models were compared to each other and to the 2HELPS2B model. All models were evaluated by calculating the area under the curve (AUC) of a ROC analysis and then compared using permutation testing of AUC with bootstrapping to generate confidence intervals. RESULTS: A total of 1528 ABI patients were included. Total seizure incidence was 13.9%. Seizure incidence among ABI subtype varied: IPH 17.2%, SDH 19.1%, aSAH 7.6%, TBI 9.2%. Age ≥ 65 (p = 0.015) and pre-cEEG acute clinical seizure (p < 0.001) positively affected seizure incidence. Clinical factors AUC = 0.65 [95% CI 0.60-0.71], EEG factors AUC = 0.82 [95% CI 0.77-0.87], and EEG and clinical factors combined AUC = 0.84 [95% CI 0.80-0.88]. 2HELPS2B AUC = 0.81 [95% CI 0.76-0.85]. The 2HELPS2B AUC did not differ from EEG factors (p = 0.51), or EEG and clinical factors combined (p = 0.23), but was superior to clinical factors alone (p < 0.001). CONCLUSIONS: Accurate seizure risk forecasting in ABI requires the assessment of EEG markers of pathologic electro-cerebral activity (e.g., sporadic epileptiform discharges and lateralized periodic discharges). The 2HELPS2B score is a reliable and simple method to quantify these EEG findings and their associated risk of seizure.

Comparison of machine learning models for seizure prediction in hospitalized patients.

OBJECTIVE: To compare machine learning methods for predicting inpatient seizures risk and determine the feasibility of 1-h screening EEG to identify low-risk patients (<5% seizures risk in 48 h). METHODS: The Critical Care EEG Monitoring Research Consortium (CCEMRC) multicenter database contains 7716 continuous EEGs (cEEG). Neural networks (NN), elastic net logistic regression (EN), and sparse linear integer model (RiskSLIM) were trained to predict seizures. RiskSLIM was used previously to generate 2HELPS2B model of seizure predictions. Data were divided into training (60% for model fitting) and evaluation (40% for model evaluation) cohorts. Performance was measured using area under the receiver operating curve (AUC), mean risk calibration (CAL), and negative predictive value (NPV). A secondary analysis was performed using Monte Carlo simulation (MCS) to normalize all EEG recordings to 48 h and use only the first hour of EEG as a "screening EEG" to generate predictions. RESULTS: RiskSLIM recreated the 2HELPS2B model. All models had comparable AUC: evaluation cohort (NN: 0.85, EN: 0.84, 2HELPS2B: 0.83) and MCS (NN: 0.82, EN; 0.82, 2HELPS2B: 0.81) and NPV (absence of seizures in the group that the models predicted to be low risk): evaluation cohort (NN: 97%, EN: 97%, 2HELPS2B: 97%) and MCS (NN: 97%, EN: 99%, 2HELPS2B: 97%). 2HELPS2B model was able to identify the largest proportion of low-risk patients. INTERPRETATION: For seizure risk stratification of hospitalized patients, the RiskSLIM generated 2HELPS2B model compares favorably to the complex NN and EN generated models. 2HELPS2B is able to accurately and quickly identify low-risk patients with only a 1-h screening EEG.