A Practical Risk Score for EEG Seizures in Hospitalized Patients (S11.002)

Objective: To use seizure risk factors from continuous EEG monitoring (cEEG) to create a simple scoring system for predicting the probability of electrographic seizures in patients with acute illness. Background: cEEG use has expanded, motivated by the high incidence of subclinical seizures in hospitalized encephalopathic patients. EEG features reported as predictors of seizures include epileptiform discharges and periodic discharges. However, no study has examined how these predictors affect seizure risk jointly. Here we propose a simple scoring system for seizure risk that we refer to as the 2HELPS2B score. Design/Methods: We used a prospective database to derive a dataset containing 24 clinical and electrographic variables for 5427 >24hr cEEG sessions. Using this dataset, we created a scoring system model to estimate seizure risk in patients undergoing cEEG. The model was built using a new machine learning method (RiskSLIM) that is designed to produce accurate, risk-calibrated, scoring systems with a limited number of variables and small integer weights. We validated the predictive accuracy and risk-calibration of our model using cross-validation, and compared its performance to models built with state-of-the-art logistic regression methods. Results: Our final model (2HELPS2B) has an AUC of 0.821 and average calibration error of 2.7%. It includes 6 variables with the following point assignments: (i) brief potentially ictal rhythmic discharges (B(I)RDs) (2 points); (ii) presence of LPD, LRDA, or BIPDs (1 point); (iii) prior seizure (1 point); (iv) sporadic epileptiform discharges (1 point); (v) frequency >2.0Hz for any periodic or rhythmic pattern (1 point); (vi) presence of “plus” features (superimposed, rhythmic, sharp, or fast activity) (1 point). The predicted seizure risk of each score is: 0: 5%, 1: 12%, 2: 27%, 3: 50%, 4: 73%, 5: 88%, 6–7: >95%. Conclusions: The 2HELPS2B score allows accurate prediction of seizures by adding points for 6 readily available variables from the patient history and initial EEG. Study Supported by: This study was supported by a Research Infrastructure award from the American Epilepsy Society and the Epilepsy Foundation. Disclosure: Dr. Struck has nothing to disclose. Dr. Ustun has nothing to disclose. Dr. Rodriguez-Ruiz has nothing to disclose. Dr. Lee has nothing to disclose. Dr. LaRoche has received royalty, license fees, or contractual rights payments from Demos Publishing. Dr. Hirsch has nothing to disclose. Dr Gilmore has nothing to disclose. Dr. Vlachy has nothing to disclose. Dr. Haider has nothing to disclose. Dr. Rudin has nothing to disclose. Dr. Westover has nothing to disclose.