Novel cEEG-based scoring system predicts inpatient seizure risk

LOS ANGELES – A novel scoring system based on six readily available seizure risk factors from a patient’s history and continuous electroencephalogram (cEEG) monitoring appears to accurately predict seizures in acutely ill hospitalized patients.

The final model of the system, dubbed the 2HELPS2B score, has an area under the curve (AUC) of 0.821, suggesting a “good overall fit,” Aaron Struck, MD, reported at the annual meeting of the American Academy of Neurology.

However, more relevant than the AUC and suggestive of high classification accuracy is the low calibration error of 2.7%, which shows that the actual incidence of seizures within a particular risk group is, on average, within 2.7% of predicted incidence, Dr. Struck of the University of Wisconsin, Madison, explained in an interview.

The use of cEEG has expanded, largely because of a high incidence of subclinical seizures in hospitalized patients with encephalopathy; EEG features believed to predict seizures include epileptiform discharges and periodic discharges, but the ways in which these variables may jointly affect seizure risk have not been studied, he said.

He and his colleagues used a prospective database to derive a dataset containing 24 clinical and electroencephalographic variables for 5,427 cEEG sessions of at least 24 hours each, and then, using a machine-learning method known as RiskSLIM, created a scoring system model to estimate seizure risk in patients undergoing cEEG.

The name of the scoring system – 2HELPS2B – represents the six variables included in the final model:

• 2 H is for frequency greater than 2.0 Hz for any periodic rhythmic pattern (1 point).
• E is for sporadic epileptiform discharges (1 point).
• L is for the presence of lateralized periodic discharges, lateralized rhythmic delta activity, or bilateral independent periodic discharges (1 point).
• P is for the presence of “plus” features, including superimposed, rhythmic, sharp, or fast activity (1 point).
• S is for prior seizure (1 point).
• 2B is for brief, potentially ictal, rhythmic discharges (2 points).

The predicted seizure risk rose with score, such that the seizure risk was less than 5% for a score of 0, 12% for 1, 27% for 2, 50% for 3, 73% for 4, 88% for 5, and greater than 95% for 6-7, Dr. Struck said. “Really, anything over 2 points, you’re at substantial risk for having seizures.”

Limitations of the study, which are being addressed in an ongoing, multicenter, prospective validation trial through the Critical Care EEG Monitoring Research Consortium, are mainly related to the constraints of the database; the duration of EEG needed to accurately calculate the 2HELPS2B score wasn’t defined, and cEEGs were of varying length.

“So in our validation study moving forward, these are two things we will address,” he said. “We also want to show that this is something that’s useful on a day-to-day basis – that it accurately gauges the degree of variability or potential severity of the ictal-interictal continuum pattern.”

With validation, Dr. Struck said that the 2HELPS2B score could ultimately be used to rapidly communicate seizure potential based on EEG severity and to guide decision making with respect to initiation of empiric antiseizure medication.

Findings from the validation study are “trending in the right direction,” but the confidence intervals are wide, as only 404 patients have been included at this point, Dr. Struck said.

This study was supported by a research infrastructure award from the American Epilepsy Society and the Epilepsy Foundation.

[email protected]

SOURCE: Struck A et al. Neurology. 2018 Apr 90(15 Suppl.):S11.002.

LOS ANGELES – A novel scoring system based on six readily available seizure risk factors from a patient’s history and continuous electroencephalogram (cEEG) monitoring appears to accurately predict seizures in acutely ill hospitalized patients.

The final model of the system, dubbed the 2HELPS2B score, has an area under the curve (AUC) of 0.821, suggesting a “good overall fit,” Aaron Struck, MD, reported at the annual meeting of the American Academy of Neurology.

However, more relevant than the AUC and suggestive of high classification accuracy is the low calibration error of 2.7%, which shows that the actual incidence of seizures within a particular risk group is, on average, within 2.7% of predicted incidence, Dr. Struck of the University of Wisconsin, Madison, explained in an interview.

The use of cEEG has expanded, largely because of a high incidence of subclinical seizures in hospitalized patients with encephalopathy; EEG features believed to predict seizures include epileptiform discharges and periodic discharges, but the ways in which these variables may jointly affect seizure risk have not been studied, he said.

He and his colleagues used a prospective database to derive a dataset containing 24 clinical and electroencephalographic variables for 5,427 cEEG sessions of at least 24 hours each, and then, using a machine-learning method known as RiskSLIM, created a scoring system model to estimate seizure risk in patients undergoing cEEG.

The name of the scoring system – 2HELPS2B – represents the six variables included in the final model:

• 2 H is for frequency greater than 2.0 Hz for any periodic rhythmic pattern (1 point).
• E is for sporadic epileptiform discharges (1 point).
• L is for the presence of lateralized periodic discharges, lateralized rhythmic delta activity, or bilateral independent periodic discharges (1 point).
• P is for the presence of “plus” features, including superimposed, rhythmic, sharp, or fast activity (1 point).
• S is for prior seizure (1 point).
• 2B is for brief, potentially ictal, rhythmic discharges (2 points).

The predicted seizure risk rose with score, such that the seizure risk was less than 5% for a score of 0, 12% for 1, 27% for 2, 50% for 3, 73% for 4, 88% for 5, and greater than 95% for 6-7, Dr. Struck said. “Really, anything over 2 points, you’re at substantial risk for having seizures.”

Limitations of the study, which are being addressed in an ongoing, multicenter, prospective validation trial through the Critical Care EEG Monitoring Research Consortium, are mainly related to the constraints of the database; the duration of EEG needed to accurately calculate the 2HELPS2B score wasn’t defined, and cEEGs were of varying length.

“So in our validation study moving forward, these are two things we will address,” he said. “We also want to show that this is something that’s useful on a day-to-day basis – that it accurately gauges the degree of variability or potential severity of the ictal-interictal continuum pattern.”

With validation, Dr. Struck said that the 2HELPS2B score could ultimately be used to rapidly communicate seizure potential based on EEG severity and to guide decision making with respect to initiation of empiric antiseizure medication.

Findings from the validation study are “trending in the right direction,” but the confidence intervals are wide, as only 404 patients have been included at this point, Dr. Struck said.

This study was supported by a research infrastructure award from the American Epilepsy Society and the Epilepsy Foundation.

[email protected]

SOURCE: Struck A et al. Neurology. 2018 Apr 90(15 Suppl.):S11.002.

LOS ANGELES – A novel scoring system based on six readily available seizure risk factors from a patient’s history and continuous electroencephalogram (cEEG) monitoring appears to accurately predict seizures in acutely ill hospitalized patients.

The final model of the system, dubbed the 2HELPS2B score, has an area under the curve (AUC) of 0.821, suggesting a “good overall fit,” Aaron Struck, MD, reported at the annual meeting of the American Academy of Neurology.

However, more relevant than the AUC and suggestive of high classification accuracy is the low calibration error of 2.7%, which shows that the actual incidence of seizures within a particular risk group is, on average, within 2.7% of predicted incidence, Dr. Struck of the University of Wisconsin, Madison, explained in an interview.

The use of cEEG has expanded, largely because of a high incidence of subclinical seizures in hospitalized patients with encephalopathy; EEG features believed to predict seizures include epileptiform discharges and periodic discharges, but the ways in which these variables may jointly affect seizure risk have not been studied, he said.

He and his colleagues used a prospective database to derive a dataset containing 24 clinical and electroencephalographic variables for 5,427 cEEG sessions of at least 24 hours each, and then, using a machine-learning method known as RiskSLIM, created a scoring system model to estimate seizure risk in patients undergoing cEEG.

The name of the scoring system – 2HELPS2B – represents the six variables included in the final model:

• 2 H is for frequency greater than 2.0 Hz for any periodic rhythmic pattern (1 point).
• E is for sporadic epileptiform discharges (1 point).
• L is for the presence of lateralized periodic discharges, lateralized rhythmic delta activity, or bilateral independent periodic discharges (1 point).
• P is for the presence of “plus” features, including superimposed, rhythmic, sharp, or fast activity (1 point).
• S is for prior seizure (1 point).
• 2B is for brief, potentially ictal, rhythmic discharges (2 points).

The predicted seizure risk rose with score, such that the seizure risk was less than 5% for a score of 0, 12% for 1, 27% for 2, 50% for 3, 73% for 4, 88% for 5, and greater than 95% for 6-7, Dr. Struck said. “Really, anything over 2 points, you’re at substantial risk for having seizures.”

Limitations of the study, which are being addressed in an ongoing, multicenter, prospective validation trial through the Critical Care EEG Monitoring Research Consortium, are mainly related to the constraints of the database; the duration of EEG needed to accurately calculate the 2HELPS2B score wasn’t defined, and cEEGs were of varying length.

“So in our validation study moving forward, these are two things we will address,” he said. “We also want to show that this is something that’s useful on a day-to-day basis – that it accurately gauges the degree of variability or potential severity of the ictal-interictal continuum pattern.”

With validation, Dr. Struck said that the 2HELPS2B score could ultimately be used to rapidly communicate seizure potential based on EEG severity and to guide decision making with respect to initiation of empiric antiseizure medication.

Findings from the validation study are “trending in the right direction,” but the confidence intervals are wide, as only 404 patients have been included at this point, Dr. Struck said.

This study was supported by a research infrastructure award from the American Epilepsy Society and the Epilepsy Foundation.

[email protected]

SOURCE: Struck A et al. Neurology. 2018 Apr 90(15 Suppl.):S11.002.