User login
NEW ORLEANS – For people with epilepsy, “the sudden and apparently unpredictable nature of seizures is one of the most disabling aspects of having the disorder,” said Michael Privitera, MD.
If a patient knew that “tomorrow will be a dangerous day” with a 50% chance of having a seizure, the patient could avoid hazardous activities, try to reduce stress, or increase supervision to reduce the risk of sudden, unexpected death in epilepsy, said Dr. Privitera, professor of neurology and director of the epilepsy center at the University of Cincinnati Gardner Neuroscience Institute. Physicians might be able to intervene during high-risk periods by altering antiepileptic drug regimens.
Evidence suggests that seizure prediction is possible today and that advances in wearable devices and analysis of chronic EEG recordings likely will improve the ability to predict seizures, Dr. Privitera said at the annual meeting of the American Epilepsy Society. Studies have found that some patients can predict the likelihood of seizures in the next 24 hours better than chance. In the future, algorithms that incorporate variables such as pulse, stress, mood, electrodermal activity, circadian rhythms, and EEG may further refine seizure prediction.
A complex picture
One problem with predicting seizures is that “you can have substantial changes in the seizure tendency, but not have a seizure,” Dr. Privitera said. Stress, alcohol, and missed medications, for example, may affect the seizure threshold. “They may be additive, and it may be when those things all hit at once that a seizure happens.”
Many patients report prodromal or premonitory symptoms before a seizure. “Most of us as clinicians will say, ‘Well, maybe you have some inkling, but I don’t think you’re really able to predict it,’ ” Dr. Privitera said.
Sheryl R. Haut, MD, professor of neurology at the Albert Einstein College of Medicine, New York, and her colleagues prospectively looked at patient self-prediction in 2007 (Neurology. 2007 Jan 23;68[4]:262-6). The investigators followed 74 people with epilepsy who completed a daily diary in which they predicted the likelihood of a seizure occurring in the next 24 hours. Their analysis included approximately 15,000 diary days and 1,400 seizure days.
A subset of participants, about 20%, was significantly better than chance at predicting when a seizure would happen. If a patient in this subgroup said that a seizure was extremely likely, then a seizure occurred approximately 37% of the time. If a patient predicted that a seizure was extremely unlikely, there was about a 10% chance of having a seizure.
“This was a pretty substantial difference,” Dr. Privitera said. Combining patients’ predictions with their self-reported stress levels seemed to yield the most accurate predictions.
Stress and the SMILE study
About 90% of people with epilepsy identify at least one seizure precipitant, and the most commonly cited trigger is stress. When Dr. Privitera and his colleagues surveyed patients in their clinic, 82% identified stress as a trigger (Epilepsy Behav. 2014 Dec;41:74-7). More than half of these patients had used some form of stress reduction, such as exercise, yoga, or meditation; 88% of those patients thought that stress reduction helped their seizures.
Underlying anxiety was the only difference between patients who thought that their seizures were triggered by stress and those who did not. Patients who did not think that stress triggered their seizures had significantly lower scores on the Generalized Anxiety Disorders–7.
Subsequently, Dr. Haut, Dr. Privitera, and colleagues conducted the Stress Management Intervention for Living with Epilepsy (SMILE) study, a prospective, controlled trial assessing the efficacy of a stress reduction intervention for reducing seizures, as well as measuring seizure self-prediction (Neurology. 2018 Mar 13;90[11]:e963-70). The researchers randomized patients to a progressive muscle relaxation intervention or to a control group; patients in the control group wrote down their activities for the day.
Patients posted diary entries twice daily into a smartphone, reporting stress levels and mood-related variables. As in Dr. Haut’s earlier study, patients predicted whether having a seizure was extremely unlikely, unlikely, neutral, likely, or extremely likely. Mood and stress variables (such as feeling unpleasant or pleasant, relaxed or stressed, and not worried or extremely worried) were ranked on a visual analog scale from 0 to 100.
The trial included participants who had at least two seizures per month and any seizure trigger. Medications were kept stable throughout the study. During a 2-month baseline, patients tracked their seizures and stress levels. During the 3-month treatment period, patients received the active or control intervention.
In all, 64 subjects completed the study, completing all diary entries on 94% of the days. In the active-treatment group, median seizure frequency decreased by 29%, compared with a 25% decrease in the control group. However, the difference between the groups was not statistically significant. Although the 25% reduction in the control group probably is partly attributable to the placebo effect, part of the decrease may be related to a mindfulness effect from completing the diary, Dr. Privitera said.
The active-treatment group had a statistically significant reduction in self-reported stress, compared with the control group, but this decrease did not correlate with seizure reduction. Changes in anxiety levels also did not correlate with seizures.
“It does not disprove the [stress] hypothesis, but it does tell us that there is more going on with stress and seizure triggers than just patients’ self-reported stress,” Dr. Privitera said.
Patients’ predictions
The seizure prediction findings in SMILE were similar to those of Dr. Haut’s earlier study. Among the 10 highest predictors out of the 64 participants, “when they said that a seizure was extremely likely, they were 8.36 times more likely to have a seizure than when they said a seizure was extremely unlikely,” Dr. Privitera said.
Many patients seemed to increase their predicted seizure probabilities in the days after having a seizure. In addition, feeling sad, nervous, worried, tense, or stressed significantly increased the likelihood that a patient would predict that a seizure was coming. However, these feelings were “not very accurate [for predicting] actual seizures,” he said. “Some people are better predictors, but really the basis of that prediction remains to be seen. One of my hypotheses is that some of these people may actually be responding to subclinical EEG changes.”
Together, these self-prediction studies include data from 4,500 seizures and 26,000 diary entries and show that “there is some information in patient self-report that can help us in understanding how to predict and when to predict seizures,” Dr. Privitera said.
Incorporating cardiac, EEG, and other variables
Various other factors may warrant inclusion in a seizure forecasting system. A new vagus nerve stimulation system responds to heart rate changes that occur at seizure onset. And for decades, researchers have studied the potential for EEG readings to predict seizures. A 2008 analysis of 47 reports concluded that limited progress had been made in predicting a seizure from interictal EEG (Epilepsy Behav. 2008 Jan;12[1]:128-35). Now, however, long-term intracranial recordings are providing new and important information about EEG patterns.
Whereas early studies examined EEG recordings from epilepsy monitoring units – when patients may have been sleep deprived, had medications removed, or recently undergone surgery – chronic intracranial recordings from devices such as the RNS (responsive neurostimulation) System have allowed researchers to look long term at EEG changes that are more representative of patients’ typical EEG patterns.
The RNS System detects interictal spikes and seizure discharges and then provides an electrical stimulation to stop seizures. “When you look at these recordings, there are a lot more electrographic seizures than clinical seizures that trigger these stimulations,” said Dr. Privitera. “If you look at somebody with a typical RNS, they may have 100 stimulations in a day and no clinical seizures. There are lots and lots of subclinical electrographic bursts – and not just spikes, but things that look like short electrographic seizures – that occur throughout the day.”
A handheld device
Researchers in Melbourne designed a system that uses implanted electrodes to provide chronic recordings (Lancet Neurol. 2013 Jun;12[6]:563-71). An algorithm then learned to predict the likelihood of a seizure from the patient’s data as the system recorded over time. The system could indicate when a seizure was likely by displaying a light on a handheld device. Patients were recorded for between 6 months and 3 years.
“There was a statistically significant ability to predict when seizures were happening,” Dr. Privitera said. “There is information in long-term intracranial recordings in many of these people that will help allow us to do a better prediction than what we are able to do right now, which is essentially not much.”
This research suggests that pooling data across patients may not be an effective seizure prediction strategy because different epilepsy types have different patterns. In addition, an individual’s patterns may differ from a group’s patterns. Complicating matters, individual patients may have multiple seizure types with different onset mechanisms.
“Another important lesson is that false positives in a deterministic sense may not represent false positives in a probabilistic sense,” Dr. Privitera said. “That is, when the seizure prediction program – whether it is the diary or the intracranial EEG or anything else – says the threshold changed, but you did not have a seizure, it does not mean that your prediction system was wrong. If the seizure tendency is going up … and your system says the seizure tendency went up, but all you are measuring is actual seizures, it looks like it is a false positive prediction of seizures. But in fact it is a true positive prediction of the seizure tendency changing but not necessarily reaching seizure threshold.”
Multiday patterns
Recent research shows that “we are just at the start,” Dr. Privitera said. “There are patterns underlying seizure frequency that … we are only beginning to be able to look at because of these chronic recordings.”
Baud et al. analyzed interictal epileptiform activity and seizures in patients who have had responsive neurostimulators for as long as 10 years (Nat Commun. 2018 Jan 8;9[1]:88). “What they found was that interictal spikes and rhythmic discharges oscillate with circadian and multiday periods that differ from person to person,” Dr. Privitera said. “There were multiday periodicities, most commonly in the 20- to 30-day duration, that were relatively stable over periods of time that lasted up to years.”
Researchers knew that seizures in women of childbearing age can cluster in association with the menstrual cycle, but similar cycles also were seen in men. In addition, the researchers found that seizures “occur preferentially during the rising phase of these multiday interictal rhythms,” which has implications for seizure forecasts, Dr. Privitera noted.
Stress biomarkers and wearables
Future seizure prediction methods may incorporate other biomarkers, such as stress hormones. A researcher at the University of Cincinnati, Jason Heikenfeld, PhD, is conducting research with a sensor that sticks to the wrist and measures sweat content, Dr. Privitera said. The technology originally was developed to measure sodium and potassium in sweat, but Dr. Privitera’s group has been working with him to measure cortisol, which may be a biomarker for stress and be useful for seizure prediction.
“Multivariate models are needed. We have lots of different ways that we can look at seizure prediction, and most likely the most accurate seizure prediction programs will incorporate multiple different areas,” Dr. Privitera said. “Seizure forecasting is possible. We can do it now. We can probably do it better than chance in many patients. ... It is important because changes in seizure likelihood could lead to pharmacologic or device or behavioral interventions that may help prevent seizures.”
Dr. Privitera reported conducting contracted research for Greenwich and SK Life Science and receiving consulting fees from Upsher-Smith and Astellas.
SOURCE: Privitera M. AES 2018, Judith Hoyer Lecture in Epilepsy.
NEW ORLEANS – For people with epilepsy, “the sudden and apparently unpredictable nature of seizures is one of the most disabling aspects of having the disorder,” said Michael Privitera, MD.
If a patient knew that “tomorrow will be a dangerous day” with a 50% chance of having a seizure, the patient could avoid hazardous activities, try to reduce stress, or increase supervision to reduce the risk of sudden, unexpected death in epilepsy, said Dr. Privitera, professor of neurology and director of the epilepsy center at the University of Cincinnati Gardner Neuroscience Institute. Physicians might be able to intervene during high-risk periods by altering antiepileptic drug regimens.
Evidence suggests that seizure prediction is possible today and that advances in wearable devices and analysis of chronic EEG recordings likely will improve the ability to predict seizures, Dr. Privitera said at the annual meeting of the American Epilepsy Society. Studies have found that some patients can predict the likelihood of seizures in the next 24 hours better than chance. In the future, algorithms that incorporate variables such as pulse, stress, mood, electrodermal activity, circadian rhythms, and EEG may further refine seizure prediction.
A complex picture
One problem with predicting seizures is that “you can have substantial changes in the seizure tendency, but not have a seizure,” Dr. Privitera said. Stress, alcohol, and missed medications, for example, may affect the seizure threshold. “They may be additive, and it may be when those things all hit at once that a seizure happens.”
Many patients report prodromal or premonitory symptoms before a seizure. “Most of us as clinicians will say, ‘Well, maybe you have some inkling, but I don’t think you’re really able to predict it,’ ” Dr. Privitera said.
Sheryl R. Haut, MD, professor of neurology at the Albert Einstein College of Medicine, New York, and her colleagues prospectively looked at patient self-prediction in 2007 (Neurology. 2007 Jan 23;68[4]:262-6). The investigators followed 74 people with epilepsy who completed a daily diary in which they predicted the likelihood of a seizure occurring in the next 24 hours. Their analysis included approximately 15,000 diary days and 1,400 seizure days.
A subset of participants, about 20%, was significantly better than chance at predicting when a seizure would happen. If a patient in this subgroup said that a seizure was extremely likely, then a seizure occurred approximately 37% of the time. If a patient predicted that a seizure was extremely unlikely, there was about a 10% chance of having a seizure.
“This was a pretty substantial difference,” Dr. Privitera said. Combining patients’ predictions with their self-reported stress levels seemed to yield the most accurate predictions.
Stress and the SMILE study
About 90% of people with epilepsy identify at least one seizure precipitant, and the most commonly cited trigger is stress. When Dr. Privitera and his colleagues surveyed patients in their clinic, 82% identified stress as a trigger (Epilepsy Behav. 2014 Dec;41:74-7). More than half of these patients had used some form of stress reduction, such as exercise, yoga, or meditation; 88% of those patients thought that stress reduction helped their seizures.
Underlying anxiety was the only difference between patients who thought that their seizures were triggered by stress and those who did not. Patients who did not think that stress triggered their seizures had significantly lower scores on the Generalized Anxiety Disorders–7.
Subsequently, Dr. Haut, Dr. Privitera, and colleagues conducted the Stress Management Intervention for Living with Epilepsy (SMILE) study, a prospective, controlled trial assessing the efficacy of a stress reduction intervention for reducing seizures, as well as measuring seizure self-prediction (Neurology. 2018 Mar 13;90[11]:e963-70). The researchers randomized patients to a progressive muscle relaxation intervention or to a control group; patients in the control group wrote down their activities for the day.
Patients posted diary entries twice daily into a smartphone, reporting stress levels and mood-related variables. As in Dr. Haut’s earlier study, patients predicted whether having a seizure was extremely unlikely, unlikely, neutral, likely, or extremely likely. Mood and stress variables (such as feeling unpleasant or pleasant, relaxed or stressed, and not worried or extremely worried) were ranked on a visual analog scale from 0 to 100.
The trial included participants who had at least two seizures per month and any seizure trigger. Medications were kept stable throughout the study. During a 2-month baseline, patients tracked their seizures and stress levels. During the 3-month treatment period, patients received the active or control intervention.
In all, 64 subjects completed the study, completing all diary entries on 94% of the days. In the active-treatment group, median seizure frequency decreased by 29%, compared with a 25% decrease in the control group. However, the difference between the groups was not statistically significant. Although the 25% reduction in the control group probably is partly attributable to the placebo effect, part of the decrease may be related to a mindfulness effect from completing the diary, Dr. Privitera said.
The active-treatment group had a statistically significant reduction in self-reported stress, compared with the control group, but this decrease did not correlate with seizure reduction. Changes in anxiety levels also did not correlate with seizures.
“It does not disprove the [stress] hypothesis, but it does tell us that there is more going on with stress and seizure triggers than just patients’ self-reported stress,” Dr. Privitera said.
Patients’ predictions
The seizure prediction findings in SMILE were similar to those of Dr. Haut’s earlier study. Among the 10 highest predictors out of the 64 participants, “when they said that a seizure was extremely likely, they were 8.36 times more likely to have a seizure than when they said a seizure was extremely unlikely,” Dr. Privitera said.
Many patients seemed to increase their predicted seizure probabilities in the days after having a seizure. In addition, feeling sad, nervous, worried, tense, or stressed significantly increased the likelihood that a patient would predict that a seizure was coming. However, these feelings were “not very accurate [for predicting] actual seizures,” he said. “Some people are better predictors, but really the basis of that prediction remains to be seen. One of my hypotheses is that some of these people may actually be responding to subclinical EEG changes.”
Together, these self-prediction studies include data from 4,500 seizures and 26,000 diary entries and show that “there is some information in patient self-report that can help us in understanding how to predict and when to predict seizures,” Dr. Privitera said.
Incorporating cardiac, EEG, and other variables
Various other factors may warrant inclusion in a seizure forecasting system. A new vagus nerve stimulation system responds to heart rate changes that occur at seizure onset. And for decades, researchers have studied the potential for EEG readings to predict seizures. A 2008 analysis of 47 reports concluded that limited progress had been made in predicting a seizure from interictal EEG (Epilepsy Behav. 2008 Jan;12[1]:128-35). Now, however, long-term intracranial recordings are providing new and important information about EEG patterns.
Whereas early studies examined EEG recordings from epilepsy monitoring units – when patients may have been sleep deprived, had medications removed, or recently undergone surgery – chronic intracranial recordings from devices such as the RNS (responsive neurostimulation) System have allowed researchers to look long term at EEG changes that are more representative of patients’ typical EEG patterns.
The RNS System detects interictal spikes and seizure discharges and then provides an electrical stimulation to stop seizures. “When you look at these recordings, there are a lot more electrographic seizures than clinical seizures that trigger these stimulations,” said Dr. Privitera. “If you look at somebody with a typical RNS, they may have 100 stimulations in a day and no clinical seizures. There are lots and lots of subclinical electrographic bursts – and not just spikes, but things that look like short electrographic seizures – that occur throughout the day.”
A handheld device
Researchers in Melbourne designed a system that uses implanted electrodes to provide chronic recordings (Lancet Neurol. 2013 Jun;12[6]:563-71). An algorithm then learned to predict the likelihood of a seizure from the patient’s data as the system recorded over time. The system could indicate when a seizure was likely by displaying a light on a handheld device. Patients were recorded for between 6 months and 3 years.
“There was a statistically significant ability to predict when seizures were happening,” Dr. Privitera said. “There is information in long-term intracranial recordings in many of these people that will help allow us to do a better prediction than what we are able to do right now, which is essentially not much.”
This research suggests that pooling data across patients may not be an effective seizure prediction strategy because different epilepsy types have different patterns. In addition, an individual’s patterns may differ from a group’s patterns. Complicating matters, individual patients may have multiple seizure types with different onset mechanisms.
“Another important lesson is that false positives in a deterministic sense may not represent false positives in a probabilistic sense,” Dr. Privitera said. “That is, when the seizure prediction program – whether it is the diary or the intracranial EEG or anything else – says the threshold changed, but you did not have a seizure, it does not mean that your prediction system was wrong. If the seizure tendency is going up … and your system says the seizure tendency went up, but all you are measuring is actual seizures, it looks like it is a false positive prediction of seizures. But in fact it is a true positive prediction of the seizure tendency changing but not necessarily reaching seizure threshold.”
Multiday patterns
Recent research shows that “we are just at the start,” Dr. Privitera said. “There are patterns underlying seizure frequency that … we are only beginning to be able to look at because of these chronic recordings.”
Baud et al. analyzed interictal epileptiform activity and seizures in patients who have had responsive neurostimulators for as long as 10 years (Nat Commun. 2018 Jan 8;9[1]:88). “What they found was that interictal spikes and rhythmic discharges oscillate with circadian and multiday periods that differ from person to person,” Dr. Privitera said. “There were multiday periodicities, most commonly in the 20- to 30-day duration, that were relatively stable over periods of time that lasted up to years.”
Researchers knew that seizures in women of childbearing age can cluster in association with the menstrual cycle, but similar cycles also were seen in men. In addition, the researchers found that seizures “occur preferentially during the rising phase of these multiday interictal rhythms,” which has implications for seizure forecasts, Dr. Privitera noted.
Stress biomarkers and wearables
Future seizure prediction methods may incorporate other biomarkers, such as stress hormones. A researcher at the University of Cincinnati, Jason Heikenfeld, PhD, is conducting research with a sensor that sticks to the wrist and measures sweat content, Dr. Privitera said. The technology originally was developed to measure sodium and potassium in sweat, but Dr. Privitera’s group has been working with him to measure cortisol, which may be a biomarker for stress and be useful for seizure prediction.
“Multivariate models are needed. We have lots of different ways that we can look at seizure prediction, and most likely the most accurate seizure prediction programs will incorporate multiple different areas,” Dr. Privitera said. “Seizure forecasting is possible. We can do it now. We can probably do it better than chance in many patients. ... It is important because changes in seizure likelihood could lead to pharmacologic or device or behavioral interventions that may help prevent seizures.”
Dr. Privitera reported conducting contracted research for Greenwich and SK Life Science and receiving consulting fees from Upsher-Smith and Astellas.
SOURCE: Privitera M. AES 2018, Judith Hoyer Lecture in Epilepsy.
NEW ORLEANS – For people with epilepsy, “the sudden and apparently unpredictable nature of seizures is one of the most disabling aspects of having the disorder,” said Michael Privitera, MD.
If a patient knew that “tomorrow will be a dangerous day” with a 50% chance of having a seizure, the patient could avoid hazardous activities, try to reduce stress, or increase supervision to reduce the risk of sudden, unexpected death in epilepsy, said Dr. Privitera, professor of neurology and director of the epilepsy center at the University of Cincinnati Gardner Neuroscience Institute. Physicians might be able to intervene during high-risk periods by altering antiepileptic drug regimens.
Evidence suggests that seizure prediction is possible today and that advances in wearable devices and analysis of chronic EEG recordings likely will improve the ability to predict seizures, Dr. Privitera said at the annual meeting of the American Epilepsy Society. Studies have found that some patients can predict the likelihood of seizures in the next 24 hours better than chance. In the future, algorithms that incorporate variables such as pulse, stress, mood, electrodermal activity, circadian rhythms, and EEG may further refine seizure prediction.
A complex picture
One problem with predicting seizures is that “you can have substantial changes in the seizure tendency, but not have a seizure,” Dr. Privitera said. Stress, alcohol, and missed medications, for example, may affect the seizure threshold. “They may be additive, and it may be when those things all hit at once that a seizure happens.”
Many patients report prodromal or premonitory symptoms before a seizure. “Most of us as clinicians will say, ‘Well, maybe you have some inkling, but I don’t think you’re really able to predict it,’ ” Dr. Privitera said.
Sheryl R. Haut, MD, professor of neurology at the Albert Einstein College of Medicine, New York, and her colleagues prospectively looked at patient self-prediction in 2007 (Neurology. 2007 Jan 23;68[4]:262-6). The investigators followed 74 people with epilepsy who completed a daily diary in which they predicted the likelihood of a seizure occurring in the next 24 hours. Their analysis included approximately 15,000 diary days and 1,400 seizure days.
A subset of participants, about 20%, was significantly better than chance at predicting when a seizure would happen. If a patient in this subgroup said that a seizure was extremely likely, then a seizure occurred approximately 37% of the time. If a patient predicted that a seizure was extremely unlikely, there was about a 10% chance of having a seizure.
“This was a pretty substantial difference,” Dr. Privitera said. Combining patients’ predictions with their self-reported stress levels seemed to yield the most accurate predictions.
Stress and the SMILE study
About 90% of people with epilepsy identify at least one seizure precipitant, and the most commonly cited trigger is stress. When Dr. Privitera and his colleagues surveyed patients in their clinic, 82% identified stress as a trigger (Epilepsy Behav. 2014 Dec;41:74-7). More than half of these patients had used some form of stress reduction, such as exercise, yoga, or meditation; 88% of those patients thought that stress reduction helped their seizures.
Underlying anxiety was the only difference between patients who thought that their seizures were triggered by stress and those who did not. Patients who did not think that stress triggered their seizures had significantly lower scores on the Generalized Anxiety Disorders–7.
Subsequently, Dr. Haut, Dr. Privitera, and colleagues conducted the Stress Management Intervention for Living with Epilepsy (SMILE) study, a prospective, controlled trial assessing the efficacy of a stress reduction intervention for reducing seizures, as well as measuring seizure self-prediction (Neurology. 2018 Mar 13;90[11]:e963-70). The researchers randomized patients to a progressive muscle relaxation intervention or to a control group; patients in the control group wrote down their activities for the day.
Patients posted diary entries twice daily into a smartphone, reporting stress levels and mood-related variables. As in Dr. Haut’s earlier study, patients predicted whether having a seizure was extremely unlikely, unlikely, neutral, likely, or extremely likely. Mood and stress variables (such as feeling unpleasant or pleasant, relaxed or stressed, and not worried or extremely worried) were ranked on a visual analog scale from 0 to 100.
The trial included participants who had at least two seizures per month and any seizure trigger. Medications were kept stable throughout the study. During a 2-month baseline, patients tracked their seizures and stress levels. During the 3-month treatment period, patients received the active or control intervention.
In all, 64 subjects completed the study, completing all diary entries on 94% of the days. In the active-treatment group, median seizure frequency decreased by 29%, compared with a 25% decrease in the control group. However, the difference between the groups was not statistically significant. Although the 25% reduction in the control group probably is partly attributable to the placebo effect, part of the decrease may be related to a mindfulness effect from completing the diary, Dr. Privitera said.
The active-treatment group had a statistically significant reduction in self-reported stress, compared with the control group, but this decrease did not correlate with seizure reduction. Changes in anxiety levels also did not correlate with seizures.
“It does not disprove the [stress] hypothesis, but it does tell us that there is more going on with stress and seizure triggers than just patients’ self-reported stress,” Dr. Privitera said.
Patients’ predictions
The seizure prediction findings in SMILE were similar to those of Dr. Haut’s earlier study. Among the 10 highest predictors out of the 64 participants, “when they said that a seizure was extremely likely, they were 8.36 times more likely to have a seizure than when they said a seizure was extremely unlikely,” Dr. Privitera said.
Many patients seemed to increase their predicted seizure probabilities in the days after having a seizure. In addition, feeling sad, nervous, worried, tense, or stressed significantly increased the likelihood that a patient would predict that a seizure was coming. However, these feelings were “not very accurate [for predicting] actual seizures,” he said. “Some people are better predictors, but really the basis of that prediction remains to be seen. One of my hypotheses is that some of these people may actually be responding to subclinical EEG changes.”
Together, these self-prediction studies include data from 4,500 seizures and 26,000 diary entries and show that “there is some information in patient self-report that can help us in understanding how to predict and when to predict seizures,” Dr. Privitera said.
Incorporating cardiac, EEG, and other variables
Various other factors may warrant inclusion in a seizure forecasting system. A new vagus nerve stimulation system responds to heart rate changes that occur at seizure onset. And for decades, researchers have studied the potential for EEG readings to predict seizures. A 2008 analysis of 47 reports concluded that limited progress had been made in predicting a seizure from interictal EEG (Epilepsy Behav. 2008 Jan;12[1]:128-35). Now, however, long-term intracranial recordings are providing new and important information about EEG patterns.
Whereas early studies examined EEG recordings from epilepsy monitoring units – when patients may have been sleep deprived, had medications removed, or recently undergone surgery – chronic intracranial recordings from devices such as the RNS (responsive neurostimulation) System have allowed researchers to look long term at EEG changes that are more representative of patients’ typical EEG patterns.
The RNS System detects interictal spikes and seizure discharges and then provides an electrical stimulation to stop seizures. “When you look at these recordings, there are a lot more electrographic seizures than clinical seizures that trigger these stimulations,” said Dr. Privitera. “If you look at somebody with a typical RNS, they may have 100 stimulations in a day and no clinical seizures. There are lots and lots of subclinical electrographic bursts – and not just spikes, but things that look like short electrographic seizures – that occur throughout the day.”
A handheld device
Researchers in Melbourne designed a system that uses implanted electrodes to provide chronic recordings (Lancet Neurol. 2013 Jun;12[6]:563-71). An algorithm then learned to predict the likelihood of a seizure from the patient’s data as the system recorded over time. The system could indicate when a seizure was likely by displaying a light on a handheld device. Patients were recorded for between 6 months and 3 years.
“There was a statistically significant ability to predict when seizures were happening,” Dr. Privitera said. “There is information in long-term intracranial recordings in many of these people that will help allow us to do a better prediction than what we are able to do right now, which is essentially not much.”
This research suggests that pooling data across patients may not be an effective seizure prediction strategy because different epilepsy types have different patterns. In addition, an individual’s patterns may differ from a group’s patterns. Complicating matters, individual patients may have multiple seizure types with different onset mechanisms.
“Another important lesson is that false positives in a deterministic sense may not represent false positives in a probabilistic sense,” Dr. Privitera said. “That is, when the seizure prediction program – whether it is the diary or the intracranial EEG or anything else – says the threshold changed, but you did not have a seizure, it does not mean that your prediction system was wrong. If the seizure tendency is going up … and your system says the seizure tendency went up, but all you are measuring is actual seizures, it looks like it is a false positive prediction of seizures. But in fact it is a true positive prediction of the seizure tendency changing but not necessarily reaching seizure threshold.”
Multiday patterns
Recent research shows that “we are just at the start,” Dr. Privitera said. “There are patterns underlying seizure frequency that … we are only beginning to be able to look at because of these chronic recordings.”
Baud et al. analyzed interictal epileptiform activity and seizures in patients who have had responsive neurostimulators for as long as 10 years (Nat Commun. 2018 Jan 8;9[1]:88). “What they found was that interictal spikes and rhythmic discharges oscillate with circadian and multiday periods that differ from person to person,” Dr. Privitera said. “There were multiday periodicities, most commonly in the 20- to 30-day duration, that were relatively stable over periods of time that lasted up to years.”
Researchers knew that seizures in women of childbearing age can cluster in association with the menstrual cycle, but similar cycles also were seen in men. In addition, the researchers found that seizures “occur preferentially during the rising phase of these multiday interictal rhythms,” which has implications for seizure forecasts, Dr. Privitera noted.
Stress biomarkers and wearables
Future seizure prediction methods may incorporate other biomarkers, such as stress hormones. A researcher at the University of Cincinnati, Jason Heikenfeld, PhD, is conducting research with a sensor that sticks to the wrist and measures sweat content, Dr. Privitera said. The technology originally was developed to measure sodium and potassium in sweat, but Dr. Privitera’s group has been working with him to measure cortisol, which may be a biomarker for stress and be useful for seizure prediction.
“Multivariate models are needed. We have lots of different ways that we can look at seizure prediction, and most likely the most accurate seizure prediction programs will incorporate multiple different areas,” Dr. Privitera said. “Seizure forecasting is possible. We can do it now. We can probably do it better than chance in many patients. ... It is important because changes in seizure likelihood could lead to pharmacologic or device or behavioral interventions that may help prevent seizures.”
Dr. Privitera reported conducting contracted research for Greenwich and SK Life Science and receiving consulting fees from Upsher-Smith and Astellas.
SOURCE: Privitera M. AES 2018, Judith Hoyer Lecture in Epilepsy.
REPORTING FROM AES 2018