A new study in patients with epilepsy has shown that seizure prediction can be achieved through an implantable device that records and analyses EEG signals. “This is the first time seizure prediction has been demonstrated in humans,” says lead author Mark Cook, Professor and Director of Neurology at St Vincent's Hospital, Melbourne, Australia.

Seizures are inadequately controlled by available treatments in 30–40% of patients with epilepsy, representing a substantial unmet medical need. Moreover, the unpredictable nature of seizures is associated with increased risk of injury and psychological comorbidities such as depression. Previous studies had suggested that changes occur in the brain prior to epileptic attacks, but the underlying relationships between such changes and seizure onset were unclear.

For the current study, 15 patients with treatment-resistant epilepsy from three specialist centres in Australia were implanted with the seizure advisory system. A small craniotomy was performed to enable placement of two implantable lead assemblies over the brain quadrant thought to contain the epileptogenic focus, on the basis of previous testing. The leads collected intracranial EEG signals from the cortical surface via electrode arrays, and conducted the signal to a telemetry unit implanted below the clavicle. This unit, in turn, wirelessly transmitted the EEG data to a hand-held advisory device. Blue, white and red lights on the advisory device notified patients of a low, moderate or high risk of seizure, respectively, and were accompanied by an audible signal.

The seizure advisory system: EEG electrode arrays on the cortical surface connected to a subclavicular telemetry unit. Credit: Image courtesy of M. Cook.

Following device implantation, participants entered a data collection phase during which EEG data were recorded, enabling generation of a patient-specific algorithm for seizure prediction. The algorithm was then tested against predefined performance criteria—namely, the sensitivity of the red (high-risk) advisory indicator had to be at least 65% and superior to a time-matched random indicator, and the false-negative rate of the blue (low-risk) indicator had to be superior to a time-matched random indicator. The algorithm met the performance criteria in 11 of the patients, who then entered the advisory phase of the trial, during which time they received advisory signals from the hand-held device on their risk of seizure.

The primary end point was the number of device-related adverse events 4 months post-implantation, and secondary end points included clinical effectiveness at 4 months and long-term device safety.

“The most significant finding was that accurate prediction of seizures was possible in most people—always in some, most of the time in others,” says Cook. The advisory system met the performance criteria in eight of the 11 patients who entered the advisory phase, and the red advisory signal achieved 100% sensitivity in two of the patients—that is, all seizures occurred during the red advisory phase, with a mean warning time of 114 min. Moreover, the blue advisory signal had a negative predictive value of 98–100% in the five patients in which it was enabled.

11 device-related adverse events occurred within the first 4 months after device implantation, two of which were classified as serious. In one patient, the telemetry unit migrated, while another patient experienced fluid accumulation around the surgical wound. Procedures to correct these events led to recovery in both patients.

Two further serious adverse events were noted at 12 months post-implantation: infection in one patient, and lead tautness in another. In both cases, the events were followed by device explantation. Overall, complication rates were similar to those observed with implantable deep brain stimulation devices for patients with Parkinson disease. Moreover, the electrodes used in the current study have previously been employed long-term in patients with epilepsy, without causing complications.

During the study, patients recorded seizure events in a diary. “One very surprising finding was the large discrepancy between events reported by patients and actual events,” notes Cook, which indicated that most patients underestimated seizures. “This finding has implications for daily management of patients and for drug trials, which often rely on patient reporting of events.”

Cook's team is aiming to take the work forward through a larger study with a less invasive device.