One of the most devastating aspects of epilepsy is its presumed unpredictability, where seizures can occur at any moment—in public, at school, at work, or even while driving—and most patients are unable to predict with any degree of certainty when a seizure is likely to strike. Consequently, developing a reliable way to gauge a patient’s risk for experiencing a seizure on a specific day, potentially providing the patient with the opportunity to take higher doses of their medications or to take other action, would be a game changer in the treatment of epilepsy.

Some patients with epilepsy report being able to predict when they are more likely to have seizures based on factors such as stress levels. In fact, a colleague here at the University of Cincinnati, Michael Privitera, MD, has been investigating such seizure self-prediction. Patients’ electronic seizure diaries can be analyzed to understand how stress might relate to seizure occurrence and to learn what other factors may be useful to predict seizures. Algorithms are also being studied that could potentially help clinicians to determine levels of seizure control stability and help to guide treatment decisions. In addition, techniques such as deep brain stimulation, which involve the use of implantable recording devices, might uncover electrophysiological biomarkers that can show when a patient is at an increased risk for seizures. The idea that these biomarkers might be paired with electronic seizure diaries is very exciting.

Using these methods, more accurate ways to predict seizure risk may be designed that can then lead to clinical trials of new interventions to prevent seizures. The challenge is that these devices generate voluminous amounts of data; artificial intelligence, data engineers, and even statisticians will be needed to process all of that information. While these all are promising and exciting developments, a multidisciplinary approach will be required to make sense of the data so that more can be learned about seizures with the goal of predicting and preventing them.

Epilepsy treatment decisions are based on seizure frequency, which may appear to be straightforward, but seizure frequency can actually be challenging to assess in practice. The inadequacies of seizure logs are well established. Having to record each seizure (ie, keeping a seizure log) is difficult and may reflect the patient’s best recollection rather than a true incidence over time. Further, even if the seizure log is completely reliable, it might still be unclear whether a patient is improving on the current treatment. Seizures often seem to occur randomly, and seeing a change in random occurrences is challenging. Clinical experience makes a difference, and physicians can become better at evaluating patients based on seizure logs with more experience. Nevertheless, the need still exists for seizure risk assessment tools that are more reliable than seizure logs.

Currently, the largest data sets are patient report–based and use mobile device applications (eg, Seizure Tracker). Work published by Cook et al involving implanted devices with intracranial electroencephalogram monitoring to look for signatures of when seizures occur provides a small data set but one with observations collected over a long period of time. Data sets have also been derived from brain-responsive neurostimulation (RNS System, NeuroPace, Inc). With these approaches, seizure detection is reliant on electrode placement, and some seizures may be missed because of electrode location. There are also seizure monitoring devices that are physiologic but not electroencephalographic (eg, they monitor heart rate, skin conduction, and movement) and sometimes are used in combination to maximize reliability. Separate from the seizure detection approaches, seizure risk assessment algorithms show some promise for measuring seizure risk based on the pattern of identified seizures, irrespective of how the seizure is identified. The Epilepsy Seizure Assessment Tool (EpiSAT) for outpatient seizure risk assessment using seizure counting data was recently evaluated against 24 specialized epilepsy clinicians. The results showed that the EpiSAT exhibited substantial observed agreement (75.4%) with clinicians for assessing seizure risk. While these are all promising developments, better tools to objectively measure seizure occurrence are sorely needed.

The ability to predict seizure occurrence would be a significant advancement in epilepsy treatment. Efforts are underway to utilize big data and artificial intelligence to predict either seizures themselves or, at minimum, high-risk periods for seizure occurrence. Still, significant challenges remain. For example, even if a high-risk pattern or period could be identified for a given patient, what advice should that patient be given? Stay home during the high-risk period? Prescribe the patient a different medicine that could be taken during that period? Epilepsy care is not yet at that point, even when patients document their seizures reliably in seizure logs.

If there is a 75% chance of having a seizure today, should the patient call out sick from work? Even if they did, a seizure could still occur the next day. In that case, the forecasted seizure has not been prevented, but rather it was delayed. Implantable devices, based on the concepts governing defibrillators and pacemakers, hold promise, but the complexity of the brain is greater than that of the heart.

Correlating specific brain function with seizures is daunting. Many brain changes are related to normal adaptive functions, and many of those are similar to pattern changes observed prior to or at the beginning of a seizure. Another challenge is to tease out the differences between the brains of children and adults. A pattern considered normal in an 8-year-old child may be abnormal in an adult, and findings from studies of implantable recording devices in adults are not necessarily generalizable to children. Further, while there are patients who say that they can predict their seizures, the distinction between triggers and predictors is an important one. For some patients, triggers might include a stressful day at work. Patients with migraines are similar in that many report their headaches often occur during stressful periods. These are triggers, not predictors; nonetheless, by learning triggers, patients with epilepsy can learn what circumstances to avoid. So, we are making progress, but we are clearly not there yet.