The majority of people use a smartwatch to track their steps and view notifications. It may eventually be the signal for someone with epilepsy that a seizure is imminent, giving them enough time to sit down, call someone, or just get away from danger. That possibility, which is still developing in research labs on at least two continents, is starting to feel more like engineering with a deadline than science fiction.
Approximately 1% of people worldwide suffer from epilepsy. Up until you realize what it truly means to live with uncontrollable seizures, the numbers are sufficiently large to be abstracted away. It becomes impossible to drive. It is not acceptable to swim by yourself. Cooking is risky. The risk of sudden death in epilepsy, or SUDEP, is 27 times higher than that of sudden death in the general population. This statistic tends to change after two readings. Unpredictability is the primary cruelty. Despite taking medication, nearly one-third of people with epilepsy still have seizures. For them, the daily concern is not whether a seizure will occur, but rather when, where, and whether anyone nearby will know what to do.
Seizures aren’t quite as random as they seem, according to what researchers have begun to piece together. Both the body and the brain send signals. Seizures tend to follow circadian and multiday rhythms, which are biological cycles that can be observed in heart rate, sleep patterns, and even step counts. When fed months’ worth of this data, a machine learning model can learn to identify when those cycles are approaching a window of high risk. It cannot read minds. However, it can recognize patterns, and sometimes patterns are sufficient.
Wearable smartwatches combined with machine learning models trained on heart rate cycles and sleep data were found to predict seizures above chance in every single participant when forecasting by the hour in a study that followed individuals with refractory epilepsy for an average of almost fifteen months. That model’s average warning time was 37 minutes, which is easily enough to sit down, get in touch with someone, or stay away from a kitchen knife but not long enough to reroute a life. It’s difficult to ignore that particular, human number for even a brief moment.
An even more advanced wearable headset that simultaneously analyzes cardiac signals and brainwaves is being developed by a research team at Glasgow Caledonian University. Up to 95% accuracy is claimed by their AI algorithm, which was trained on thousands of hours of EEG and ECG recordings. Earlier this year, the team was awarded additional funding through the UK’s Research and Innovation program. The lead researcher has expressed a desire for the finished product to be discreet, wireless, and resemble a lightweight cap. It’s still unclear if it will take that form, but the underlying science seems strong enough to sustain funding.
There are actual issues that the scientific community has yet to fully address. Wearable device adherence is more difficult over the long term than early trials indicate. When the novelty wears off, people either forget to charge them, find them uncomfortable, or just move on. Concerns about data security and privacy in relation to ongoing physiological monitoring are still real and not incidental. Furthermore, it is still difficult to identify seizure types other than tonic-clonic events, which are the dramatic, convulsive type that sensors most accurately identify. The way that focal seizures and absence seizures shake the wrist differs.
However, over the past few years, something has changed. Research teams at King’s College London have been conducting studies in which participants wear portable caps every day and control their own EEG recordings at home without clinical supervision. In the end, the messy and imperfect data from those home environments might yield forecasting models that are more helpful than anything produced in a hospital setting. The field appears to be heading in that direction, moving away from tools that people visit and toward tools that people actually live in.
Piece by piece, a future where the morning forecast encompasses more than just the weather is being constructed. This research gives the impression that the gap between what technology can measure and what medicine can do is closing more quickly than most people outside the field are aware.
