Ultrafast diagnosis of epilepsy: a randomised controlled trial of AI-assisted interpretation of ambulatory EEG recordings versus routine care.

Diagnosis of epilepsy is not straightforward; many epilepsy patients receive
their diagnosis years after their first symptoms. Current standard of care for
the diagnosis of epilepsy generally includes a routine electroencephalogram
(EEG). This EEG is visually inspected for signatures of epilepsy: interictal
epileptiform discharges (IEDs). These are short, characteristic, transient
anomalies in the EEG, associated with an increased likelihood of seizures. The
sensitivity of a routine EEG is low, however, due to the short duration of the
recording. Furthermore, IEDs occur more frequently during sleep than during
wakefulness, whereas routine EEGs are performed while the patient is awake.
After a normal or inconclusive routine EEG, a second routine EEG, a
sleep-deprivation EEG or a 24-hour ambulatory EEG may be considered. As an
ambulatory EEG has a longer duration and captures EEG data during sleep, it
greatly improves the sensitivity for IED capture. Although it is generally
accepted that ambulatory recordings are highly valuable for diagnostics in
epilepsy, they are not routinely used in most hospitals due to the time
consuming nature of ambulatory EEG interpretation that takes 2-3 hours for
review. Our group recently developed an artificial intelligence (AI) algorithm
for IED detection. This algorithm performs on par with experts and realizes a
50- to 75-fold time reduction to analyse ambulatory EEG. We hypothesize that
using AI-assisted interpretation of ambulatory EEG recordings is more efficient
in comparison to the current clinical practice (starting with a routine EEG) to
diagnose epilepsy and will significantly reduce the time to diagnosis in
patients referred for possible epilepsy.

To evaluate the effect of AI-assisted interpretation of ambulatory EEG
recordings compared to standard care (starting with a routine EEG) on the time
to diagnose epilepsy (measured as the time in weeks between initial referral
for an EEG and the final diagnose).

Randomized controlled trial.

Patients are randomized into two treatment arms: the standard care arm and the
ultra-fast AI arm. In the ultra-fast AI arm, an ambulant EEG recording is
performed and analyzed by an artificial intelligence algorithm. This algorithm
highlights events in the EEG with a high suspicion of interictal epileptiform
abnormalities. These marked events are visually assessed by a clinical
neurophysiologist. In addition, at least 20 minutes of the ambulatory EEG will
be visually assessed (similar to that in the standard care arm). The clinical
neurophysiologist will give the final conclusion.

Participants will be randomized to one of the study arms, the standard care or
the ultrafast diagnostics arm. Participants in the ultrafast diagnostics arm
will undergo a 24-hour routine EEG. The first 20 minutes of the ambulatory EEG
will be visually similar to standard care. In addition, the complete EEG will
be reviewed with a hybrid approach; in which the complete EEG is analysed by
the AI algorithm and all detected events are shown to an expert. Patients in
both study arms will have one extra outpatient clinic appointment after 1 year
to follow-up on their diagnosis. If the ultrafast diagnosis arm is proven to
reduce time-to-diagnosis significantly this will greatly impact future care.
Patients will receive their diagnosis sooner, which allows for earlier
treatment. Therefore, the (limited) risks and burden for the participating
capacitated adults are in proportion with the potential value of this study.