Pre-Seizure Symptoms: What an Aura Is, What a Prodrome Is, and Why the Difference Matters Diagnostically

Under the International League Against Epilepsy 2017 classification, the subjective experience patients describe as an aura is treated as a focal aware seizure: sensory, autonomic, emotional, or cognitive onset with awareness retained. It is not a warning before a seizure. It is the beginning of one, and the earliest prominent manifestation defines the seizure type.

A prodrome is different. Prodromal or premonitory features occur before electrographic seizure onset, sometimes by hours, and they do not localize. The most commonly reported ones are unglamorous: tiredness, a hazy look, sleepiness.

The distinction has direct diagnostic value. An aura, correctly described, tells a neurologist where in the brain the seizure starts, and in the workup for epilepsy surgery that information carries real weight. A prodrome does not localize, but it may create a window for safety measures.

It also protects against a specific error. Patients who experience auras sometimes conclude that they have a reliable early warning system and that their medication has become optional. The evidence on seizure self-prediction does not support that conclusion, and the consequences of acting on it can be severe.

The International League Against Epilepsy published a revised operational classification of seizure types in Epilepsia in 2017, led by Robert Fisher, with a companion instruction manual the same year. The revision retired several familiar terms: partial became focal, and simple partial, complex partial, dyscognitive, psychic, and secondarily generalized were eliminated outright.

Awareness became the classifier for focal seizures. A focal seizure in which awareness is retained throughout is a focal aware seizure. If awareness is impaired during any segment, it is a focal impaired awareness seizure. Focal seizures are further described by their earliest prominent manifestation, which may be motor (automatisms, atonic, clonic, epileptic spasms, hyperkinetic, myoclonic, tonic) or nonmotor (autonomic, behavior arrest, cognitive, emotional, sensory).

That framework is what places the aura inside the seizure rather than before it. The rising epigastric sensation, the sudden unexplained fear, the familiar smell, the deja vu, the visual disturbance: each of these is a nonmotor focal onset, occurring with awareness retained. When the event then spreads, it becomes a focal to bilateral tonic-clonic seizure, which is the term that replaced secondarily generalized.

This matters for what a patient should report. The sequence of sensations, in order, from the very first thing noticed, is the clinical data. A description that begins at the point of collapse discards the part that localizes.

Separate from the aura is a longer and vaguer period that some patients and many caregivers report recognizing. The research on this is smaller than the topic’s prominence would suggest.

A 2017 study in Epilepsy and Behavior by Puja Patel, Sheryl Haut, and colleagues at Montefiore Medical Center surveyed caretakers of patients aged 0 to 21 who had experienced at least one seizure in the previous year, excluding those with non-epileptic seizures or daily seizures. Of 150 qualifying questionnaires, 32 caretakers, or 21.6 percent, reported that they could predict seizures. The most commonly reported pre-ictal features were being tired, a hazy look, and sleepiness. Age at seizure onset was earlier in the group reporting prediction, 3.3 years on average compared with 5.3 years, a statistically significant difference. Separately, 76.6 percent of caretakers reported at least one seizure precipitant.

The authors noted that the rate of caretaker prediction in children was similar to rates of self-prediction reported in adult studies, which is itself informative. Roughly one in five, whether the observer is the patient or the parent, is the ceiling this literature supports.

The vocabulary in this area is unsettled and worth naming plainly. Prodrome, premonitory symptom, and pre-ictal symptom are used somewhat interchangeably in the literature to describe features occurring before electrographic onset. None of them is an aura, and none of them has the localizing value an aura has.

The most rigorous work on this question is a prospective diary study published in Neurology in 2007 by Sheryl Haut, Richard Lipton, and colleagues. Seventy-one adults with localization-related epilepsy kept daily diaries, returning 15,179 complete diary days, and each night estimated the likelihood of a seizure the next day.

Three findings emerged. For each additional hour of sleep the previous night, the odds of a seizure the following day fell, with an odds ratio of 0.91 and a confidence interval of 0.82 to 0.99. Each one-unit increase in self-reported stress or anxiety on a ten-point scale was associated with increased seizure risk the next day, odds ratios of 1.06 and 1.07 respectively. And self-prediction itself carried an odds ratio of 3.7, with a confidence interval of 1.8 to 7.2.

The instructive detail is what happened when self-prediction was added to the model. Self-prediction and hours of sleep remained significant; stress and anxiety did not. In other words, a substantial part of what patients were sensing when they predicted a seizure appears to have been the same information the stress and anxiety measures captured, rather than something independent of it.

An odds ratio of 3.7 is a real signal and a poor forecast. It means a predicted day carried meaningfully higher risk than an unpredicted one. It does not mean a predicted seizure occurred, or that an unpredicted day was safe. No patient should adjust medication, driving, swimming, or bathing decisions on the strength of that number.

The assumption that seizures occur at random has not survived long-term brain recording either. A 2021 review in Nature Reviews Neurology by Philippa Karoly, Maxime Baud, and colleagues synthesized evidence from implanted devices, electronic seizure diaries, and animal neurophysiology and described cycles of epileptic brain activity operating on daily, multiday, and yearly timescales. Observations that seizures are cyclical date back to antiquity; what is new is the ability to quantify them in individual patients from direct brain recordings.

A 2023 study in Epilepsia by Nicholas Gregg, Benjamin Brinkmann, and colleagues at Mayo Clinic connected those brain cycles to signals that can be measured without surgery. Ten patients wore a multimodal wrist sensor recording heart rate, accelerometry, electrodermal activity, and temperature while an implanted responsive neurostimulation system recorded brain activity, for a mean of 232 days. Multiday cycles were present in every wearable signal in every subject. Among the seven patients with reliable electrographic seizure detections, seizure timing was phase locked to multiday cycles in temperature in five, in heart rate and phasic electrodermal activity in four, and in accelerometry and heart rate variability in three. After regressing behavioral covariates out of the heart rate signal, six of seven still showed phase locking to the residual.

This is the honest state of seizure forecasting: a real physiological substrate, individualized, measurable with consumer-grade hardware, and not yet a clinical tool. The word to hold onto is risk. These methods estimate whether a person is in a higher-risk or lower-risk phase. They do not say a seizure is coming.

Premonitory symptoms are not specific to epilepsy, and assuming they are can delay a correct diagnosis by years.

A 2022 case-control study in Seizure by Lindsay Stager, Aaron Fobian, and colleagues at the University of Alabama at Birmingham compared 26 children aged 13 to 18 with video-EEG-confirmed functional seizures to 26 matched controls. Ninety-eight percent of the patients with functional seizures endorsed premonitory symptoms. That single number should settle the question of whether a described warning sign distinguishes epileptic from functional events. It does not.

The same study found that children with functional seizures had slower reaction times on a modified Stroop task using seizure symptom words and a poorer sense of control on an experimental task, which the authors proposed as potential treatment targets. Functional seizures are a real and treatable condition, and they are not a diagnosis of exclusion made by asking about auras.

Other conditions produce similar pre-event experiences. Migraine aura, syncope with a prodrome of warmth and visual greying, panic attacks with derealization, and cardiac arrhythmia can each be described by a patient in language that sounds epileptic. The distinguishing work is done by history, semiology in sequence, electroencephalography, and where necessary prolonged video monitoring.

Cannabis comes up constantly in epilepsy clinics, and the evidence base is narrower than the conversation implies. It concerns one purified compound, at pharmaceutical dose, added to existing treatment, in specific severe syndromes.

The Dravet syndrome trial published in the New England Journal of Medicine in 2017 by Orrin Devinsky and colleagues randomized 120 children and young adults with drug-resistant seizures to cannabidiol oral solution at 20 mg per kilogram per day or placebo, in addition to standard antiepileptic treatment. Median monthly convulsive seizures fell from 12.4 to 5.9 on cannabidiol and from 14.9 to 14.1 on placebo, an adjusted median difference of 22.8 percentage points (95 percent CI 41.1 to 5.4; P = 0.01). Forty-three percent of the cannabidiol group achieved at least a 50 percent reduction compared with 27 percent on placebo, a difference that did not reach significance. Nonconvulsive seizures were not significantly reduced. Five percent became seizure-free versus none on placebo.

The Lennox-Gastaut trial published in The Lancet in 2018 by Elizabeth Thiele and colleagues randomized 171 patients to the same dose or placebo for 14 weeks. Median monthly drop seizure frequency fell 43.9 percent on cannabidiol and 21.8 percent on placebo, an estimated median difference of 17.21 percentage points (95 percent CI 30.32 to 4.09; p = 0.0135). Adverse events occurred in 86 percent of the cannabidiol group and 69 percent of the placebo group, and 14 percent of the cannabidiol group withdrew for adverse events compared with 1 percent on placebo.

Three facts follow from those trials and belong in every conversation about them. Every participant stayed on their existing anticonvulsants; cannabidiol was tested as an addition, never as a replacement. The measured benefit was on convulsive and drop seizure frequency, and no trial has tested whether any cannabinoid reduces auras, aborts a focal aware seizure in progress, or affects prodromal symptoms. And the product studied was a purified pharmaceutical cannabidiol at a weight-based dose, which is not what is sold in a dispensary.

A randomized dose-ranging safety trial published in Neurology in 2018 by Orrin Devinsky and colleagues studied 34 children with Dravet syndrome across 5, 10, and 20 mg per kilogram per day. Exposure was dose-proportional. Cannabidiol did not affect concomitant antiepileptic drug levels with one exception: it raised N-desmethylclobazam, the active metabolite of clobazam, an interaction the authors attributed to inhibition of cytochrome P450 2C19. Six patients taking cannabidiol together with valproate developed elevated transaminases. None met criteria for drug-induced liver injury and all recovered.

A 2022 practical guide in CNS Drugs by Adam Strzelczyk and Susanne Schubert-Bast describes the clobazam interaction as bidirectional, raising both 7-hydroxy-cannabidiol and N-desmethylclobazam, with the potential for increased somnolence and sedation, and reiterates that transaminase elevations occur particularly with concomitant valproate. In the European Union, cannabidiol is approved for Dravet syndrome in combination with clobazam.

The practical implication is that cannabidiol in epilepsy is a drug with a drug’s obligations: baseline and follow-up liver function testing, awareness of the clobazam interaction, attention to sedation, and management by a clinician who knows the patient’s full regimen. Self-directed use of over-the-counter cannabidiol products alongside anticonvulsants carries the same interaction risks without the monitoring.

The sentence that matters most in this entire page: nothing in this literature supports reducing or stopping an anticonvulsant. Withdrawal of antiseizure medication without neurological supervision can precipitate prolonged seizures and status epilepticus, and no cannabis product has been shown to substitute for that medication.

The classification framework is consensus rather than experiment, which the International League Against Epilepsy states openly: the 2017 scheme is operational and practical because current knowledge is insufficient to support a scientifically derived classification. It is the right vocabulary to use and it should not be mistaken for a biological discovery.

The cannabidiol evidence is genuinely strong within a narrow scope. Two adequately powered randomized placebo-controlled trials with prespecified primary endpoints, in defined syndromes, at a defined dose, as add-on therapy, is a high standard of evidence. It applies to exactly those conditions and that product, and it does not generalize to other epilepsies, other cannabinoids, or other formulations.

The prediction literature is the weakest of the three. Diary studies and caregiver questionnaires depend on self-report, and prediction accuracy was never established prospectively against a blinded standard. The device-based cycle work is stronger methodologically and involves very small numbers.

Self-prediction studies have an unavoidable circularity problem. A patient who predicts a seizure may alter behavior, sleep, medication timing, or stress exposure in response to that prediction, which changes the outcome being measured. The Montefiore diary study’s own finding that stress and anxiety lost significance once self-prediction entered the model is consistent with prediction partly encoding those same variables.

The caregiver study in children asked caretakers to recall whether they could predict seizures, a design that selects for memorable successes and forgets the misses. Twenty-one percent reporting the ability is not the same as twenty-one percent demonstrating it.

In the cannabidiol trials, the adverse event profile, particularly somnolence, diarrhea, and elevated liver enzymes, was distinctive enough that functional unblinding is a reasonable concern in trials whose primary endpoint is a caregiver-recorded seizure count. This is an acknowledged difficulty across antiseizure drug trials and not specific to cannabidiol.

No study cited here shows that any cannabinoid prevents, shortens, or aborts an aura, alters prodromal symptoms, or improves the accuracy of seizure prediction. Those questions have not been asked in a controlled trial.

Nothing here supports whole-plant cannabis, dispensary CBD products, or THC-containing preparations as antiseizure treatment. The trials used a purified pharmaceutical cannabidiol at 20 mg per kilogram per day. Nothing here supports reducing or discontinuing an anticonvulsant, and no trial tested cannabidiol as anything other than an addition to existing therapy.

The wearable and cycle research does not establish a validated forecasting device. It establishes that physiological cycles exist, that they are individual, and that seizure timing is phase locked to them in some patients.

Epilepsy care has been moving for a decade from a model of random events toward a model of fluctuating risk, and the pre-seizure literature is where that shift becomes visible to patients. Multiday cycles, sleep, stress, and individual physiological rhythms all contribute to whether today is a higher-risk day. That is a more useful mental model than waiting for a warning sign, and it points toward chronotherapy rather than rescue.

The cannabidiol story sits alongside that as a reminder of what an adequate evidence base looks like. It took randomized placebo-controlled trials in defined syndromes with prespecified endpoints to establish a modest add-on benefit, and even then the effect on nonconvulsive seizures was not significant. That is the standard against which every other cannabis claim in neurology should be measured.

The most valuable thing a patient with epilepsy can bring to an appointment is a careful account of the first few seconds. Not the collapse, not what the room looked like afterward, but the exact order of the first sensations. That sequence is diagnostic information nobody else can provide, and most people have never been asked for it that way.

On cannabis, I try to be equally specific. There is a real, trial-supported role for pharmaceutical cannabidiol as an addition to treatment in two severe syndromes, with liver monitoring and attention to the clobazam interaction. There is no evidence for cannabis touching auras, prodromes, or prediction, and there is no version of this where a cannabis product replaces an anticonvulsant. I have never met a patient who regretted staying on their medication while we figured something out. I have met people who regretted the opposite.

An aura is the start of a focal aware seizure, not a warning before one, and describing it in sequence helps localize where seizures begin. A prodrome is a separate, non-localizing phenomenon reported by roughly one in five patients or caregivers. Prediction is real but weak, on the order of a threefold change in odds, and is not a basis for altering medication or safety behavior. Purified cannabidiol has trial-supported add-on benefit in Dravet and Lennox-Gastaut syndromes and has never been shown to substitute for anticonvulsant therapy.

The findings to carry forward are diagnostic: the aura is part of the seizure, its sequence localizes, the prodrome does not, and premonitory symptoms do not distinguish epileptic from functional events. The claims to set aside are that a warning sign constitutes reliable prediction, and that any cannabis product has a role in the pre-seizure period. Neither has been demonstrated.