Seizure Management

EEG Fundamentals taught you to read the record. This course teaches what to do for the patient: how seizures and epilepsy are evaluated and treated, and the evidence behind each step.

About 4 hours 55 minutes28 short lessons · read in order or choose a topic from Contents

Each lesson explains the reasoning, shows the trial or guideline that supports it, links back to the electroencephalography (EEG) lesson it builds on, and ends with a quick check. It is written for neurology residents; fellows can use single lessons for review.

Course topics

1. The first seizureEvaluation, and when to treat
2. DiagnosisSemiology, the cause, functional seizures, seizures early in life
3. Antiseizure medicationsChoosing, mechanisms, profiles, levels
4. Status epilepticusThe clock, treatment, continuous EEG
5. Drug-resistant epilepsyReferral, evaluation, surgery and devices
6. Stopping treatmentWhen and how to discuss it
7. Living with epilepsyClusters and rescue medicines, sudden death, safety
8. Women's health and hormonesCycle, contraception, pregnancy, menopause

Written for teaching by the NeuroLogic team from primary studies and guidelines, cited in each lesson. Reviewed October 2026. Use local protocols and current prescribing information for patient care.

A first event raises three questions, in order: was it an epileptic seizure, was it provoked, and what does the workup show? The answers decide whether this is epilepsy and whether to treat.

1. Was it an epileptic seizure?

Most of the answer comes from the history, ideally from someone who saw the event. Transient ischemic attack, migraine aura, parasomnias and hypoglycemia can all be mistaken for seizures. Two mimics deserve particular attention:

2. Was it provoked?

An acute symptomatic, or provoked, seizure happens at the time of an acute brain insult or systemic disturbance. It is treated by treating the cause, and it does not count toward a diagnosis of epilepsy. The International League Against Epilepsy (ILAE) definitions, written for epidemiological research, with metabolic cutoffs chosen to favor specificity, include, among others (Beghi 2010):

• within 7 days of a stroke, traumatic brain injury, intracranial surgery or anoxic encephalopathy, or during an active infection of the central nervous system;
• 7 to 48 hours after the last drink in alcohol withdrawal in someone with chronic alcohol use, or during withdrawal from benzodiazepines or barbiturates; also intoxication or exposure to a drug that provokes seizures;
• with a severe metabolic disturbance in blood drawn within 24 hours: glucose below 36 mg/dL, or above 450 mg/dL with ketoacidosis; sodium below 115 mmol/L; calcium below 5 mg/dL; magnesium below 0.8 mg/dL; urea nitrogen above 100 mg/dL; or creatinine above 10 mg/dL.

A seizure with a milder metabolic disturbance does not meet the definition of acute symptomatic, but the report does not class it as unprovoked either: it falls into an "unknown" category that does not count toward epilepsy. A seizure months or years after a stroke or brain injury is not provoked: it is a remote symptomatic seizure, and it may already mean epilepsy, as the next lesson explains.

3. The workup

The 2007 American Academy of Neurology parameter on evaluating a first seizure (the source of the 23% and 10% yields) was retired in 2018; its figures remain the clearest summary of what each test finds. The emergency neuroimaging review (Harden 2007) was reaffirmed in 2025.

Sources: as cited. Seizure terms follow the ILAE classification as updated in 2025 (Beniczky 2025).

After one unprovoked seizure, the question is how likely a second one is. That probability decides both the diagnosis of epilepsy and the case for starting treatment.

The ILAE definition of epilepsy

The International League Against Epilepsy (ILAE) practical definition says a person has epilepsy after any of (Fisher 2014):

1. at least two unprovoked (or reflex) seizures more than 24 hours apart;
2. one unprovoked (or reflex) seizure and a probability of further seizures similar to the risk after two unprovoked seizures, at least 60%, over the next 10 years;
3. the diagnosis of an epilepsy syndrome.

The 60% is a guide, not a sharp line; after two unprovoked seizures the risk of more is about 60 to 90%. The report's example for the second criterion is a single unprovoked seizure at least a month after a stroke, and it warns that one seizure with a lesion, or with epileptiform discharges, does not automatically qualify. Seizures clustered within 24 hours count as one. Epilepsy is resolved in someone who has outgrown an age-dependent syndrome, or who has been seizure-free for 10 years and off medication for the last 5.

How likely is a second seizure?

Level A, B, C: the American Academy of Neurology grades each guideline recommendation by the evidence behind it, from A (the strongest) to C (the weakest).

After a first unprovoked seizure in an adult, the risk of another is highest early: 21 to 45% within 2 years (Level A). These figures come from cohorts that mixed treated and untreated people, and the risk appeared lower in those treated (Krumholz 2015). After a second unprovoked seizure, about three-quarters of people had another within 4 years (Hauser 1998).

What immediate treatment does, and does not, do

The American Academy of Neurology and American Epilepsy Society guideline sums it up: immediate treatment lowers the absolute risk of recurrence over the next 2 years by about 35%, the guideline's estimate (Level B), is unlikely to improve the chance of long-term remission (Level B), may not improve quality of life (Level C), and causes adverse effects in 7 to 31%, mostly mild and reversible (Level B) (Krumholz 2015).

Making the decision together

Treatment after a first seizure is a choice, not a rule. It weighs the individual recurrence risk against the burden of daily medication and its adverse effects, and it depends on what a second seizure would cost this person: driving, work, living alone, pregnancy plans. Many people choose to wait; others, especially those who drive for work, choose to start.

Sources: as cited. The 2015 guideline was reaffirmed by the American Academy of Neurology in 2024.

A seizure's signs, read in the order they appear, show where the discharge is and where it goes. Before any test, they often tell which lobe, and which side, the seizures come from.

Describe the whole sequence

The International League Against Epilepsy (ILAE) seizure classification, updated in 2025, sorts a focal seizure by consciousness, preserved or impaired, judged from two things: whether the person recalls the seizure (awareness) and whether they respond normally during it (responsiveness). If either is impaired, consciousness is impaired. In its expanded version the seizure is then described by listing its signs in the order they appear, joined by arrows, because the sequence characterizes the epileptic network better than the first sign alone (Beniczky 2025).

Two rules sit behind localization. A sign shows where the discharge is when the sign appears, the symptomatogenic zone, which may lie some way from where the seizure began (the zones are set out in Presurgical evaluation). And primary cortex produces elementary symptoms while association cortex produces complex ones: among 20 surgically treated patients with visual auras, complex visual hallucinations never came from occipital lobe seizures, while elementary hallucinations, illusions and visual loss came from occipital, occipitotemporal and anteromedial temporal onsets alike (Bien 2000).

Where the first sign points

Signs that tell the side

Some signs point to the hemisphere where the seizure starts. They differ in how often they appear, how well observers agree on them and how often they are right.

Signs are stronger together. In a blinded review of 47 seizures that became bilateral, version, unilateral tonic or clonic activity, an asymmetric end to the clonic phase and Todd paresis each pointed to the correct side more than 80% of the time, while hand dystonia and the figure-of-4 were less reliable in that setting. When two or more reliable signs agreed, the side was correct in every patient, but no sign separated temporal from extratemporal onset (Marashly 2016).

Sources: as cited. Seizure terms follow the ILAE classification as updated in 2025 (Beniczky 2025).

From the first seizure on, the question is not only whether this is epilepsy but what causes it. The cause often decides treatment: a lesion for surgery, a diet or a vitamin that treats the defect itself, immunotherapy, or a medication that must be avoided.

Six groups of causes

The International League Against Epilepsy (ILAE) classification of the epilepsies asks for the etiology at every step, from the first seizure on, in six groups chosen for their treatment consequences: structural, genetic, infectious, metabolic, immune and unknown. The groups are not exclusive: tuberous sclerosis complex is structural, which matters for surgery, and genetic, which matters for counseling and for mechanistic target of rapamycin (mTOR) inhibitors (Scheffer 2017).

Groups, definitions and examples from the classification (Scheffer 2017); glucose transporter 1 deficiency from its consensus statement (Klepper 2020).

The scan, and who reads it

Who should have genetic testing

The National Society of Genetic Counselors guideline strongly recommends genetic testing for everyone with unexplained epilepsy, at any age: epilepsy not attributed to an acquired cause such as trauma or infection. First-tier testing is exome or genome sequencing, or a multigene panel of more than 25 genes, followed by chromosomal microarray, with exome or genome sequencing conditionally preferred over a panel. Tests should be chosen, ordered and interpreted by a qualified clinician with genetic counseling before and after (Smith 2022).

Counseling matters because a result changes more than the label. A de novo variant means the parents are unlikely to have epilepsy, but the patient may pass it on: with a dominant variant, each child has a 50% chance of inheriting it, though whether the child has epilepsy depends on its penetrance (Scheffer 2017).

Results that change treatment

Some results change the next prescription. The same drug class can harm one genetic epilepsy and treat another.

Once a pyridoxine trial has started, it is continued until biochemical or genetic testing has excluded pyridoxine-dependent epilepsy, so that a late or masked response is not missed (Stockler 2011).

Immune causes

Sources: as cited. Etiology groups follow the 2017 ILAE classification of the epilepsies (Scheffer 2017).

Functional/dissociative seizures look like epileptic seizures but arise without an epileptic discharge. The diagnosis rests on positive features, recorded on video-electroencephalography (video-EEG) when possible, and it leads to its own treatment. Missing it is not harmless: in the Established Status Epilepticus Treatment Trial (ESETT), 10% of the patients enrolled for benzodiazepine-refractory status epilepticus had functional/dissociative seizures (Kapur 2019).

The name

In 2025 an International League Against Epilepsy (ILAE) task force proposed functional/dissociative seizures for what were called psychogenic nonepileptic seizures, to replace a range of labels that confused professionals and patients. The slash lets the clinician use "functional", "dissociative" or both, depending on the patient and the language (Hingray 2025). Studies cited here used the older term.

How certain is the diagnosis?

Video-EEG is the standard, but it is not available everywhere or for every patient, so the ILAE Nonepileptic Seizures Task Force set out four levels of diagnostic certainty. Each needs a history consistent with the diagnosis; they differ in who saw an event and what the EEG showed (LaFrance 2013).

The qualifier in the third level matters. Focal seizures with preserved consciousness often show nothing on the scalp EEG: only 18 of 87 (21%) did in one video-EEG series (Devinsky 1988). So a normal EEG during an event counts against epilepsy only when the event is one that would show an epileptic discharge (LaFrance 2013).

Signs at the bedside, weighed

A review of 34 video-EEG studies found good support for a set of signs; it also concluded that the diagnosis should rest on all the evidence, never on one sign alone (Avbersek 2010).

Giving the diagnosis, and treatment

When epilepsy and functional seizures coexist

Reports of how many patients with functional seizures also have epilepsy have ranged from 10% to more than 50%. In one video-EEG unit, 3 of 32 patients diagnosed in a year (9.4%) had interictal epileptiform discharges supporting coexisting epilepsy (Benbadis 2001). This is why antiseizure medications are withdrawn only once coexisting epilepsy has been excluded, and why follow-up continues after withdrawal (Oto 2005).

Sources: as cited. The term follows the 2025 ILAE task force proposal (Hingray 2025).

The two commonest settings for a seizure early in life ask opposite things. In a neonate, look harder: most seizures show no clinical sign, so the diagnosis rests on the electroencephalogram (EEG). After a brief generalized seizure with fever in a toddler, look less: the evaluation is of the fever, not of the seizure.

Seizures in the newborn

Seizures in the neonatal period are often provoked by an acute cause and may be electrographic only. The International League Against Epilepsy (ILAE) classification for neonates therefore rests the diagnosis on EEG: seizures with no clinical feature are included, and a clinical event without an EEG correlate is not classified as a seizure. Because neonatal seizures have a focal onset, there is no division into focal and generalized; a seizure is named by its predominant feature, motor (automatisms, clonic, epileptic spasms, myoclonic, tonic), nonmotor (autonomic, behavior arrest) or sequential (Pressler 2021).

Watching is not enough. In term infants recorded with video-EEG from the first hours of life, only 34% of 526 electrographic seizures had any clinical sign on the video, and only 27% of 177 events the nursery staff suspected were seizures on EEG; in all, only 9% of the electrographic seizures were recognized and documented (Murray 2008).

The ILAE treatment guideline makes six recommendations (Pressler 2023):

Febrile seizures

A febrile seizure is a seizure with fever, a temperature of at least 38 °C by any method, without infection of the central nervous system, in a child aged 6 to 60 months; it affects 2 to 5% of children. It is simple when it is generalized, lasts under 15 minutes and does not recur within 24 hours, and complex when it is prolonged, focal or recurs within 24 hours (American Academy of Pediatrics, 2008 and 2011).

Sources: American Academy of Pediatrics 2011 (evaluation); American Academy of Pediatrics 2008 (long-term management).

What parents ask next is whether it will happen again, and whether it is epilepsy. About 30% of children have a second febrile seizure when the first comes after 12 months of age, and about 50% when it comes before. After simple febrile seizures, the risk of epilepsy by age 7 is about the same as in the general population, 1%; children who have had several simple febrile seizures, were younger than 12 months at the first and have a family history of epilepsy are at higher risk, with afebrile generalized seizures by age 25 in 2.4% (American Academy of Pediatrics 2008).

Sources: as cited. Neonatal seizure terms follow the ILAE neonatal classification (Pressler 2021); febrile seizure terms follow the American Academy of Pediatrics.

The first drug is chosen in three steps: classify the seizures and the epilepsy, match the drug's spectrum to them, then fit it to the person who will take it.

1. Classify the seizures and the epilepsy

Focal or generalized seizures, and if possible the epilepsy syndrome, using the classification of the International League Against Epilepsy (ILAE classification). The history and the electroencephalogram (EEG) do most of this work.

2. Match the spectrum

Some drugs treat both focal and generalized seizures; others treat focal seizures only, and ethosuximide treats absence seizures only. For a generalized epilepsy, the wrong spectrum can make things worse. The National Institute for Health and Care Excellence (NICE) lists carbamazepine, oxcarbazepine, phenytoin, gabapentin, pregabalin, tiagabine and vigabatrin as drugs that may exacerbate absence and myoclonic seizures, including in juvenile myoclonic epilepsy (and phenobarbital for absence seizures), advises against them for myoclonic seizures, and notes that lamotrigine can occasionally worsen myoclonic seizures (NICE NG217).

Case series show what happens when a generalized epilepsy is misclassified as focal:

3. Fit the drug to the person

What the trials show

NICE suggests considering lamotrigine or levetiracetam first for focal seizures (a weaker "consider" recommendation); it offers a choice of lamotrigine, levetiracetam or valproate for generalized tonic–clonic seizures, within the valproate rules above; ethosuximide first for absence seizures; and levetiracetam or valproate for myoclonic seizures (NICE NG217).

What to expect

In a Glasgow cohort of 1,795 people followed for up to 30 years after starting treatment, 50.5% became seizure-free for at least a year on their first regimen. The second regimen added 11.6% and the third 4.4%; all later regimens together added about 2%. At the end, 63.7% were seizure-free, so more than a third were not, and outcomes had not improved over three decades of new drugs (Chen 2018). Why the first drug failed matters: in the original report from the same clinic, 41% of people whose first drug failed because of adverse effects later became seizure-free, against 11% when it failed for lack of efficacy (Kwan 2000).

Sources: as cited. NICE NG217 is quoted as updated in August 2026; US labeling does not carry the UK valproate age rule.

Where a drug acts explains much of its clinical profile: which seizures it treats, which it can make worse, and which adverse effects to expect.

Seizures start when excitation outruns inhibition in a network. Antiseizure medications shift that balance back in one of four broad ways:

A fifth target explains a whole seizure type. Thalamic relay neurons fire in bursts through T-type calcium channels, and that burst firing drives the thalamocortical loop behind generalized spike-and-wave. Ethosuximide reduces the T-type current, which is why its use is limited to absence seizures; its label warns that, used alone in mixed epilepsies, it can increase tonic–clonic seizures.

Choose a target or a drug in the model to see where it acts.

Every antiseizure medication used in practice, with its mechanism and the adverse effects that shape the choice. Read it as a set of profiles: spectrum, target and the safety problems to know before prescribing.

Some drugs are broad spectrum (they treat focal and generalized seizures), some treat focal seizures only, and a few are reserved for one syndrome.

Adverse effects and boxed warnings from the US prescribing information, linked from each drug's name. Every antiseizure medication also carries a class warning for suicidal thoughts and behavior. Mechanisms from the labels where they state one; otherwise from the literature (Sills 2020; valproate and the thalamic calcium current, Coulter 1989; eslicarbazepine and slow inactivation, Hebeisen 2015; brivaracetam binding, Gillard 2011; cannabidiol, Gray 2020; fenfluramine and sigma-1, Martin 2020). Boxed warnings and monitoring are gathered in the next lesson, and drug levels in Drug levels and interactions.

A boxed warning is the strongest warning in US prescribing information, and each one carries an action: a test before the first dose, monitoring during treatment, a limit on titration or infusion rate, or a restricted program. After the boxed warnings come the organ effects that examinations return to, grouped by system.

The boxed warnings

Oxcarbazepine, eslicarbazepine, topiramate, zonisamide, lacosamide, levetiracetam, cenobamate, rufinamide, ethosuximide, cannabidiol, stiripentol and ganaxolone carry no boxed warning. The warning on suicidal thoughts and behavior covers every antiseizure medication, but it sits in Warnings and Precautions, not in a box (see Medication profiles).

Organ effects by system

Sources: as cited; boxed warnings and warnings from the current US prescribing information on DailyMed, checked October 2026.

Once an epilepsy syndrome is recognized, the syndrome rather than the seizure type chooses the medicine. Several drugs are approved by the US Food and Drug Administration (FDA) for one syndrome only, and in some syndromes a drug chosen by seizure type alone makes the seizures worse.

Infantile epileptic spasms syndrome

Dravet syndrome: no sodium-channel blockers

More than 85% of people with Dravet syndrome carry a pathogenic variant in SCN1A, which leaves them with too little of the Nav1.1 sodium channel (Wirrell 2022). In a mouse model, losing Nav1.1 reduced the sodium current in inhibitory interneurons but not in excitatory pyramidal neurons, which points to failing inhibition as the cause of the seizures (Yu 2006).

Lennox–Gastaut syndrome

Tuberous sclerosis complex and CDKL5 deficiency disorder

Childhood absence and juvenile myoclonic epilepsy

Both are generalized epilepsies, so the drugs that worsen absence and myoclonic seizures are the ones to avoid (see Choosing a first antiseizure medication). Since January 2025, NICE follows the UK medicines regulator: valproate is not started in anyone under 55, of either sex, unless two specialists independently agree, and in women and girls who could become pregnant only if other treatments fail, with the Pregnancy Prevention Programme (NICE NG217, box 2).

Sources: as cited; indications and ages from the current US labels on DailyMed. The 2012 infantile spasms guideline has since been retired by the American Academy of Neurology and is cited for its evidence review; NICE NG217 recommendations are numbered as on the NICE website.

A drug level answers a specific question; it is not a target to chase. And because several antiseizure medications induce or inhibit liver enzymes, they change the levels of each other and of many other drugs.

When a level helps

The International League Against Epilepsy (ILAE) position paper on drug monitoring lists when a level is most likely to help (Patsalos 2008):

• Once seizures are controlled, to record the patient's own effective level, so that later changes can be interpreted.
• To diagnose suspected toxicity.
• To check adherence, especially after a breakthrough seizure.
• When pharmacokinetics vary: children, older adults, other illness, a change of formulation.
• When a large change is expected: pregnancy, or an interacting drug added or stopped.
• For drugs whose kinetics depend on the dose, above all phenytoin.

The laboratory reference range is not a therapeutic range. It marks a population band below which a response is less likely and above which toxicity is more likely, and some patients are controlled below it while others need and tolerate levels above it. Treat the patient, not the level (Patsalos 2018).

Total and free levels

Laboratories usually report the total concentration, bound and unbound together, but only the unbound drug acts. Phenytoin is about 92% bound to plasma proteins, mostly albumin (Patsalos 2018). With low albumin, or with kidney or liver disease, the unbound fraction rises, so a total level inside the range can hide toxicity; the label advises monitoring unbound phenytoin in these patients (phenytoin label). In kidney failure the cause is weaker binding: in serum from patients with chronic kidney failure treated conservatively, measured in the laboratory, the free fraction of phenytoin was about twice normal at a creatinine near 800 µmol/L (about 9 mg/dL) and up to 4 to 5 times normal in uremia (Tiula 1987). Valproate both displaces phenytoin from albumin and slows its metabolism, and valproate's own binding saturates, so its free fraction rises at higher concentrations and with low albumin (valproate label).

Phenytoin's metabolism also saturates, at a level that differs from patient to patient and is sometimes within the usual range, so a modest increase in dose can cause a very large rise in the level, and toxicity (phenytoin label; Patsalos 2018).

Enzyme inducers and inhibitors

Interactions worth knowing by name

Kidney and liver

Drugs cleared mainly by the kidney accumulate as creatinine clearance falls, and several are removed by hemodialysis. Drugs metabolized by the liver need care in hepatic impairment. The labels give the doses; the patterns are what to know.

Sources: as cited; labels are the current US prescribing information on DailyMed.

Status epilepticus is defined by two clocks. The first says when to start treating; the second, when there is a risk of lasting injury. Both depend on the type of seizure.

The “five minute rule” is not a rule about seizures in general — it is t₁ for convulsive status only. The International League Against Epilepsy (ILAE) 2015 operational definition sets two separate clocks, and they differ by seizure type:

• t₁ — the point beyond which the seizure is unlikely to stop on its own. This is when emergency treatment starts.
• t₂ — the point beyond which there is a risk of long-term consequences: neuronal injury and death, and altered neuronal networks. This is what you are racing.

Use the clocks correctly: these operational estimates vary by seizure type and do not establish a safe waiting period. Clinical deterioration, recurrent seizures without recovery, or uncertain onset requires prompt assessment and treatment. The time associated with injury is less certain in nonconvulsive status.

Trinka E, Cock H, Hesdorffer D, et al. A definition and classification of status epilepticus — Report of the ILAE Task Force on Classification of Status Epilepticus. Epilepsia 2015;56(10):1515–23.

Types of status epilepticus

The same report classifies status along four axes: its semiology, its etiology, its electroencephalographic (EEG) correlates and the patient's age. The first axis divides status into two groups that are treated differently:

Treatment is a sequence run against the clock: stabilize, give a benzodiazepine early and in full, add a second-line medication if convulsions continue, and escalate to anesthesia if that fails. The common errors are delay and underdosing.

The minute marks differ a little between guidelines; the American Epilepsy Society guideline gives phases of stabilization from 0 to 5 minutes, initial therapy from 5 to 20, second therapy from 20 to 40 and third therapy from 40 to 60 minutes (Glauser 2016). The order is the same in all of them, and the times are targets, not waiting periods.

Why the benzodiazepine must be early and full

In the American Epilepsy Society guideline, a benzodiazepine is the initial therapy of choice (Level A). In adults, the second therapy is less effective than the first and the third substantially less (Level A); in children, the second also appears less effective, on weaker evidence (Level C) (Glauser 2016). Two prehospital trials show why it should come early:

In practice, benzodiazepines are often given late and in too small a dose:

Choosing the second-line medication

None of the three was superior; they had similar efficacy, so the choice is made by the patient's other conditions and by local practice. The labels set the limits:

Sources: treatment phases and evidence levels, Glauser 2016; clocks, Trinka 2015 (previous lesson); trials as cited above. The clock follows the American Epilepsy Society phases; for doses, see the Emergency Neurological Life Support status epilepticus protocol (version 6.0, 2024) or your institution's protocol.

Stopping the movements is not the same as stopping the seizure. And when first- and second-line treatment both fail, the problem moves to the intensive care unit, where the EEG becomes the way to follow treatment.

When consciousness does not return

In a prospective study of 164 patients monitored with continuous electroencephalography (EEG) for at least 24 hours after their convulsive status had been controlled, 48% still had electrographic seizures and more than 14% were in nonconvulsive status epilepticus; these patients were comatose, with no visible sign of convulsions (DeLorenzo 1998). The American Clinical Neurophysiology Society (ACNS) consensus therefore recommends continuous EEG when mental status stays abnormal after convulsive status, and suggests considering it when alertness has not clearly improved within 10 minutes, or consciousness is still impaired 30 minutes after the clinical seizures stop (Herman 2015).

So when a patient treated for convulsive status is not returning toward baseline, arrange an urgent EEG rather than attributing the coma to the medications. Subtle signs, such as nystagmoid eye jerks or subtle clonic movements of the face or a limb, can be the only clinical trace of ongoing seizures.

Refractory and super-refractory status

Status that continues after an initial benzodiazepine and a second antiseizure medication is refractory: it is defined by failed treatment, not by the clock (Brophy 2012; Herman 2015). Status that continues or returns 24 hours or more after anesthesia began, including when the anesthetic is being reduced, is super-refractory (Shorvon 2011).

The principles of care in the intensive care unit:

When there is no obvious cause

The practical point is to search early for autoimmune and infectious causes, because some of them can be treated. An international consensus advises first-line immunotherapy within 72 hours of onset and, if seizures continue, second-line immunotherapy and the ketogenic diet early (Wickstrom 2022).

Sources: as cited. The Neurocritical Care Society guideline (Brophy 2012) also sets a goal of definitive control within 60 minutes of onset.

In the intensive care unit most seizures have no visible sign. Continuous electroencephalography (EEG) finds them, and it also finds patterns that are neither clearly seizures nor clearly harmless. Deciding what to do with those is where EEG reading turns into management.

Who needs continuous EEG, and for how long

In 570 critically ill patients monitored with continuous EEG, 19% had seizures, and in 92% of those every seizure was nonconvulsive. The first seizure came within 24 hours in 88%, but it came later in 20% of comatose patients against 5% of those not in coma (Claassen 2004). A routine 30 to 60 minute EEG finds only about half (45 to 58%) of the patients who go on to have seizures (Herman 2015).

Seizure, status or continuum: the definitions

The ACNS 2021 terminology separates what is certainly a seizure from what might be (Hirsch 2021):

Not every pattern carries the same risk. In 4,772 critically ill adults, lateralized periodic discharges were most strongly associated with seizures at every frequency, especially with plus features; generalized periodic discharges and lateralized rhythmic delta became riskier as they got faster; and generalized rhythmic delta carried no more risk than no pattern at all (Rodriguez Ruiz 2017).

Diagnosing nonconvulsive status: the Salzburg criteria

For patients without a known epileptic encephalopathy, the Salzburg criteria turn the definitions into a diagnosis (Beniczky 2013; Leitinger 2015):

In the validation study's 120 patients with suspected nonconvulsive status, the criteria had a sensitivity of 97.7% and a specificity of 89.6%; in 100 controls with abnormal EEGs the specificity was 97.0% (Leitinger 2016).

Does treating the pattern help?

These trials do not show that seizures in critically ill patients are harmless. They show that finding and suppressing a pattern does not by itself change the outcome, which depends at least as much on the cause and the patient.

A reasonable approach to a continuum pattern

No trial tells you exactly what to do, so the approach is a judgment that weighs the pattern against the patient:

1. Describe it precisely. Name, frequency, prevalence, plus modifiers and any evolution or fluctuation. Higher frequency, plus features and lateralized patterns carry more seizure risk.
2. Look for a clinical correlate: a focal deficit, fluctuating responsiveness or subtle clonic movements time-locked to the pattern.
3. Consider a diagnostic medication trial while watching the EEG and the examination. Improvement in both supports treating the pattern as nonconvulsive status.
4. Treat the cause and limit harm. A non-sedating antiseizure medication is a common choice; escalating to anesthesia for a continuum pattern alone is weighed carefully against its risks.

Sources: as cited. Pattern names follow the ACNS 2021 terminology taught in EEG Fundamentals.

About a third of people with epilepsy keep having seizures despite medication. Recognizing that early, and referring, is the most useful thing a general neurologist can do for them.

The International League Against Epilepsy (ILAE) defines drug-resistant epilepsy as failure of adequate trials of two antiseizure medication schedules to achieve sustained seizure freedom. The schedules must have been tolerated, appropriately chosen and appropriately used, alone or in combination (Kwan 2010).

• Appropriate: a drug with proven efficacy for the patient's seizure type, preferably in randomized trials.
• Adequate: enough of it, for long enough.
• Seizure freedom: no seizures for three times the longest interval between seizures before the change, or 12 months, whichever is longer.

Why two medications are enough to call it

Because each drug that fails makes the next one less likely to work. In 1,795 people followed from their first treatment, the first regimen made 50.5% of the whole cohort seizure-free, the second a further 11.6% and the third a further 4.4%; every later regimen together added about 2%. Failure of the first drug raised the odds that each later regimen would fail too (Chen 2018).

Rule out pseudoresistance

Some apparent drug resistance is something else. Before calling it, check four things, which the epilepsy center will also review:

Why refer early

The evidence for surgery is strongest when it comes early, yet patients still reach it late. At one epilepsy surgery center, the mean time from diagnosis to referral for temporal lobe epilepsy was about 17 years in 1995–1998 and about 19 years in 2005–2008, no shorter after the 2001 randomized trial and the 2003 practice parameter below (Haneef 2010).

The American Academy of Neurology practice parameter (Level A) says that patients with disabling focal seizures with impaired consciousness who have failed appropriate first-line drugs should be considered for referral to an epilepsy surgery center (Engel 2003). The 2022 ILAE expert consensus goes further: referral for a surgical evaluation should be offered to every patient with drug-resistant epilepsy up to 70 years of age as soon as drug resistance is established, whatever the seizure type, epilepsy type, duration or comorbidities (Jehi 2022). Referral is for an evaluation; it does not commit anyone to surgery.

Uncontrolled seizures also carry a risk of death. Sudden unexpected death in epilepsy (SUDEP) affects about 1 in 1,000 adults and 1 in 4,500 children with epilepsy each year. Generalized tonic–clonic seizures are its major risk factor, and the risk rises with their frequency. The American Academy of Neurology and American Epilepsy Society guideline asks clinicians to give patients, or the parents of young children, these numbers; to keep working to reduce seizures in anyone who still has tonic–clonic seizures; and to explain that freedom from seizures, particularly tonic–clonic seizures, is strongly associated with a lower risk (Level B; Harden 2017).

Sources: as cited; the pathway figure follows Kwan 2010.

Before anyone operates, the epilepsy team has to agree on where the seizures begin and what that cortex does. No single test answers either question.

What the evaluation looks for

The aim is to find the epileptogenic zone, the cortex whose removal or disconnection stops the seizures, and to map the function around it. No test shows it directly. Each test defines a different zone, the team infers the epileptogenic zone from where they overlap, and it is only proven when the patient is seizure-free after surgery (Rosenow & Lüders 2001).

Stereo EEG (SEEG)

When the noninvasive results disagree, show no lesion, suggest a deep or bilateral onset, or point to cortex near language or motor areas, the team states a hypothesis about where seizures start and how they spread, and tests it with depth electrodes. Each lead passes through a small drill hole along a trajectory planned on MRI and vascular imaging to avoid vessels, and is placed with a stereotactic frame or robot. About a dozen leads is typical: 902 electrodes in 76 implantations in one series (van der Loo 2017).

For drug-resistant epilepsy the choice is between removing or disconnecting the source, modulating the network, and dietary or syndrome-specific treatment. Only removing, ablating or disconnecting a localized source regularly brings freedom from seizures.

Remove or disconnect the source

Modulate the network

All three are palliative: the benefit grows over years and seizure freedom is uncommon. Diathermy is contraindicated with each device, and MRI is possible only under the manufacturer's conditions for the implanted system.

Diet and syndrome-specific treatment

Neuromodulation devices: where they sit

Pivotal trials: seizure reduction, active versus control

What the devices look like

Sources: definition, International League Against Epilepsy (ILAE) task force (Kwan 2010); vagus nerve stimulation, American Academy of Neurology guideline update (Morris 2013); device indications from the US Food and Drug Administration approvals and the manufacturers' labeling; trial results as linked in each card.

Many people who become seizure-free will ask whether they can stop. The decision trades a higher chance of recurrence against freedom from the medication, and it belongs to the patient once the risks are clear.

What the evidence shows

The American Academy of Neurology advisory asks clinicians to cover these points (Gloss 2021):

Sources: as cited. The online tool is the published model of Lamberink 2017.

Some people have seizures in bursts, closer together than their usual pattern. A written plan and a rescue benzodiazepine that needs no intravenous access let the family or caregivers act at once, which can help avoid progression to status epilepticus and visits to the emergency department.

No single definition

Studies have defined a seizure cluster as 3 or more seizures in 24 hours, 2 or more in 24 hours, or 2 or more in 6 hours (Jafarpour 2018). The UK National Institute for Health and Care Excellence (NICE) manages repeated or cluster seizures, typically 3 or more self-terminating seizures in 24 hours, as a medical emergency (NICE NG217). The US labels of the rescue medicines describe their target as intermittent, stereotypic episodes of frequent seizure activity, distinct from the patient's usual seizure pattern: the comparison is with the patient's own seizures.

Clusters are more likely in severe, poorly controlled epilepsy, and most studies come from tertiary epilepsy centers. They increase health care use and lower the quality of life of patients and caregivers, yet rescue medicines are underused (Jafarpour 2018).

The written plan

Rescue medicines on the US label

All four are benzodiazepines for the acute treatment of seizure clusters in people with epilepsy. They are named here by route and labeled age only; doses come from the label and the prescriber.

Each label says the medicine is not recommended for chronic, daily use as an anticonvulsant, and each carries the boxed warning on concomitant use with opioids; abuse, misuse and addiction; and dependence and withdrawal reactions. Patients and caregivers mostly preferred non-rectal routes (Jafarpour 2018). In the UK, NICE names buccal midazolam or rectal diazepam as the first-line benzodiazepine in the community for a convulsive seizure of 5 minutes or more (NICE NG217).

Needing rescue medicine often is a signal about the epilepsy itself. The labels cap use at five episodes a month, clusters are more likely when epilepsy is poorly controlled, and NICE advises seeking expert guidance when clusters recur: ask whether the epilepsy is drug resistant (see Drug-resistant epilepsy).

Sources: as cited. Labeled ages, frequency limits and warnings are from the US prescribing information on DailyMed; no doses are given.

Uncontrolled tonic–clonic seizures carry a risk of sudden death and of injury. Recordings of sudden deaths in monitoring units, and studies of the people who died, point to what lowers the risk: control of tonic–clonic seizures, and someone nearby at night.

How often, and how to tell patients

Sudden unexpected death in epilepsy (SUDEP) is the leading cause of death in chronic refractory epilepsy (Ryvlin 2013). The American Academy of Neurology and American Epilepsy Society guideline puts the risk at 0.22 per 1,000 patient-years in children and 1.2 per 1,000 in adults, and asks clinicians to tell parents that SUDEP typically affects 1 in 4,500 children with epilepsy a year, and adults that it affects 1 in 1,000 adults with epilepsy a year. Generalized tonic–clonic seizures are the major risk factor, and the risk rises with their frequency; freedom from seizures, particularly tonic–clonic seizures, is strongly associated with a lower risk (Level B; Harden 2017). The numbers were introduced in Drug-resistant epilepsy; this lesson is about the mechanism and the night.

What the recordings show

MORTEMUS, a study of cardiorespiratory arrests in epilepsy monitoring units, surveyed the units of Europe, Israel, Australia and New Zealand, 147 of 160 of which responded, and invited units elsewhere to report cases. 29 cardiorespiratory arrests were reported, including 16 SUDEP, 14 of them at night. In the 10 deaths with cardiorespiratory recordings, a consistent sequence followed a generalized tonic–clonic seizure (Ryvlin 2013):

1. Rapid breathing. After the seizure, 18 to 50 breaths a minute.
2. Breathing and heart falter. Within 3 minutes of the seizure's end, transient or terminal failure of breathing and heart function: an early postictal, centrally mediated shutdown.
3. Terminal apnea, then cardiac arrest. Where the failure was transient, it recurred: terminal apnea within 11 minutes of the end of the seizure, followed by cardiac arrest.

In adult monitoring units, SUDEP occurred at 5.1 per 1,000 patient-years, about 1.2 per 10,000 video-electroencephalography (video-EEG) monitorings, probably aggravated by suboptimal supervision and possibly by withdrawal of antiseizure medication. The authors called for better supervision in monitoring units, particularly at night (Ryvlin 2013).

Who is at risk

Two case-control studies show that the danger is a tonic–clonic seizure with no one there.

The Swedish authors estimated that 69% of these deaths in people with tonic–clonic seizures who lived alone could be prevented if they were not unattended at night or were free of tonic–clonic seizures; a previous diagnosis of substance abuse or alcohol dependence also carried excess risk (Sveinsson 2020).

The International League Against Epilepsy (ILAE) and the International Federation of Clinical Neurophysiology recommend clinically validated wearable devices for detecting generalized and focal to bilateral tonic–clonic seizures, especially in people who are unsupervised, where an alarm can bring rapid help. The recommendation is weak: detection is accurate, but it is uncertain whether the alarms improve outcomes, and the devices are not recommended for other seizure types (Beniczky 2021).

Everyday safety

Sources: as cited. SUDEP incidence and counseling follow the American Academy of Neurology and American Epilepsy Society guideline (Harden 2017).

Sex hormones and antiseizure medications act on each other in both directions, and the questions change across a lifetime: the menstrual cycle, contraception, planning and carrying a pregnancy, breastfeeding, gender-affirming hormones and menopause. Most people with epilepsy who plan a pregnancy have a healthy one. Do not stop an antiseizure medication abruptly because of pregnancy.

Most of the evidence comes from cisgender women, so this topic says “women”. The same principles apply to anyone who menstruates, can become pregnant or takes sex hormones, including transgender and nonbinary people.

Most women with epilepsy who plan a pregnancy have a healthy one. The decisive work happens before conception, in choosing the medication and recording a baseline level, and continues with levels through the nine months.

The questions continue after the birth: doses raised in pregnancy, breastfeeding and sleep, then bone health, gender-affirming hormones and menopause.

Primary guidance: American Academy of Neurology, American Epilepsy Society and Society for Maternal-Fetal Medicine practice guideline on teratogenesis, perinatal and neurodevelopmental outcomes after in-utero antiseizure medication exposure (Pack 2024); US Medical Eligibility Criteria for Contraceptive Use of the Centers for Disease Control and Prevention (CDC 2024); RCOG Green-top Guideline 68. This topic supports counseling and does not replace individualized neurology, obstetric, endocrine and maternal-fetal-medicine care.

The guidelines and consensus statements this course draws on most, and one review for the history and pharmacology of the drugs.

Background reading

For how each antiseizure medication was found and how it acts, from bromide to cenobamate, read the review by Löscher and Klein 2021. It is a history and pharmacology review, not a prescribing guide: for the choice of a first drug, follow Choosing a first antiseizure medication and the guidelines below.

Guidelines and consensus statements

• Definition of epilepsy: International League Against Epilepsy (ILAE) official report, a practical clinical definition (Fisher 2014).
• First seizure in adults: American Academy of Neurology and American Epilepsy Society evidence-based guideline on management of an unprovoked first seizure (Krumholz 2015).
• Choosing medications: National Institute for Health and Care Excellence (NICE) guideline NG217, epilepsies in children, young people and adults (NICE 2022, amended since, including in January 2025).
• Status epilepticus: definition and classification (Trinka 2015); American Epilepsy Society treatment guideline for convulsive status (Glauser 2016); the Emergency Neurological Life Support (version 6.0) status epilepticus protocol, which also gives doses.
• Critical care EEG: American Clinical Neurophysiology Society (ACNS) standardized critical care EEG terminology (Hirsch 2021) and consensus on continuous EEG in critically ill patients (Herman 2015).
• Drug-resistant epilepsy: ILAE consensus definition (Kwan 2010).
• Pregnancy: American Academy of Neurology, American Epilepsy Society and Society for Maternal-Fetal Medicine guideline on teratogenesis and outcomes after in-utero exposure (Pack 2024).