Epilepsy is a tendency to recurrent unprovoked seizures — brief episodes of abnormally synchronised electrical activity in the brain. It is one of the most common serious neurological conditions, and its global burden has been quantified across 195 countries [1] [2], with a disproportionate share falling on low-income settings where treatable causes go untreated [3].
It is also one of the most misunderstood. Epilepsy is not one disease. It is a symptom of many — genetic channelopathies, structural damage from stroke or trauma or tumour, infections, malformations of cortical development, autoimmune encephalitis, and in a large fraction of cases nothing identifiable at all. And a seizure is not automatically epilepsy: a seizure provoked by an acute insult — low blood sugar, alcohol withdrawal, acute stroke — is a different thing from an enduring predisposition to seizures, and treating it as epilepsy is a common error.
The hopeful fact, which deserves stating early because the stigma is largely built on its absence, is that most people with epilepsy become seizure-free on medication. In two large cohorts nearly two-thirds did [4] [5]. The important structural fact, which the centerpiece is about, is when that happens: almost all of it on the first drug or the second, and almost none afterwards.
Start here: seizures, and what generates them
A seizure arises when the normal balance between excitation and inhibition in a population of neurons fails and they fire in pathological synchrony. Both sides of that balance are implicated: GABAergic inhibitory mechanisms [6] and glutamatergic excitatory ones [7], and most antiseizure medications act on one or the other, or on the ion channels that set neuronal excitability [8].
Where the discharge starts and how far it spreads determines what the seizure looks like. The ILAE's operational classification divides seizures first by onset — focal, generalised, or unknown — then by whether awareness is retained, then by motor or non-motor features [9]. This replaced older terminology whose vocabulary ("grand mal", "petit mal", "partial") persists in public usage and obscures more than it conveys [10] [11].
A focal seizure may be a rising epigastric sensation, a smell, a sense of familiarity, a hand movement, a period of unresponsiveness with automatisms — none of which looks like the convulsion most people picture. A generalised tonic-clonic seizure does look like it. Both are epilepsy; only one is recognisable to a bystander, which is a large part of why diagnosis is delayed.
Above the level of individual seizures sit epilepsy syndromes — recognisable constellations of seizure type, age at onset, EEG pattern and prognosis, which carry far more information than any single seizure description [12] [13].
Pillar 1: measurement and diagnosis
The diagnosis is clinical, and the definition is explicit
The ILAE's practical definition states that epilepsy is a disease of the brain defined by any of: at least two unprovoked seizures more than 24 hours apart; one unprovoked seizure with a probability of further seizures of at least 60 percent over the next ten years — the same as the general recurrence risk after two; or diagnosis of an epilepsy syndrome [14].
That second clause matters. It means a single seizure in someone with a structural lesion and an abnormal EEG can be epilepsy, and can be treated as such, without waiting for a second event. The same document defines epilepsy as resolved for someone past the age of an age-dependent syndrome, or seizure-free for ten years and off medication for five — carefully distinguished from "cured" [14].
The diagnosis rests overwhelmingly on description: what the person did, in what order, whether they responded, what happened afterwards. A witness account is often more valuable than any test.
EEG, and its limits
The electroencephalogram detects interictal epileptiform discharges between seizures, supports classification, and identifies syndromes. Its two limitations must be stated together, because misunderstanding either causes harm.
A normal EEG does not exclude epilepsy. Discharges are intermittent, and a routine recording samples twenty minutes of a life.
An abnormal EEG does not establish it. Non-specific abnormalities occur in people who will never have a seizure, and treating an EEG rather than a patient is a recognised route to misdiagnosis.
The most common differential is psychogenic non-epileptic seizures, which resemble epileptic seizures, do not arise from epileptiform discharge, and require an entirely different treatment — a distinction that video-EEG usually settles and that has its own trial literature [15].
MRI, and looking for a cause
MRI seeks the structural causes: hippocampal sclerosis, cortical malformations, tumours, vascular lesions, old infarcts and trauma. Finding one changes the recurrence risk, the syndrome classification, and — critically — whether surgery might be curative.
Stroke is a leading cause of adult-onset epilepsy, and the risk is now predictable: the SeLECT score estimates the probability of late seizures after ischaemic stroke from routinely available clinical variables [16]. Whether epilepsy can be prevented after such an insult is a distinct and unsolved question [17].
Centerpiece: a simple simulatable model of what medication can do
The most useful number in epilepsy care is not how many people respond to treatment but how the response is distributed across successive drugs, because that shape dictates when to stop trying drugs.
A prospective cohort of 525 newly diagnosed patients followed at a single Glasgow centre found that 333 (63 percent) became seizure-free. Among the 470 previously untreated, 222 (47 percent) became seizure-free on their first antiepileptic drug, and 67 (14 percent) on a second or third [4].
Seventeen years later the same unit reported a cohort more than three times larger — 1,795 patients treated between 1982 and 2012, spanning the introduction of more than a dozen new drugs. 1,144 (63.7 percent) were seizure-free. Of the total pool, 50.5 percent achieved it on the initial drug; the second and third regimens added 11.6 and 4.4 percent; everything after that added 2.12 percent [5].
The model is checked twice against numbers it was not given. The 2017 cohort separately reports that 89.9 percent of those achieving control did so with the first or second regimen; recomputing that share from its own stepwise percentages gives 90.5 percent. And the two cohorts' overall seizure-freedom rates — 63 percent and 63.7 percent — agree to within a percentage point, as do their first-drug rates (47 and 50.5 percent), despite seventeen years and a dozen new medications between them. That agreement is the later paper's own headline: "Despite the availability of many new AEDs with differing mechanisms of action, overall outcomes in newly diagnosed epilepsy have not improved" [5].
This is the teaching point, and it has a direct operational consequence. After two failed drugs the remaining yield from further drug trials is about 6 percent in total. That is precisely why the ILAE defines drug-resistant epilepsy as failure of adequate trials of two tolerated, appropriately chosen and used antiseizure medication schedules [18] — and why reaching that point should trigger referral for surgical evaluation rather than a third, fourth and fifth prescription. In practice it does not: patients are typically referred for surgery after about 20 years of seizures, often too late to prevent disability [19].
The right-hand panel adds the refinement that keeps this from being applied crudely. Among patients whose first drug failed, later seizure freedom depended sharply on why it failed: 11 percent when the drug lacked efficacy, but 41 percent when it was stopped for intolerable side effects and 55 percent after an idiosyncratic reaction [4]. A drug abandoned at a sub-therapeutic dose because of a rash has not really been tried — which is exactly why the ILAE definition specifies tolerated and adequately used trials [18].
Two honest limits. Both cohorts come from a single specialist centre, so the case mix is not a general population, and the stepwise percentages count ever achieving a year of freedom on a given regimen while the overall rate counts being seizure-free in the final year — different quantities, shown separately here and deliberately not summed to reconcile. And "seizure freedom" is not the only outcome that matters; the drugs have cognitive, teratogenic and psychiatric costs that this curve does not show.
Pillar 2: treatment
Antiseizure medications
The mainstay, and there are now more than twenty, acting through a handful of mechanisms: sodium-channel blockade (carbamazepine, lamotrigine, lacosamide), GABA enhancement (benzodiazepines, phenobarbital), calcium-channel effects (ethosuximide in absence seizures), broad or multiple mechanisms (valproate, topiramate), and synaptic vesicle protein 2A binding — the target identified for levetiracetam, an unusual case where a drug found by screening had its binding site worked out afterwards [20] [21].
Choice is driven by seizure and syndrome type, not by potency. Drugs effective in focal epilepsy can worsen generalised epilepsies. The ILAE has assembled the evidence for initial monotherapy [22], and lamotrigine and others have monotherapy trial data [23]; the newer agents brought tolerability and interaction advantages rather than higher efficacy [24], which the cohort comparison above demonstrates unusually starkly.
Valproate deserves separate mention because it is both among the most effective drugs, particularly in generalised epilepsy, and a major teratogen: prenatal exposure is associated with substantially increased risk of autism spectrum disorder and childhood autism [25]. Its use in people who could become pregnant is now tightly restricted, and this is one of the clearest risk-benefit trade-offs in neurology.
Surgery, which can be curative
For drug-resistant focal epilepsy with an identifiable, resectable focus, surgery is not a last resort — it is the treatment with the best evidence.
A randomised controlled trial assigned 80 patients with temporal-lobe epilepsy to surgery or a further year of medical therapy: 58 percent of the surgical group were free of awareness-impairing seizures at one year against 8 percent of the medical group, with better quality of life and adverse effects of surgery in 10 percent [26]. ERSET tested the same question early — within two years of failing two drugs — and found 0 of 23 medical patients seizure-free in year two against 11 of 15 surgical patients, though it was halted prematurely for slow accrual and memory decline occurred in 36 percent after surgery [19]. Long-term series describe the durability of remission and its patterns [27] [28].
Neurostimulation, when surgery is not possible
Vagus nerve stimulation delivers intermittent stimulation via an implanted device, with efficacy in partial-onset seizures established in randomised trials and durability in long-term follow-up [29] [30]. Deep brain stimulation of the anterior nucleus of the thalamus was tested in the SANTE trial [31] [32]. Responsive neurostimulation detects electrographic seizure onset and delivers stimulation in a closed loop [33]. None is typically curative; all reduce seizure frequency in people with no surgical option.
Diet, and everything else
The ketogenic diet and its variants have a real evidence base, particularly in childhood epilepsies, and the mechanisms are under active investigation [34].
Counselling is treatment. Driving restrictions, safety around water and heights, sleep deprivation and alcohol as triggers, medication adherence, and pregnancy planning are all part of care, as are the psychiatric and cognitive comorbidities that accompany epilepsy far more often than chance [35] [36], and the specific risks around delivery [37].
SUDEP, which should be discussed
Sudden unexpected death in epilepsy is the leading cause of epilepsy-related death in people with chronic uncontrolled seizures. Its definitions have been unified [38], its incidence and risk factors characterised [39] [40] [41] [42], and the dominant modifiable risk factor identified: frequency of generalised tonic-clonic seizures. Seizure control and mortality are linked [43], adjunctive medication that reduces seizures reduces SUDEP risk [44], and the mechanism appears to involve peri-ictal respiratory compromise — ictal hypoxaemia is common and severe in focal epilepsy [45].
That chain — more seizures, more risk; better control, less risk — is the strongest argument for not settling for partial control, and it is the reason the two-drug rule matters clinically rather than merely administratively.
Pillar 3: what is unresolved
Why a third of patients are drug-resistant. Proposed mechanisms include drug transporters, target alterations and network reorganisation, and none satisfactorily explains the phenomenon [46] [47]. That the failure rate has not moved despite a dozen new drugs with different mechanisms [5] suggests the obstacle is not the absence of the right molecule.
Precision therapy for genetic epilepsies. Genome-wide analysis has identified loci and implicated diverse mechanisms [48], and for a small number of specific channelopathies the genetic diagnosis now selects the drug — the clearest existing example of precision medicine in epilepsy, and still a small fraction of cases.
Preventing epilepsy after brain injury. Post-stroke and post-traumatic epilepsy have a latent period between the insult and the first seizure — an obvious window for prevention, and no intervention has been shown to use it [17] [16].
Seizure prediction and closed-loop treatment. Long-term intracranial recording has made forecasting seizures a tractable problem [49], and responsive stimulation is the first therapy to act on that signal [33].
Getting people to surgery. The evidence that surgery works in appropriately selected drug-resistant focal epilepsy is stronger than for most neurological interventions [26] [19]. A twenty-year median delay to referral is not a scientific problem.
Dig deeper in lmmol
Stroke is a leading cause of adult-onset epilepsy, and the risk of late seizures after ischaemic stroke can now be estimated from routine clinical variables [16] — the two diseases share both a causal link and the unsolved question of whether the latent period between them can be used [17]. Among the other neurological reviews, Alzheimer's disease and Parkinson's disease offer the contrast of progressive degeneration against a disorder that is episodic and frequently static, while multiple sclerosis shares epilepsy's dependence on MRI for both diagnosis and cause-finding. Depression and anxiety are markedly more common in epilepsy than chance would predict and materially affect quality of life [35] — see depression. Obstructive sleep apnea is common in medically refractory epilepsy and is a treatable contributor to poor seizure control [36] — see obstructive sleep apnea. The full collection is at health.