Does Therapeutic-Dose Lisdexamfetamine Lower the Seizure Threshold? · vicente.md
vicente.md/2026-001 [q-bio.NC] · 26 Jul 2026
Preprint · not peer reviewed 8 pages
Does Therapeutic-Dose Lisdexamfetamine Lower the Seizure Threshold?
Vicente González B.†
Engineer and computational neuroscience researcher · Santiago, Chile
† correspondence · hi@vicente.md · no institutional funding, no competing interests
Abstract
The proposition that stimulants lower the convulsive threshold has circulated in prescribing labels, clinical folklore, and specialist practice for decades. This appraisal asks whether it survives contact with the modern evidence base as applied to lisdexamfetamine (LDX) at therapeutic oral doses. It does not. Three large studies using orthogonal designs converge: a within-individual analysis of 801,838 patients with ADHD found odds ratios of 0.71 for seizure events during medicated months among patients both with (95% CI 0.60–0.85) and without (0.62–0.82) a prior seizure history; a matched Medicaid cohort of 18,166 stimulant users against 54,197 non-users, all children with epilepsy, found a hazard ratio of 0.95 (0.83–1.09); and a Swedish register study of 21,557 individuals with a seizure history found no increase following ADHD-medication initiation. Every adequately powered estimate lies at or below the null. None indicates harm. The warning's provenance has been independently traced to preclinical animal work and stimulant-overdose case reports rather than therapeutic-dose observation, and the U.S. Food and Drug Administration has since removed the seizure entry from the Warnings and Precautions section of the LDX label, retaining it only under Overdosage.
Two qualifications are load-bearing and are frequently dropped in secondary summaries. First, the evidence is settled at the amphetamine-class level: LDX was inside the named exposure set of the largest study but has never been estimated as its own stratum, and no molecule-specific seizure estimate exists in the literature. Second, the finding is scoped to a defined exposure envelope — oral, intact prodrug, renally adjusted dose, and epilepsy controlled at initiation. We argue that the operationally correct framing is not a dose threshold but an exposure envelope, since renal impairment, urinary alkalinization, and CYP2D6 inhibition can each produce supratherapeutic exposure at a nominally therapeutic milligram count. We conclude with an explicit map of where the negative finding holds, where it thins, and where no data exist at all — a distinction the secondary literature routinely collapses.
Keywords — `lisdexamfetamine` `seizure threshold` `ADHD pharmacotherapy` `epilepsy comorbidity` `within-individual design` `pharmacovigilance` `evidence appraisal`
Keywords:
lisdexamfetamine
adhd
seizure threshold
pharmacovigilance
Introduction
A clinical proposition can survive for decades without ever having been tested. The claim that psychostimulants lower the convulsive threshold is one of these. It appeared in North American and European package inserts for both methylphenidate and amphetamine products, was reproduced in prescribing references and specialist teaching, and functioned in practice as a soft contraindication: clinicians declined to treat ADHD in patients with epilepsy, and patients with a remote seizure history were told a stimulant was off the table. The proposition was never supported by therapeutic-dose human data. It was inferred.
The inference had two sources. Mid-century preclinical work established that amphetamine analogues can induce convulsions in rodents at doses far above any therapeutic equivalent1. Separately, poison-centre and emergency-department series established that stimulant overdose produces seizures as part of a sympathomimetic toxidrome. Regulators applied both to the therapeutic label. The International League Against Epilepsy has since stated the position plainly: the inserts warn of a threshold effect, preclinical studies and abuse-or-overdose reports are partly responsible for those warnings, and empirical data at therapeutic doses were sparse2.
That gap closed between 2017 and 2020. Three population-scale studies, using designs specifically constructed to survive confounding by indication, tested the proposition directly. This paper appraises what they found, what they did not cover, and what the residual uncertainty actually consists of — with attention to lisdexamfetamine specifically, since LDX is a dextroamphetamine prodrug whose pharmacokinetics differ from the immediate-release amphetamines on which most class inference rests.
Evidence base and study designs
Seizures are a rare outcome. This is the single fact that shaped the literature: pre-2017 studies were mostly open-label trials of 10–120 patients, chronically underpowered, and unable to distinguish a null result from a failure to detect. An intermediate generation of pharmacoepidemiologic cohorts improved on this but compared across individuals rather than within them3. The modern studies solved the power problem with administrative data, and the confounding problem with within-individual comparison.
Why within-individual designs matter here
Patients with ADHD carry an elevated baseline seizure risk independent of treatment. Wiggs et al. quantified this: relative to non-ADHD controls, odds of any seizure were 2.33 (2.24–2.42) in males and 2.31 (2.22–2.42) in females4. Any between-person comparison of medicated against unmedicated patients is therefore contaminated, because the decision to medicate correlates with severity, with comorbidity, and with healthcare contact intensity. The within-individual design compares the same person during medicated months against their own unmedicated months, holding constant every time-invariant confounder — genotype, epilepsy syndrome, socioeconomic position, baseline severity. What remains is the drug effect plus time-varying confounding, a far smaller residual.
Table 1. Inventory of adequately powered studies bearing on therapeutic-dose stimulant exposure and seizure risk. Design heterogeneity is the strength of this set: claims-based within-individual analysis, propensity-matched cohort, and national register self-comparison fail in different directions, so their agreement is more informative than three replications of one method.
§ First 30 days only (95% CI 2.09–7.68); no elevation beyond three months. Absolute incidence of new-onset seizure in that cohort was 0.2%, or 4.4 per 10,000 patient-years.
Results
Convergence across designs
The three estimates in Figure 1 agree in sign and approximately in magnitude. Liu et al. additionally report the underlying incidences, which are worth stating in absolute terms because ratio measures obscure scale: seizure-related hospitalization occurred at 3.6, 3.5, and 4.3 per 100 patient-years in current users, former users, and non-users respectively5. The direction is unfavourable to the proconvulsant hypothesis and the magnitudes are small.
We caution explicitly against the inverse error. Several secondary accounts read the sub-unity estimates as evidence that stimulants are anticonvulsant, sometimes proposing a mechanism in which treating ADHD improves antiseizure-medication adherence and sleep regularity. That mechanism is plausible and untested. Residual confounding by time-varying indication, healthy-adherer effects, and differential ascertainment during treated periods are at least as parsimonious. The defensible reading of Figure 1 is absence of harm, not presence of benefit.
The regulatory record
Label archaeology provides an independent line of evidence, because it records a regulator's own revision of a prior position. The 2012 U.S. lisdexamfetamine label carried, under Warnings and Precautions §5.3, the statement that the drug may lower the convulsive threshold and should be discontinued in the presence of seizures8. The current label, revised April 2026, contains no seizure entry in §5 at all9. Seizures now appear in exactly three places: as a manifestation of serotonin syndrome (§5.7), among voluntary postmarketing reports where frequency and causality cannot be established (§6.2), and under Overdosage (§10), where they are listed alongside psychomotor agitation, hallucinations, cerebrovascular accident, and coma. Epilepsy is not a listed contraindication.
Boundaries of the finding
A negative safety finding is only as good as its scope statement. Three distinct epistemic categories are routinely merged in secondary summaries, and separating them is the main practical contribution of this section: evidence of no effect, evidence of effect at the boundary, and no evidence in either direction.
Where the negative finding holds
Oral administration of the intact prodrug, at or below 70 mg daily, with the label's renal reductions applied (50 mg for GFR 15 to <30 mL/min/1.73m²; 30 mg for end-stage renal disease), in ADHD patients aged six and over, with or without a prior seizure history. The prior-seizure stratum was explicitly estimated rather than assumed, which is what elevates this from extrapolation to evidence. Where epilepsy is present, the finding holds when seizures are controlled at initiation.
Where the finding thins
Two soft spots are real and are usually omitted. The first is active, uncontrolled epilepsy at initiation. In the tabulated literature10, the only two studies recording increased risk both involved actively seizing patients: a double-blind controlled trial of 33, and an open-label trial in which the increase was confined to the actively seizing subgroup1112. Both used methylphenidate; both are small and old, and a later open-label series in difficult-to-treat epilepsy found reduced seizure severity with increased frequency in only four of twenty-two patients13. The pattern is nonetheless consistent enough that establishing seizure control before adding a stimulant remains the standard sequence.
The second is the initiation window. Man et al. found an incidence rate ratio of 4.01 (2.09–7.68) during the first 30 days of methylphenidate treatment, with no elevation beyond three months7. Whether this reflects pharmacology or ascertainment — newly treated patients have more clinical contact and more opportunities for an event to be recorded — is unresolved. It warrants a monitoring window, not a contraindication.
Where no evidence exists
Children under six, in whom steady-state dextroamphetamine exposure runs approximately 44% higher than in ages 6–11 at identical dosing9; adults over 65, for whom trials did not enrol sufficient numbers; pregnancy and lactation; adults with epilepsy, since the epilepsy-specific register work was conducted in individuals under 19; and the binge-eating-disorder population as a separately analysed seizure cohort. The correct statement for each is not studied, which is an argument for clinical caution and not for assumed danger.
Dose is not exposure
The framing that dominates both the label and the secondary literature is a dose threshold: below 70 mg the finding holds, above it one is in overdose territory. This framing is operationally misleading. Lisdexamfetamine is an inactive prodrug hydrolysed by red blood cells to dextroamphetamine, with roughly 96% of an administered dose recovered in urine9. The quantity that governs pharmacodynamic effect is systemic dextroamphetamine exposure, and several routine clinical circumstances decouple that quantity from the milligram count on the prescription.
Key insight. The therapeutic-dose seizure evidence is best read as defining an exposure envelope rather than a dose ceiling. Three routine circumstances — unadjusted dosing in renal impairment, co-administration of urinary alkalinizers, and CYP2D6 inhibition — can place a patient outside that envelope while the prescription remains nominally therapeutic. Conversely, a patient at 20 mg is well inside it, notwithstanding secondary accounts that erroneously define the therapeutic range as beginning at 30 mg; 20 mg is a marketed strength and a legitimate individualized dose. The clinically actionable question is never “how many milligrams” in isolation.
Table 2. Conditions that place a patient outside the exposure envelope to which the negative finding applies. Column three distinguishes mechanisms that raise systemic exposure from those that act on seizure threshold through an independent pathway — a distinction that matters because the first class is dose-correctable and the second is not.
The fourth and fifth rows deserve a note. Carbonic anhydrase–inhibiting antiseizure medications, notably topiramate and zonisamide, alkalinize urine. The label documents that urinary alkalinizing agents increase amphetamine blood levels and potentiate its action9, and the composition of the interaction is therefore predictable: a patient with epilepsy on either agent may sit at higher amphetamine exposure than the prescribed dose implies. We are not aware of a study quantifying this specific combination, and flag it as an inference from the interaction table rather than an established finding.
Working in the opposite direction, LDX has two properties that are favourable relative to immediate-release amphetamines. Hydrolysis by red blood cells rather than absorption of an active moiety flattens the concentration–time profile; if any proconvulsant effect is Cmax-driven — which the overdose toxicology implies — the prodrug architecture is protective. Second, lisdexamfetamine is not metabolized by cytochrome P450 enzymes9, which eliminates most of the enzyme-induction interactions that complicate stimulant co-prescription with older antiseizure medications. Neither observation is evidence. Both are reasons to expect that a class-level null generalizes to this molecule rather than failing at it.
The molecule-level gap
Lisdexamfetamine was inside the exposure definition of the largest study. Reviews tabulating Wiggs et al. record the exposure set as amphetamine salt combination, dexmethylphenidate, dextroamphetamine sulfate, lisdexamfetamine dimesylate, methamphetamine hydrochloride, methylphenidate, and methylphenidate hydrochloride10. LDX contributed person-time to the pooled odds ratio of 0.71. It was never estimated separately, and the published subgroup analyses stratify by seizure history, sex, and antiseizure co-medication rather than by molecule.
Three structural reasons explain the absence, and they compound. Seizures are the rare outcome, so slicing by molecule collapses statistical power precisely where it is needed; a molecule-specific null would carry intervals wide enough to be uninformative. LDX launched in 2007, after much of the observation window in the U.S. and Swedish data. And the question was framed at class level because the proposition under test was a class warning. A dedicated LDX monotherapy cohort of approximately 430,000 patients does exist across four U.S. databases14, but its prespecified outcomes are suicidality, psychosis, and substance abuse. The infrastructure to answer the molecule-level question has been built and has not been pointed at this endpoint.
The International League Against Epilepsy grades the resulting state of knowledge conservatively: methylphenidate carries level B evidence for safety and tolerability in children with epilepsy, while amphetamines are graded as having insufficient data to support use in that population2. This is a statement about evidential coverage, not about observed harm, and the two are frequently conflated. A recent Cochrane review reaches the same conclusion from the other direction, identifying only two eligible trials and calling for studies including adults and a wider range of stimulant agents15.
Discussion
What does it mean to call a null safety finding “settled”? We propose three conditions, all of which are met here. The proposition must have been tested by designs adequate to detect the claimed effect — satisfied, with three orthogonal designs and combined sample sizes approaching one million. The origin of the contrary belief must have been identified and explained, rather than merely outvoted — satisfied, since the preclinical and overdose provenance is documented and the inferential leap is visible. And an independent adjudicator with an institutional bias toward caution must have revised its position — satisfied, in that a regulator moved the statement from Warnings to Overdosage.
Notably, none of these three conditions requires a positive demonstration of safety, which is unobtainable for rare outcomes. They require that the specific claim under examination be shown to lack support and that the reason for its former support be understood. This is the appropriate epistemic bar for deprecating a warning, and it is a lower bar than the one implicitly demanded by clinicians who continue to withhold stimulants from patients with remote seizure histories.
The cost of the residual caution is asymmetric and largely invisible. Untreated ADHD in the epilepsy population is associated with reduced antiseizure-medication adherence and worse psychosocial outcomes10, and the patients most likely to have comorbid ADHD are those with the most complex epilepsy. Withholding treatment on the basis of a deprecated warning produces harm that is never attributed to the decision that produced it, whereas a seizure occurring during stimulant treatment is attributed immediately and often incorrectly. This asymmetry in attribution, rather than any asymmetry in the underlying evidence, is the most likely reason the proposition has outlived its refutation.
A final methodological note concerns the propagation of this literature through secondary summaries. In preparing this appraisal we compared several independent syntheses of the same source material and found a recurring failure mode: correct headline conclusions accompanied by fabricated supporting specifics — precise seizure-inducing dose thresholds that appear in no source, animal-study routes and dosages that do not match the cited work, and mechanistic explanations for the sub-unity estimates presented as established findings rather than hypotheses. In each case the fabricated detail made the correct conclusion sound better supported than it is. Readers of any synthesis on this topic, including this one, should verify the numbered claims against the primary sources rather than the reverse.
Conclusions
At therapeutic oral doses, lisdexamfetamine has not been shown to lower the seizure threshold or increase seizure risk. The correct formulation is absence of evidence of increased risk, not biological impossibility; rare individual events occur with almost any CNS-active agent.
The evidence is settled at class level and unreplicated at molecule level. LDX contributed to the pooled estimates but has no stratum of its own. This is a gap in granularity rather than a live controversy, and two pharmacokinetic properties of the prodrug — flattened Cmax and absence of CYP metabolism — argue against it failing where the class succeeded.
The operative variable is exposure, not dose. Unadjusted renal dosing, urinary alkalinization, and CYP2D6 inhibition can place a patient outside the studied envelope at a nominally therapeutic milligram count. Conversely, doses below 30 mg are inside the envelope, not below a therapeutic floor.
Two boundaries deserve explicit clinical handling: establish seizure control before initiating in a patient with active epilepsy, and monitor the first thirty days, during which the only positive signal in the modern literature was observed.
For patients whose epilepsy is surgically resolved and who are off antiseizure medication, the studied populations are a conservative reference. Those cohorts consisted of patients with active diagnoses on active treatment; a post-surgical, medication-free, aura-free patient sits below that risk floor rather than inside it, though no dataset names this stratum. Being off antiseizure medication additionally removes nearly the entire interaction surface described in Section 5.
The dominant modifiable risk in such a patient is sleep, not threshold pharmacology. Insomnia was reported in 27% of adults in the LDX registration trials9, and sleep deprivation is among the most reliable seizure triggers known. Morning-only dosing, treating degraded sleep as a dose-adjustment signal rather than a tolerable side effect, and attention to appetite suppression are the interventions with the largest expected effect.
Reproducibility — derived quantities
The exposure multipliers plotted in Figure 4 are derived from the label's renal dose caps under an exposure-equivalence assumption. The forest-plot log-axis positions and the inverted renal multipliers are reproduced below.
# Exposure multiplier implied by label dose caps.# Assumption: the capped dose in impairment is intended to yield# the same systemic exposure as 70 mg in normal renal function.
NOMINAL = 70.0
CAPS = {'normal': 70.0, 'severe_ckd': 50.0, 'esrd': 30.0}
implied = {k: NOMINAL / v for k, v in CAPS.items()}
# → {'normal': 1.00, 'severe_ckd': 1.40, 'esrd': 2.33}# Forest-plot estimates (log10 ratio scale).import math
EST = {'Wiggs 2018': (0.71, 0.60, 0.85),
'Brikell 2019': (0.73, None, None),
'Liu 2018': (0.95, 0.83, 1.09)}
for name, (pe, lo, hi) in EST.items():
crosses = (lo isNone) or (lo < 1.0 < hi)
harm = (lo isnotNone) and (lo > 1.0)
print(name, round(math.log10(pe), 3), 'harmful:', harm)
# → no estimate satisfies harm == True
References
Footnotes
Zagnoni PG, Albano C. Psychostimulants and epilepsy. Epilepsia. 2002;43(Suppl 2):28–31. ↩
Auvin S, Wirrell E, Donald KA, Berl M, Hartmann H, Valente KD, Van Bogaert P, Cross JH, et al. Systematic review of the screening, diagnosis, and management of ADHD in children with epilepsy. Consensus paper of the Task Force on Comorbidities of the ILAE Pediatric Commission. Epilepsia. 2018;59(10):1867–1880. PMID 30178479. free full text. ↩↩2
McAfee AT, Landon J, Jones M, et al. A cohort study of the risk of seizures in a pediatric population treated with atomoxetine or stimulant medications. Pharmacoepidemiology and Drug Safety. 2013;22(4):386–393. ↩
Wiggs KK, Chang Z, Quinn PD, Hur K, Gibbons R, Dunn D, Brikell I, Larsson H, D'Onofrio BM. Attention-deficit/hyperactivity disorder medication and seizures. Neurology. 2018;90(13):e1104–e1110. .
Figure 1. Effect estimates for stimulant exposure on seizure outcomes, plotted on a logarithmic ratio axis against the null value of 1.0. Markers are point estimates; horizontal bars are 95% confidence intervals where published. Estimates to the left of the dashed null line indicate fewer seizure events during medicated periods. The dominant signal is that no interval excludes the null on the harmful side, and two of the three lie entirely below it — the pattern one expects from a null-or-protective exposure, not from a proconvulsant. The Brikell interval is shown open because the published analysis reports a within-individual rate ratio across pre- and post-initiation windows rather than a single symmetric interval.
Figure 2. Divergence of the evidence stream from the regulatory stream, 1990–2026. The upper track marks the accumulation of human therapeutic-dose evidence; the lower track marks the status of the seizure-threshold statement in the U.S. lisdexamfetamine label. The warning persisted for roughly two decades on inferential grounds alone, was contradicted within three years of the first adequately powered studies, and was removed from the Warnings section thereafter. The lag between the leftmost evidence marker and the label revision is the quantity of interest: it is the cost, in patient-years of withheld treatment, of a proposition that was never tested at the dose it governed.
Figure 3. Evidence coverage by population stratum and evidence tier. Filled markers denote direct estimation within that stratum; half-filled denote inclusion in a pooled analysis without separate estimation; open denote absence of data. The diagonal structure is the finding: coverage is dense at class level for the core ADHD populations, sparse for epilepsy-specific and adult-epilepsy strata, and empty across the entire molecule-specific column. No stratum has an LDX-specific seizure estimate.
Figure 4. Relative systemic exposure at a nominal 70 mg daily dose under four modifying conditions. The renal bars are derived by inverting the label's own dose caps: if the manufacturer judges 50 mg in severe impairment and 30 mg in ESRD to be exposure-equivalent to 70 mg in normal function, then administering the unreduced 70 mg implies approximately 1.4× and 2.3× the intended exposure respectively. The two hatched bars denote interactions whose direction is documented in the label but whose magnitude is not quantified for LDX; their heights are illustrative, not estimated. The practical implication is that a prescription reading “70 mg” can correspond to markedly different internal exposures, and the seizure evidence applies to the exposure, not to the label.
Liu X, Carney PR, Bussing R, Segal R, Cottler LB, Winterstein AG. Stimulants do not increase the risk of seizure-related hospitalizations in children with epilepsy. Journal of Child and Adolescent Psychopharmacology. 2018;28(2):111–116. PMID 29028437. ↩↩2
Brikell I, Chen Q, Kuja-Halkola R, D'Onofrio BM, Wiggs KK, Lichtenstein P, Almqvist C, Quinn PD, Chang Z, Larsson H. Medication treatment for attention-deficit/hyperactivity disorder and the risk of acute seizures in individuals with epilepsy. Epilepsia. 2019;60(2):284–293. PMID 30682219. ↩
Man KKC, Lau WCY, Coghill D, et al. Association between methylphenidate treatment and risk of seizure: a population-based, self-controlled case-series study. Lancet Child & Adolescent Health. 2020;4(6):435–443. abstract. ↩↩2
U.S. Food and Drug Administration. Vyvanse (lisdexamfetamine dimesylate) prescribing information, application 021977s022. 2012. §5.3 Seizures. FDA PDF. ↩
Takeda Pharmaceuticals America. Vyvanse (lisdexamfetamine dimesylate) capsules and chewable tablets, full prescribing information. Revised April 2026. current PI. ↩↩2↩3↩4↩5↩6
Fan HC, Chiang KL, Chang KH, Chen CM, Tsai JD. Epilepsy and attention deficit hyperactivity disorder: connection, chance, and challenges. International Journal of Molecular Sciences. 2023;24(6):5270. open access. ↩↩2↩3
Gonzalez-Heydrich J, Whitney J, Waber D, et al. Adaptive phase I study of OROS methylphenidate treatment of attention deficit hyperactivity disorder with epilepsy. Epilepsy & Behavior. 2010;18(3):229–237. ↩
Gross-Tsur V, Manor O, van der Meere J, Joseph A, Shalev RS. Epilepsy and attention deficit hyperactivity disorder: is methylphenidate safe and effective? Journal of Pediatrics. 1997;130(4):670–674. ↩
Santos K, Palmini A, Radziuk AL, et al. The impact of methylphenidate on seizure frequency and severity in children with attention-deficit–hyperactivity disorder and difficult-to-treat epilepsies. Developmental Medicine & Child Neurology. 2013;55(7):654–660. ↩
ClinicalTrials.gov identifier NCT04132557. A study on suicidality, psychosis or substance abuse with methylphenidate, atomoxetine, amphetamine/dextroamphetamine or lisdexamfetamine. Retrospective cohort, n ≈ 430,000. ClinicalTrials.gov. ↩
Eaton C, Yong K, Walter V, et al. Stimulant and non-stimulant drug therapy for people with attention deficit hyperactivity disorder and epilepsy. Cochrane Database of Systematic Reviews. 2022;CD013136. full text. ↩