Alpha-GPC and Stroke: What the Korean Cohorts Actually Show

You have been taking it for a few months. Then a headline, or a comment under a video, or a line in a forum thread: alpha-GPC raises stroke risk 43%. Sometimes 46%. The number is real and it comes from a real study of twelve million people. Almost nothing said around it is precise, including — and this is the part worth knowing — the version on the page you would check first.

There are two Korean cohorts, not one. They point in opposite directions. Neither is about you, and the reasons they are not about you are more interesting than the headline.

A word on why a publication for people who work with their heads is writing about stroke registries. Alpha-GPC is sold for the afternoon, and the case that it does anything for a healthy person’s afternoon is thin — that is a separate article. This one exists because a compound sold on a cognitive promise has a cardiovascular literature attached to it, and the two never travel together. This piece is about the safety evidence only. What is actually inside a capsule compared with what is inside an egg is a separate question with a clean arithmetical answer, and it lives in its own article: a standard capsule delivers less choline than a single egg.

The cohort of twelve million, in full

A retrospective cohort built on Korean national insurance claims followed 12,008,977 people aged 50 and over from 2009 to 2018, of whom 108,877 had been prescribed alpha-GPC. After 1:1 matching on age, sex, household income and Charlson comorbidity index, the published abstract reports an adjusted hazard ratio for total stroke of 1.43 (95% CI 1.41–1.46); 1.34 (1.31–1.37) ischaemic, 1.37 (1.29–1.46) haemorrhagic. Longer prescription duration went with higher risk. Note that this is duration, not dose — the analysis compares how long people were on the drug, not how much they took. For ischaemic stroke, against under two months of use, more than twelve months gave 1.37 (1.28–1.46), p < .001 for trend.

Two things about that number before anything else. Both figures you will see quoted are the paper’s: 1.46 (1.43–1.48) in the whole cohort before matching, which the summary box renders as 46% higher, and 1.43 (1.41–1.46) after matching on every covariate. Choosing between them loses the interesting part: matching on age, sex, income and comorbidity barely moved the estimate. Be careful what that shows. It rules out those four variables as the explanation. It says nothing about the one that matters here — why the prescription was written — which is not in the matching set and cannot be, because claims data does not record it. An estimate insensitive to matching on a set that excludes the confounder is not an estimate cleared of that confounder. The study was funded by Korea’s National Health Insurance Service and its National Research Foundation with no conflicts declared: not industry work, in either direction.

Now the parts usually missing. The duration gradient is not clean throughout. For haemorrhagic stroke the three bands run 1.23 (1.06–1.44) at two to six months, 1.07 (0.89–1.29) at six to twelve, and 1.34 (1.11–1.61) beyond twelve — the middle band’s interval crosses 1, so the gradient is not monotonic. Be careful what that is and is not, because this is where cohort readings usually go wrong. It is not a finding that six to twelve months of use carries no risk: the intervals overlap heavily, the band is one slice of a larger analysis, and without the absolute event rates — which we could not retrieve — there is no way to judge how much it could have detected. The paper reports p = .002 for the haemorrhagic trend, and one band’s interval crossing 1 does not overturn a trend statistic. What it does is make the tidy version of this study, the one where risk climbs steadily with time on the drug, untrue as stated. We could not open the full text — the journal and PubMed Central both returned bot-challenge pages — so these figures come from the published abstract and an institutional record, and we make no claim about anything else the paper contains. Absolute event rates were not in what we retrieved, and that matters: a hazard ratio of 1.43 tells you nothing about how many strokes it represents.

Most important: this is a claims database of people aged 50 and over, prescribed alpha-GPC as a drug, where the usual regimen is 1,200 mg a day — the convention in the prescribing literature; the doses actually dispensed were not in what we could retrieve. Confounding by indication is the obvious threat — alpha-GPC is prescribed to people with cognitive complaints, who plausibly carry vascular risk that age, sex, income and a comorbidity index do not capture. The design pushes back on that harder than most write-ups admit: from 13.5 million people, the authors excluded anyone with a prior stroke, a transient ischaemic attack, a history of antidementia drugs or an earlier alpha-GPC prescription, along with those who had died, before arriving at their 12 million. It does not close the question, because the reason for the prescription is still unobserved. And nobody has studied whether any of it applies to a healthy 30-year-old taking 300 mg from a bottle. If you see “alpha-GPC raises stroke risk 43%” without that sentence attached, someone has taken a correct number and moved it to the wrong population. Further down the chain it degrades into something that no longer exists. A comparison page published by a pouch company in February 2026, which we read in full on 30 August 2026, tells readers that a 2021 study in JAMA Internal Medicine by Sun et al. found alpha-GPC associated with increased TMAO. Every identifying element is wrong, and one search establishes that: the authors are Lee and colleagues, the journal is JAMA Network Open, and the outcome measured was stroke incidence, not TMAO. The alpha-GPC-and-TMAO work does exist — it is Wang and colleagues, in a different journal, in mice. Two real papers have been folded into one citation that names neither correctly.

The sentence sits directly above a section headed “Why Nectr Uses Cognizin (Citicoline),” so a citation nobody could look up is doing work against a competing ingredient on a page that sells the other one. We are not calling that deception: a page assembled carelessly and never checked explains it just as well, and more of the internet works that way than the other. And this is one page from one seller — we point at it rather than at the category because we opened it and can show you the sentence.

A second study using the same national database points the other way, and it deserves the same prominence. Among 508,107 Koreans newly diagnosed with mild cognitive impairment between 2013 and 2016, those prescribed alpha-GPC converted to Alzheimer’s disease dementia less often than non-users (HR 0.899, 0.882–0.918) and to vascular dementia less often still (0.832, 0.801–0.865). Stroke risk did not rise in this population at all, and fell among the patients whose impairment did not progress. Funding and conflicts for this study we could not retrieve, and we read it at abstract level.

The temptation is to call this a contradiction and stop. The reason it isn’t one matters more than either result. The two studies are asking different questions about different people. The first excluded anyone with a stroke in their history, a transient ischaemic attack, or any prior antidementia prescription — it is asking what happens when alpha-GPC appears in the records of a broadly healthy older population. The second is made entirely of patients with mild cognitive impairment, which is to say entirely of the people alpha-GPC is actually prescribed for. Confounding by indication is the central threat to the first design and largely absent from the second, because in the second everyone shares the indication.

So the honest reading is this: the association appears in a population where the reason for the prescription is unobserved, and disappears in a population where everyone shares that reason. That is precisely the pattern you would expect if the signal were confounding — and also a pattern consistent with a real effect that a sicker population masks. Nobody has separated those, and neither study can.

The reference page most people would check has one of these two cohorts and not the other. Examine’s alpha-GPC page reports the stroke cohort with its numbers — twelve million participants, 108,877 users, elevated stroke risk after twelve months of use — alongside the mouse study. It goes further than that: it tells readers the causal link between TMAO and cardiovascular disease is not especially strong, that healthy people may have little to worry about, and that someone at high cardiovascular risk might reasonably weigh the added risk. That is risk-stratified advice rather than a headline, and it is more than almost any commercial page in this category manages.

What it does not have is the second cohort. We read the page in full on 30 August 2026; it carries its own last-updated date of 28 September 2022, which is three years before the MCI study was published, so this is a page being outrun by its subject rather than a page hiding anything. Examine also maintains more than one page on this compound — a separate research breakdown, now marked archived, whose safety section rests on a 2011 industry assessment of the ingredient and which presents alpha-GPC as neuroprotective given after a stroke. Two pages, one site, two different pictures. Which is the whole argument of this article in miniature: the file is bigger than the page.

The rest of the safety file

Choline intake in general has one large observational signal against it and three against that: higher intake was associated with higher mortality in around 120,000 US health professionals, in an analysis that proposed TMAO as the mechanism without measuring it, while three large cohorts found no association between choline intake and cardiovascular or peripheral-artery endpoints.

The rest of the safety picture is unremarkable. The tolerable upper intake level for choline is 3,500 mg a day, set on hypotension as the critical effect with fishy body odour secondary. The Food and Nutrition Board flags liver disease, kidney disease, Parkinson’s disease, depression and inherited trimethylaminuria as warranting more caution at high intakes. ODS states choline is not known to have clinically relevant interactions with medications, and that is a searched absence rather than an unexamined one.

The TMAO knot, which does not untie

Choline’s other complication is that gut bacteria convert some of it to trimethylamine, which the liver oxidises to TMAO, a metabolite associated with cardiovascular events. Here the evidence genuinely conflicts, and it took longer to work through than anything else across both of these articles.

Pointing one way: 40 healthy volunteers ate two hard-boiled eggs with a capsule of deuterium-labelled phosphatidylcholine and their TMAO rose over the following hours; in the six who repeated the challenge after a week of broad-spectrum antibiotics the rise was suppressed, returning once they stopped — clean evidence that the pathway is microbial. In the same paper, among 4,007 patients undergoing elective coronary angiography, the highest quartile of fasting TMAO carried a hazard ratio of 2.54 (1.96–3.28) against the lowest quartile for death, heart attack or stroke over three years. That is a biomarker-to-outcome association in cardiac patients. The challenge shows a metabolite being produced; harm from a breakfast is a further claim nobody made. We read this at abstract level.

Pointing the other way, on chemical form. A German crossover gave six healthy men four single doses of labelled choline — choline chloride (the plain salt), phosphorylcholine, alpha-GPC (the capsule molecule), or a phosphatidylcholine (the egg form) — six weeks apart, tracking the label for seven days. Neither the phosphatidylcholine form nor phosphorylcholine produced detectable labelled TMAO; alpha-GPC and choline chloride did. But read the conclusion rather than the headline: the egg form peaked latest and gave the highest plasma phosphatidylcholine exposure, and the authors report no difference in AUC over time for labelled choline and betaine after it. On the measure that matters for exposure, the egg form moved the choline differently rather than in greater quantity. Six men, one dose each, no p-values in the abstract, and the authors’ interest was preterm infants rather than nootropic users. In rats, dietary phosphatidylcholine did not raise plasma TMAO while glycerophosphocholine and choline chloride did — five rats per group, and the feeding dose was not in what we retrieved. And in two four-week human egg-feeding crossovers — Lemos and colleagues in thirty young adults, and the same group’s replication in twenty-three adults with metabolic syndrome, the pair discussed in the companion piece — fasting TMAO did not change at all. Both were funded by the Egg Nutrition Center, the research arm of the American Egg Board. The Egg Nutrition Center also publishes a longer list of egg-and-TMAO studies whose titles report the same thing; we did not open them, and a title is not a result. That is the same organisation paying for the reassurance and compiling the list of reassurances, and we are telling you because the finding suits the argument we are making.

For completeness: the study’s own abstract states that glycerophosphocholine promotes atherosclerosis in hyperlipidaemic Apoe-knockout mice — mouse work, not a human result, and a 2026 review notes that in mice engineered to express CETP, which humans have and ordinary lab mice do not, neither choline nor TMAO affected atherosclerosis at all.

The distinction that dissolves most of the apparent contradiction is one almost nobody makes: an acute spike after a single challenge and an unchanged fasting baseline after four weeks of daily eggs are different measurements. Someone saying “eggs raise TMAO” and someone saying “eggs don’t raise TMAO” can both cite real human data and both be describing what they measured. Neither has shown that eating eggs causes cardiovascular events.

What stays unresolved is the form question, and the honest count of evidence on it is two studies, not the longer list this article could have assembled. Only two compared forms head to head: the rat feeding study, which put phosphatidylcholine against glycerophosphocholine and choline chloride, and the six-man tracer crossover, which put four forms against each other. Both found the egg form producing less TMAO. Nothing else here bears on the question. The mouse work gave one compound to one set of animals and supports the narrower claim that glycerophosphocholine raises TMAO, which is where it sits two paragraphs above. The four-week fasting data is a null on both arms, eggs and salt alike, and a study in which nothing moved cannot testify about which form moves it less. One older human egg challenge points the other way. Two studies against one, all of them small, is the whole of it. I wanted this to resolve, and it doesn’t.

What to do with all of this

If you take alpha-GPC at a retail dose, the cohort is not about you and neither is the reassurance. It studied people over 50 on a prescription dose for a cognitive indication. That cuts both ways: the 43% does not transfer to you, and neither does the MCI cohort’s protective finding. Nobody has studied a healthy adult in their thirties taking 300 mg. You are absent from the data in both directions, which is a different position from being safe.

If you are over 50, or taking it for cognitive complaints, you are closer to the population that was studied — and that population has two results, not one. This is a conversation to have with a doctor who can see your vascular risk, not one to settle from a hazard ratio on a supplement page.

Do not use TMAO as a reason either way. The measurements people cite are not measuring the same thing. Nobody has shown that eating eggs, or taking a capsule, causes cardiovascular events.

Check whether the page you are reading mentions both cohorts. Search results here are dominated by pages with something to sell, and at least one of them attributes this finding to a journal, an author and an outcome that do not go together. A page that gives you the first cohort without the second is giving you half the file; a page that gives you neither is not worth your attention.

What is genuinely established here is smaller than it looks and worth holding onto anyway: a widely sold supplement has a stroke association in the population that takes it as a drug, a second study in a related population finds the opposite, and the most trusted reference page in the category carries the first and has not caught up with the second. That is the honest state of it, and it is not the state of it you will find anywhere else.

About this article

Written by Leah Elish. Leah covers food-first nutrition — what a nutrient does, how much of it is in a meal, and what the capsule adds to that. Not a dietitian or clinician — traces nutritional claims back to the studies that supposedly support them.

Medical review: None. NeuriFuel does not currently have a licensed clinician on the editorial team, and this article has not been medically reviewed. We state this rather than implying an authority we do not have. This piece describes observational evidence about a compound sold as a supplement and prescribed as a drug; it is not medical advice, and the decision it bears on is one to take with a doctor. See our About page for our full methodology.

Sources: 14 numbered sources. One entry bundles four papers, so 17 individual works are referenced. Among the 14: two national health-insurance cohorts covering more than twelve and a half million people between them, three human randomised crossover trials, one human challenge study with an attached patient cohort, four epidemiological cohorts reported as a single entry, two animal or in vitro studies, a narrative review, an industry ingredient-safety assessment cited through a third party, a government fact sheet, one independent supplement reference, and one commercial page that the article criticises and that is listed here because we read it in full and quote what it says. Neither Korean cohort’s full text could be opened; the figures come from published abstracts, summary boxes and extracted text, and the article says so wherever it matters. Two of the trials cited here were funded by the egg industry and that is disclosed in the text as well as here. Funding is given where it was retrieved and its absence is marked where it was not.

Corrections: Found an error? Write to hello@neurifuel.com with a source and we will fix it and log the correction.

Last updated: 5 September 2026

References

1. National Institutes of Health, Office of Dietary Supplements. Choline — Fact Sheet for Health Professionals. Updated 2 June 2022. Read in full.
2. Böckmann KA, Franz AR, Shunova A, Minarski M, Wiechers C, Poets CF, Bernhard W. “Different choline supplement metabolism in adults using deuterium labelling.” Eur J Nutr 2023;62(4):1795–1807. DOI 10.1007/s00394-023-03121-z · PMID 36840817 · PMC10195734 · trial DRKS00020498. Authors declare no conflict of interest; funders not retrieved. Read at abstract level plus two extracted discussion paragraphs.
3. Shirouchi B, Fukuda A, Akasaka T. “Unlike Glycerophosphocholine or Choline Chloride, Dietary Phosphatidylcholine Does Not Increase Plasma Trimethylamine-N-Oxide Levels in Sprague-Dawley Rats.” Metabolites 2022;12(1):64. DOI 10.3390/metabo12010064. Rat study, n = 5 per group. Read at abstract level; funding not retrieved.
4. Tang WHW, Wang Z, Levison BS, Koeth RA, Britt EB, Fu X, Wu Y, Hazen SL. “Intestinal microbial metabolism of phosphatidylcholine and cardiovascular risk.” N Engl J Med 2013;368(17):1575–1584. DOI 10.1056/NEJMoa1109400 · PMID 23614584. Read at abstract level; funding not retrieved.
5. Lemos BS, Medina-Vera I, Malysheva OV, Caudill MA, Fernandez ML. “Effects of Egg Consumption and Choline Supplementation on Plasma Choline and Trimethylamine-N-Oxide in a Young Population.” J Am Coll Nutr 2018;37(8):716–723. DOI 10.1080/07315724.2018.1466213 · PMID 29764315. Funded by the Egg Nutrition Center, research arm of the American Egg Board, and by CNPq. Read at abstract level.
6. Conference abstract, not a full paper. Thomas MS, DiBella M, Malysheva OV, Caudill MA, Blesso C, Fernandez ML. “Intake of 3 Eggs/Day or Equivalent Amount of Choline as Supplement for 4 Weeks Increases Plasma Choline Without Changing Plasma TMAO in Participants with Metabolic Syndrome.” Current Developments in Nutrition; PMC7258582. Funded by the Egg Nutrition Center, stated on the record itself. Published in an American Society for Nutrition supplement; the record carries no full text beyond the abstract, which is what we read. Volume and pagination not retrieved.
7. Lee G, Choi S, Chang J, Choi D, Son JS, Kim K, Kim SM, Jeong S, Park SM. “Association of L-α Glycerylphosphorylcholine With Subsequent Stroke Risk After 10 Years.” JAMA Netw Open 2021;4(11):e2136008. DOI 10.1001/jamanetworkopen.2021.36008 · PMC8613599. Funded by the National Health Insurance Service of Korea and the National Research Foundation of Korea (2017R1D1A1B03033721); no conflicts of interest reported. Full text could not be accessed directly; figures above are from the published abstract, the summary box and an extracted PDF text.
8. Wang Z, Hazen J, Jia X, et al. “The Nutritional Supplement L-Alpha Glycerylphosphorylcholine Promotes Atherosclerosis.” Int J Mol Sci 2021;22(24):13477. DOI 10.3390/ijms222413477. Mouse and cell study. Read at abstract level; funding not retrieved.
9. “A Friend or Foe…” Nutrients 2026;18(10):1526. DOI 10.3390/nu18101526. Narrative review, read at abstract level; source of the CETP-model counter-evidence, which is described rather than opened here.
10. Zheng Y, et al. Am J Clin Nutr 2016;104:173–180. PMID 27281307 — higher choline intake and mortality; TMAO was not measured. Against three large null cohorts on cardiovascular endpoints: Bertoia 2014 (PMID 24819748), Bidulescu 2007 (PMID 17629908), Dalmeijer 2008 (PMID 17375117). All four read at secondary level, through the ODS fact sheet.
11. Examine.com, alpha-GPC page, https://examine.com/supplements/alpha-gpc/ — read in full on 30 August 2026; the page carries its own last-updated date of 28 September 2022. The separate research breakdown at /supplements/alpha-gpc/research/ is marked archived and was read at Phase 1. The page is edited over time and this describes the version read; the citicoline page was not read.
12. Brownawell AM, Carmines EL, Montesano F. “Safety assessment of AGPC as a food ingredient.” Food Chem Toxicol 2011. Industry ingredient safety assessment; cited by Examine and not opened here.
13. “Association between L-α glycerylphosphorylcholine use and delayed dementia conversion: a nationwide longitudinal study in South Korea.” J Prev Alzheimers Dis; PII S2274580725000032 · PMID 40155153 · PMC12184023. Funding and conflicts not retrieved. Read at abstract level plus two extracted discussion paragraphs.
14. Nectr Energy, “Citicoline vs. Alpha-GPC: Which Nootropic Is Better for Focus?”, https://nectr.energy/blogs/life-hacks/citicoline-vs-alpha-gpc-which-nootropic-is-better-for-focus — published 9 February 2026, read in full 30 August 2026. A commercial page selling a citicoline product, listed because this article quotes and criticises it. Its citation to a 2021 JAMA Internal Medicine study by Sun et al. does not correspond to any paper we could identify.