"A glass of red wine is good for you" has been repeated so often it feels like settled science. The research behind it has been unravelling for years, and a 2025 dose-response analysis of 20 cohort studies is some of the clearest evidence yet on what alcohol does to cognition. This piece is not an argument for drinking or against it. It's an account of why the famous curve appeared, and what happened when researchers found ways to test it properly.
What a dose-response meta-analysis actually does
Rather than comparing "drinkers vs. non-drinkers" as one blunt category, a dose-response meta-analysis plots cognitive outcomes against the actual amount consumed, across a pooled 20 cohort studies. Published in Addiction in 2025, it found a systematic relationship between alcohol intake and cognitive decline without the protective dip at "moderate" levels that older, more limited studies had suggested.1 Like everything in the first half of this article, it is observational — which is why the second half matters.
The sick-quitter problem, quantified
The "J-curve" — lowest risk among light drinkers, higher among both abstainers and heavy drinkers — has a mundane explanation that has now been measured. The abstainer group in most cohorts is not a clean comparison: it contains people who stopped drinking because they became ill, people on medication that precludes alcohol, and people who are poorer and sicker for unrelated reasons.
A 2023 systematic review in JAMA Network Open pooled 107 cohort studies covering 4,838,825 participants and 425,564 deaths, split by whether they had controlled for that bias.3 Across all studies, low-volume drinkers (1.3–24.0 g of ethanol a day) showed a relative risk of all-cause mortality of 0.93 — the familiar apparent benefit, and not statistically significant (p = 0.07). Restricted to studies free of abstainer bias, the same comparison came out at 1.19 (p < 0.001): a modest increase, not a decrease. Risk was clearly elevated at 45–64 g a day (RR 1.19) and 65 g or more (RR 1.35), and larger in women (RR 1.22).
The same erasure shows up for cognition specifically. A 2025 study in Frontiers in Aging Neuroscience re-ran the moderate-drinking-and-cognition question while explicitly controlling for income and cultural factors, and the apparent protective effect disappeared.2
Once income and cultural factors were controlled for, moderate drinking's "protective" effect on cognition disappeared — it was measuring who drinks moderately, not what moderate drinking does.
Mendelian randomisation: the design built to get past confounding
There will never be a randomised trial assigning people to decades of moderate drinking. Mendelian randomisation is the closest substitute. Variants in the alcohol-metabolising genes ADH1B and ALDH2 cause flushing and nausea and so suppress intake — and they are fixed at conception, so they cannot have been caused by your income, health or culture.
The landmark application appeared in The Lancet in 2019: over 500,000 Chinese adults, roughly 160,000 genotyped for rs671 (ALDH2) and rs1229984 (ADH1B).4 Conventionally the cohort reproduced the familiar pattern — one or two drinks a day looked slightly protective against stroke. Genetically, it did not: genotype-predicted intake raised stroke risk by about 35% for every four additional drinks a day (280 g a week), with blood pressure rising in step. For heart attack the picture was, in the authors' words, less clear-cut — and they said so rather than rounding it up.
A 2025 analysis in BMJ Evidence-Based Medicine did the same for dementia.5 Observationally, across 559,559 participants in the US Million Veteran Program and UK Biobank with 14,540 incident dementia cases, the U-shape appeared: non-drinkers, people drinking more than 40 drinks a week (HR 1.41, 95% CI 1.15–1.74) and people with alcohol use disorder (HR 1.51, 1.42–1.60) all had higher risk than light drinkers. The genetic analyses — drawing on 2.4 million people across 45 GWAS cohorts — found a monotonic increase instead: a one-standard-deviation rise in genetically predicted log drinks per week carried 15% higher dementia risk (OR 1.15, 1.03–1.27), and the non-linear analysis found no U-shape at all. They also demonstrated the mechanism that manufactures one: people who went on to develop dementia had been reducing their intake beforehand, so the light-drinking group looks healthy partly because those getting ill have already left it.
Where Mendelian randomisation stops
It is a better design, not a perfect one, and both sets of authors say so. MR rests on assumptions that cannot be verified — chiefly that the variants affect the outcome only through alcohol. Power was concentrated in participants of European ancestry, and the ALDH2 flushing variant behind the Chinese analysis is largely absent in Europeans. An MR estimate also reflects lifetime accumulated exposure rather than what happens if you change your drinking now — and the lowest-dose estimates are the least precise of all.
What brain imaging adds, and what it can't
A 2022 study in Nature Communications examined 36,678 UK Biobank participants with brain imaging.6 Higher intake was associated with lower grey and white matter volume, widespread rather than localised: zero to one daily unit was a very small change, one to two units comparable to about two years of ageing for grey matter, two to three about 3.5 years. Their own limitations: cross-sectional, so no causal claim; self-reported intake; no adjustment for past alcohol use disorder.
What this doesn't mean
This isn't a claim that one glass with dinner is doing damage you would ever detect. The relationship is gradual and the effects at the low end are small and imprecisely estimated. What it does mean is narrower and firmer: "moderate drinking protects your brain" does not survive the studies designed to test it. That is not the same statement as "any alcohol is dangerous," and we are not making it.
Numbers you can actually use
The UK Chief Medical Officers' low-risk guideline is no more than 14 units a week, the same for men and women, spread over three or more days with several drink-free days. A pint of 4% beer is about two units, a single 25ml spirit measure is one, and a 175ml glass of 13% wine is roughly 2.3 — so 14 units is about six medium glasses of wine.7 "Low-risk" is the guideline's own word, not "no risk." If you'd like to drink less and find it harder than expected, that's a recognised clinical problem with real treatments: the treatments that work, and why willpower isn't one of them.
What actually protects cognition, evidence-wise
If the goal is long-term brain health, the interventions with the most consistent evidence aren't a beverage — they're sleep, aerobic exercise, hearing, and the cardiovascular risk factors that govern blood flow to the brain. The per-factor breakdown is in the 45% dementia figure and where to start; the vascular half is in the ten things that cause nine in ten strokes, where alcohol and blood pressure both appear — the pathway the Lancet genetic analysis identified. On exercise, which type actually helps is the more useful question. On sleep, the thing most people get wrong is using alcohol for it: what a nightcap actually does to your night. And if memory changes are the worry, start with what's normal ageing and what needs an assessment.
No supplement, ours included, offsets alcohol or protects the brain from it, and we won't imply one does. If you take anything from our range and drink, see are Proco supplements safe with alcohol.
The honest takeaway
Drink or don't — that's a personal choice, and this is not an argument for abstinence any more than for a daily glass. But if the reasoning is specifically "it's good for my brain," the evidence doesn't support it.
References
- "Alcohol consumption and cognitive function: A systematic review and dose-response meta-analysis of 20 cohort studies." Addiction, 2025. onlinelibrary.wiley.com
- "Moderate alcohol consumption does not protect cognitive function when controlling for income and cultural factors." Frontiers in Aging Neuroscience, 2025. pmc.ncbi.nlm.nih.gov
- Zhao J, Stockwell T, Naimi T, Churchill S, Clay J, Sherk A. "Association Between Daily Alcohol Intake and Risk of All-Cause Mortality: A Systematic Review and Meta-analyses." JAMA Network Open, 2023;6(3):e236185. 107 cohort studies, 4,838,825 participants. jamanetwork.com
- Millwood IY, Walters RG, Mei XW, et al. "Conventional and genetic evidence on alcohol and vascular disease aetiology: a prospective study of 500 000 men and women in China." The Lancet, 2019;393(10183):1831-1842. thelancet.com
- Topiwala A, Levey DF, Zhou H, et al. "Alcohol use and risk of dementia: observational and Mendelian randomisation analyses." BMJ Evidence-Based Medicine, 2025. 559,559 participants observationally; 2.4 million in the genetic analyses. brainlife.org
- Daviet R, Aydogan G, et al. "Associations between alcohol consumption and gray and white matter volumes in the UK Biobank." Nature Communications, 2022. n = 36,678. nature.com
- UK Chief Medical Officers' low-risk drinking guidelines, as summarised by Drinkaware. drinkaware.co.uk
This article is for informational purposes only and has not been evaluated by the FDA. It is not intended to diagnose, treat, cure, or prevent any disease. This is not medical advice about alcohol use; speak with a healthcare provider about your own drinking and any related health concerns.
Related reading: Alcohol and Sleep: What a Nightcap Actually Does to Your Night and Modifiable Dementia Risk: What the Research Actually Shows
See Omega-3's full dosing and cardiovascular-adjacent evidence at procohq.com.