Is moderate drinking actually healthy? The red-wine myth just collapsed under better methods

The story you were sold: the rise of the healthy drink

For a long stretch of the late twentieth and early twenty-first century, the science seemed genuinely settled, and it was lovely news. Study after study found the same curve: people who drank a little — a glass of wine most nights, a beer with dinner — had lower rates of death and heart disease than people who drank nothing at all, and lower rates than heavy drinkers. Plot risk against intake and you got a J-shape (often called the J-curve or U-curve): risk dipping below the teetotaller baseline at low intake, bottoming out somewhere around one drink a day, then climbing steeply as intake rose.

That dip was the whole story. It became the “French paradox,” the resveratrol-in-red-wine headlines, the cardiologist quietly telling patients a glass a night was fine, maybe even good. It felt mechanistically plausible — modest alcohol nudges HDL cholesterol up and can affect clotting — and it was repeated so many times it hardened into common knowledge. The comforting part was not just that drinking was allowed; it was that not drinking looked slightly risky. Abstinence itself seemed to sit on the wrong, higher part of the curve.

Here is the problem I want to walk you through carefully, because it is one of the cleanest case studies in how observational epidemiology fools itself. That dip was largely an artifact. Not a lie, not fraud — an honest statistical mirage produced by how the studies were built. And once you see how it was built, you cannot unsee it.

The flaw hiding in the J-curve

Two flaws, really, and they stack.

The first and most damaging is abstainer bias — often called the “sick-quitter” problem. Ask yourself who ends up in the “non-drinker” comparison group in a typical study. Some are lifelong teetotallers. But a large share are former drinkers — people who quit. And why do people quit drinking? Frequently because they got sick: a doctor told them to stop after a cardiac scare, a liver warning, a cancer diagnosis, the start of a medication that does not mix with alcohol. So the “non-drinker” group is quietly seeded with people who are already ill or in decline. Compare a healthy light drinker against a group padded with sick quitters, and of course the light drinker looks healthier. The alcohol did not protect them; the comparison group was rigged against abstinence by the very illnesses that caused people to abstain.

The second flaw is ordinary confounding. In many Western populations, moderate drinkers are, on average, wealthier, better educated, more likely to exercise, more likely to have good healthcare, and less likely to smoke heavily than either heavy drinkers or lifelong abstainers. Moderate drinking travels with a whole bundle of advantages that themselves lower mortality. Statistical adjustment helps, but it never fully removes this — you cannot adjust for what you did not measure well, and “the kind of person who has a glass of wine with dinner” carries advantages that are hard to fully capture on a questionnaire.

Put those two together and the J-curve’s protective dip has an entirely mundane alternative explanation that has nothing to do with alcohol being good for you: the reference group was sicker than it should have been, and the moderate drinkers were healthier to begin with. The question the field finally asked, properly, was: what happens to the dip when you correct for those biases?

The moderate drinker did not outlive the abstainer because alcohol is medicine. He outlived a comparison group secretly stocked with people who had already been told to stop drinking.

The better evidence: what bias correction did to the benefit

This is the hinge of the whole article, and it turns on one study in particular.

In 2023, Zhao, Stockwell, Naimi and colleagues published a systematic review and meta-analysis in JAMA Network Open that pooled 107 cohort studies and more than 4.8 million participants — and, crucially, examined how the apparent benefit changed depending on study quality. Their key move was to separate the studies that were vulnerable to abstainer bias and confounding from the studies that had guarded against them (for example, by using only lifetime abstainers as the reference, and by properly accounting for the age and characteristics of the cohorts).¹

The result is the punchline of the modern literature. Across all studies naively pooled, the familiar protective dip appeared. But in the analyses adjusted for those biases, the mortality reduction at low intake was no longer statistically significant — light-to-moderate drinkers did not have a meaningful survival advantage over lifetime non-drinkers. Meanwhile the harm at higher intake remained clear: a significantly raised risk of all-cause mortality kicked in for women at roughly 25 grams a day and for men at roughly 45 grams a day (a standard drink is about 10–14 grams).¹ In plain terms: the better the study, the smaller the “benefit,” until it vanished. That is exactly the fingerprint of an artifact, not a real effect. Real biological effects get stronger as the methods get cleaner; artifacts get weaker and disappear.

This did not come out of nowhere. The GBD 2016 Alcohol Collaborators, in a sweeping Lancet analysis of alcohol’s burden across 195 countries, had already reached a blunt conclusion at the population level: weighing all health outcomes together, the level of consumption that minimises health loss is zero.² Any small cardiovascular signal at low intake, they found, was outweighed across a population by alcohol’s other harms, cancer chief among them. And the Wood 2018 combined analysis of nearly 600,000 current drinkers put the threshold for lowest all-cause mortality risk down around 100 grams of alcohol per week — well below many national guidelines at the time — with a roughly linear rise in stroke, heart failure, and several other cardiovascular outcomes above that.³ The trend was already unmistakable: the safe, protective zone kept shrinking every time someone measured it more carefully.

The genetics: Mendelian randomization removes the fog

Here is the move that really closed the case, because it sidesteps abstainer bias and most confounding altogether.

The technique is Mendelian randomization (MR), and the logic is elegant. Certain gene variants affect how much alcohol people drink or how they metabolise it — in East Asian populations, for instance, variants in the ALDH2 and ADH1B genes make drinking unpleasant (the flushing response), so carriers drink far less. Crucially, you are dealt those genes at conception, at random, decades before you develop heart disease or cancer, and independently of your wealth, your diet, or whether you later got sick and quit. So if you compare health outcomes across genetically lighter versus heavier drinkers, you get something close to a natural randomized experiment — the sick-quitter and the wealthy-wine-drinker confounds mostly wash out.

The Millwood 2019 study in The Lancet did exactly this in the China Kadoorie Biobank — roughly 500,000 adults, with genotyping in a large subset. Conventional (self-reported) analysis showed the usual U-shape, with moderate male drinkers apparently protected against stroke. But the genetic analysis told a different story: alcohol had a continuous, roughly linear relationship with stroke risk, with no protective effect at low intake. The moderate-drinking “protection” against stroke was a confounding illusion; the genetics showed steady harm.⁴

The Biddinger 2022 study in JAMA Network Open ran the same logic in the UK Biobank, roughly 371,000 people, focused on cardiovascular disease. It found the genetic relationship between alcohol and cardiovascular outcomes was consistently risk-increasing — nonlinear, with only minimal increases at light intake but exponentially rising risk of hypertension and coronary artery disease as intake climbed. Again, no protective valley, no free lunch at one drink a day.⁵ When two large genetic studies in two very different populations both erase the protective dip that the self-report studies kept finding, the most parsimonious reading is that the dip was never real — it was the artifact the bias-correction work independently predicted.

The cancer angle nobody put on the wine label

Even if you wanted to cling to a small heart signal, there is a second ledger the red-wine story conveniently left off the label entirely: cancer.

Alcohol is not a “possible” or “probable” carcinogen. It is classified by the International Agency for Research on Cancer (IARC) as a Group 1 carcinogen — the same top tier as tobacco and asbestos, meaning there is sufficient evidence it causes cancer in humans. The mechanism is not exotic: ethanol is metabolised to acetaldehyde, itself a carcinogen that damages DNA and proteins, and alcohol also raises certain hormone levels and acts as a solvent that helps other carcinogens into tissues.

The part that matters for the “moderate” question is that the cancer risk does not politely wait for heavy drinking. The Bagnardi 2015 dose-response meta-analysis in the British Journal of Cancer examined light drinking specifically — up to about one drink a day — and found that even at that level there was a measurable increase in risk for several cancers, with oesophageal squamous-cell carcinoma, oral and pharyngeal cancers, and female breast cancer among the most sensitive. Heavy drinking raised risk far more (multi-fold for upper-aerodigestive cancers), but the point is that the curve for these cancers starts rising from the first regular drinks, not from some threshold of abuse.⁶ Breast cancer is the one that most directly touches the “healthy glass of wine” demographic: the risk increase per daily drink is modest in relative terms but real, and it appears with no safe lower threshold.

This is precisely why, in 2023, the World Health Organization published a statement in The Lancet Public Health with a title that leaves no wiggle room: “no level of alcohol consumption is safe for our health.” Their argument is exactly the cancer one — that because carcinogenicity begins at the very first drop and there is no identified threshold below which cancer risk does not rise, you cannot promise any amount is risk-free.⁷ Whatever small cardiovascular signal might survive at very low intake, it is being weighed against a carcinogen with no safe floor. That is the trade the wine label never mentioned.

What about the heart signal?

I want to be honest here rather than triumphalist, because overcorrecting is its own kind of dishonesty. The fair position is not “alcohol has been proven to have zero cardiovascular upside at any dose.” It is narrower and more careful than that.

There may still be a small, real cardiovascular signal at genuinely low intake for some specific outcomes — ischaemic heart disease in particular, where a modest HDL and clotting effect is biologically plausible and shows up even in some careful analyses. What has changed is the size and certainty and net weight of that signal. It is smaller than the old J-curve implied, shakier once bias is stripped out, applies to some cardiovascular endpoints while other endpoints (stroke, hypertension, atrial fibrillation, several cancers) move the wrong way even at low intake, and is outweighed at the population level by the total harm.²⁵ The Biddinger genetic data captures this nuance well: minimal risk increase at the very bottom of the intake range, but no clean protective valley, and rising harm thereafter.⁵

So the responsible sentence is not “alcohol is pure poison at every dose” — that overstates the certainty in the other direction. It is: any cardiovascular benefit at low intake is real at best in a small, uncertain, endpoint-specific way, and it does not add up to a net health benefit once you include cancer and everything else. That is why I grade the “real and clinically meaningful cardiovascular benefit” claim WEAK rather than HYPE — there is a thread of signal there, just nowhere near strong enough to hang a health recommendation on, and nowhere near enough to make a daily glass a good idea on balance.

Grey areas: absolute risk, guidelines, and honesty

Three honest caveats, because the collapse of the J-curve gets abused in both directions and I do not want to hand anyone ammunition for a moral panic.

First, absolute risk versus relative risk. Most of the scary-sounding numbers are relative risks — “X% higher risk of breast cancer per daily drink.” In absolute terms, for a single individual at low intake, the added risk is small. One glass of wine a night is not a health crisis, and this article is emphatically not telling you a birthday toast will hurt you. The J-curve’s collapse means there is no clear benefit and no risk-free floor — it does not mean light drinking is dangerous in the way heavy drinking is. Both of those can be true at once, and honesty requires holding them together.

Second, the guidelines have already moved, and they moved toward this evidence, not toward prohibition. Canada’s 2023 Guidance on Alcohol and Health reframed the whole thing around a continuum of risk: risk is low at one or two drinks a week, increases from there, and becomes high above roughly six a week — with the honest headline that when it comes to alcohol, less is better, and no amount is entirely without risk. That is not “never drink.” It is “there is no protective sweet spot, so treat every drink as carrying a little risk and let that inform how much you choose.” It is a harm-reduction frame, and it matches the science described above.

Third, the non-drinker takeaway, which is the cleanest practical conclusion in the entire field: if you do not currently drink, there is no health reason to start. The old J-curve was, for a while, an actual argument that abstainers should perhaps take up a nightly glass for their hearts. That argument is dead. Nothing in the current evidence supports a non-drinker starting for health. And a few contexts are non-negotiable regardless of the debate: no amount is considered safe in pregnancy; alcohol interacts with a long list of medications; and if drinking has stopped feeling like a choice, that is alcohol use disorder — a medical condition with real treatment, and a reason to talk to a clinician rather than white-knuckle it alone.

The verdict

The story that a moderate, daily drink is good for you — the French paradox, the heart-healthy glass of red, the sense that abstaining was faintly risky — was one of the most comforting ideas in modern health, and it has largely collapsed under better methods. The protective J-curve turned out to be built substantially on abstainer bias and confounding; correct for them, as the Zhao 2023 meta-analysis of 4.8 million people did, and the mortality benefit at low intake evaporates.¹ Genetic studies that bypass those biases entirely — Millwood in China, Biddinger in the UK — find mostly steady harm and no protective floor, especially for cardiovascular disease.⁴⁵ And alcohol’s status as a Group 1 carcinogen with no safe threshold, spelled out by Bagnardi and endorsed by the WHO, means even light drinking carries a real cancer cost the wine label never listed.⁶⁷ So the “healthy protective level” claim grades WEAK, the “robust cardiovascular benefit” claim grades WEAK, and “a daily glass of red wine is good for your health” grades HYPE.

But let me land this without the sanctimony, because the moral-panic version is as dishonest as the health-halo version. This is not a call for prohibition, and it is not a claim that your occasional drink is a health emergency — at low intake the absolute risks are small, and how you weigh a small risk against a real pleasure is genuinely your call. What has changed is only this: the reasons you may have been given to drink for your health do not survive the better evidence. There is no established healthy level, no protective sweet spot, and no risk-free floor. If you do not drink, don’t start for your heart. If you do drink, the honest, non-preachy summary is the one Canada’s own guidance landed on — less is better. Not zero as a mandate; less as a direction. Judged as what it actually is — a mild pleasure with a small, dose-dependent cost and no proven health upside — alcohol is fine to choose with your eyes open. It just isn’t medicine, and the glass of red was never the reason anyone was healthy.

For the rest of the “is this common habit actually good or bad for you?” map, our reads on coffee and longevity, the 10,000-steps myth, whether statins are overprescribed, sauna and cold contrast therapy, and forest bathing sit right next to this one — each one an honest audit of a belief that got ahead of its evidence.