Does Alcohol Cause Cancer? Yes — and the “Glass a Day Is Good for You” Finding Came From Miscounting Who Had Quit

Yes. The evidence is large, old and consistent, and the dose-response is the part that settles it. Across 572 studies and 486,538 cancer cases, more alcohol meant more cancer of the mouth, throat, oesophagus, colon, larynx and breast — steadily, not at some threshold. [1]

The interesting question is not whether. It is why so many people believe a glass a day is actively good for them, and the answer to that is one of the cleanest examples of a research error escaping into the world that we have covered.

What it raises, and by how much

One finding in that table points the other way and we are printing it anyway: the same analysis found alcohol inversely associated with Hodgkin and non-Hodgkin lymphoma. [1] It is a real result in a good paper, nobody has explained it, and leaving it out would be the selective quoting this page exists to correct.

Now the glass of red wine

For decades the data showed something odd: people who drank moderately died less often than people who did not drink at all. Plot risk against intake and you got a J — a dip at the bottom, then a climb.

That dip is where “a glass a day is good for your heart” came from. It was real in the data. It was also an artifact of who got counted as a non-drinker.

Here is the mistake. If you group everyone who is not currently drinking together, your “non-drinkers” include people who used to drink and stopped — and a lot of people stop because they have become ill. So the abstaining group quietly fills up with sick people, and moderate drinkers look healthy by comparison.

A 2023 analysis of 107 studies, 4,838,825 people and 425,564 deaths went back and adjusted for exactly that. [2]

The protection disappeared. Occasional drinking came out at 0.96 — not statistically significant. Low-volume drinking, 0.93 — also not significant. [2] The dip in the J was, to a good approximation, the sick people.

And above that the risk climbed: significantly higher mortality for women drinking 25 grams a day or more, and for men at 45 grams or more. [2]

A note on what a gram is, because this is where the reporting usually gets sloppy. These studies measure grams of pure alcohol, not “drinks”, because a drink is not a fixed quantity — it differs by country and by pour. The paper’s own translation is that 25 grams is about two Canadian standard drinks. [2] We are quoting that rather than converting it for you, because the conversion is exactly the step where these numbers usually go wrong.

Then somebody found a natural experiment

Adjusting for a bias is an argument. What settles it is a design where the bias cannot operate at all.

Some people carry gene variants that make alcohol physically unpleasant — the flush reaction — so they drink far less, for life. Which variant you carry was decided at conception, before anything about your health, your income or your habits. Comparing people by that genotype is much closer to a randomised trial than any survey can be.

A study of 512,724 adults in China, followed twelve years with 56,550 deaths, did precisely this. In the genotyped group, those variants predicted sixty-fold differences in how much people actually drank — from 4 grams a week to 255. [3]

The same study produced both answers. Its conventional analysis found the familiar pattern: non-drinkers and ex-drinkers had higher death rates than moderate drinkers. Its genetic analysis found that genotype-predicted alcohol intake was “uniformly and positively associated” with dying — of anything (1.07), of alcohol-related cancers (1.12), of liver disease (1.31), of cardiovascular disease (1.15), mostly stroke. [3]

The authors’ conclusion, in their own words: “There was no genetic evidence of protection from moderate drinking for all-cause and cause-specific mortality, including CVD.” [3]

And one result in that study did not reach significance, which we are reporting because it is the kind of detail that gets dropped: for ischaemic heart disease specifically, the genetic estimate was 1.06, with a range running from 0.99 to 1.14 — so “no effect” is still inside it. That is how we read an interval. The heart-disease case is the weakest part of the picture; the cancer and stroke case is not.

Worth knowing who paid: that study’s funders included Cancer Research UK, the Wellcome Trust and GlaxoSmithKline. [3] We keep a running count of who funds what we cite.

What this does and does not tell you

It does not tell you your odds. Everything above is a multiplier on a baseline risk, and those baselines differ enormously between cancers and between people. None of the sources here publishes the underlying rates, so we are not putting a number on your personal risk — and you should be suspicious of any page that does without showing its working.

It does tell you the direction is not in doubt, and there is no floor. Three different designs — pooled observation, bias-corrected observation, and a genetic natural experiment — point the same way, and the dose-response means less is genuinely less.

What we are not going to do is lecture you. Plenty of things worth doing carry a risk, and people are allowed to weigh that themselves once somebody has told them the truth about it. The target here is the forty years of marketing that sold wine as a health product on the back of a statistical artifact — not anybody who believed it.

One thing that is not a lifestyle question. If drinking is something you have tried to cut down and could not, that is a clinical matter with real treatments behind it, and it belongs with a doctor rather than an article.

Elsewhere on the harder questions: saturated fat, red meat, the Blue Zones — where wine was part of the story — and what actually lowers blood pressure. Sleep is the other one people underrate.