Alcohol

Alcohol was the last of the big items on this site's order-of-magnitude table without a chapter, and it is the one where the public understanding has changed most in the last decade — largely in one direction, and largely without anyone announcing it.

The short version: the belief that a moderate amount is good for your heart was built on a study design that could not distinguish between alcohol protecting people and healthy people drinking moderately. When better designs were applied, the protection largely disappeared. What did not disappear was the cancer risk, which starts lower than most people expect.

1. What the UK guidance actually says, and what the number means

The UK Chief Medical Officers advise no more than 14 units a week, the same for men and women, spread over three or more days rather than saved up, with several drink-free days. A UK unit is 10 ml or 8 g of pure alcohol — roughly half a pint of ordinary-strength beer, or a single measure of spirits. A large glass of wine is closer to three units than one, which is where most people's mental arithmetic goes wrong.

The important thing about 14 is what it is not. It is not a safety threshold and there is no cliff edge at 15. The guidelines are explicitly for low-risk drinking rather than safe drinking, and 14 units is the point at which the modelled lifetime risk of dying from an alcohol-related cause reaches about 1 in 100.

Whether 1 in 100 is acceptable is a judgement, not a medical fact, and it is yours to make. For scale, David Spiegelhalter's comment when the guidance was published is the most useful thing anyone said about it: on his estimate, an hour of television a day, or a couple of bacon sandwiches a week, carries more lifetime risk than drinking at the guideline level.

2. The J-curve, and why it mostly is not there

For thirty years the standard chart showed a J: abstainers at moderate risk, light drinkers lowest, heavy drinkers highest. It shaped a generation of advice and a great deal of marketing.

The problem is the left-hand end. The comparison group — non-drinkers — is not made up of people who happen not to drink. It contains people who stopped, often because they were unwell, and people who never started, often for reasons that travel with worse health. This is the sick-quitter effect, and it makes moderate drinkers look healthy by comparison with a group that was already ill. Moderate drinkers are also, on average, wealthier, more socially connected and in better health to begin with — and no amount of statistical adjustment fully removes that.

Mendelian randomisation gets around the problem, for the reason set out in the misconceptions chapter: gene variants affecting how much people drink are allocated at conception, so they cannot be caused by illness, wealth or anything else that confounds the observational picture.

Applied to alcohol, the picture is consistent and it is not a J. A linear and non-linear Mendelian randomisation analysis published in the International Journal of Epidemiology in 2024, in 278,093 white-British UK Biobank participants, found that each additional unit (8 g) per day of genetically predicted intake was associated with an odds ratio of 1.27 for all-cause mortality, 1.30 for cardiovascular disease and 1.20 for cancer — with an essentially straight line and no protective benefit at modest intake. Earlier individual-participant work in the BMJ reached the same conclusion for coronary heart disease specifically, and a UK Biobank analysis found alcohol was not protective for any cardiometabolic outcome by any method or in any subgroup tested. Strong

The counterweight. This is not unanimous. Reviews that weight the observational literature heavily — and that judge most of those studies to have handled the sick-quitter problem adequately — still report a protective association with ischaemic heart disease in men, maximal at around 20 g of ethanol a day. The disagreement is substantially about how much credence to give observational data once genetic data point the other way. This site takes the genetic evidence as the stronger signal, for the same reason it does with lipids, and says so rather than pretending the argument is over.

Two other things are worth separating out. Alcohol does raise HDL, which is where much of the original mechanism came from — and raising HDL has repeatedly failed to reduce events, which is covered in the misconceptions chapter. And whatever is true about ischaemic heart disease, alcohol clearly raises blood pressure and clearly causes atrial fibrillation, both dose-dependently. A cardiovascular defence of drinking has to net those off, and usually does not.

3. Cancer is the part that gets left out

Alcohol is a Group 1 carcinogen — the same International Agency for Research on Cancer category as tobacco and asbestos, which describes the strength of the evidence that it causes cancer in humans, not the size of the risk. It is causally linked to cancers of the mouth, throat, oesophagus, larynx, liver, colon and rectum, and breast. Strong

Two features make it different from the cardiovascular story. There is no threshold — risk rises from the first drink rather than from some level of excess. And breast cancer is affected at low intakes, which means the risk is not confined to heavy drinkers or to men.

A concrete example from the Sheffield modelling that underpins the 2016 CMO guidelines, because relative risks are useless here. For oesophageal cancer in men: about 6 cases per 1,000 in lifelong abstainers, about 13 per 1,000 in men drinking within the 14-unit guideline, and about 25 per 1,000 at 14–35 units a week. That is roughly a doubling at guideline level — of a number that is small. Both halves of that sentence matter, and most coverage drops one of them.

4. What this means in practice

• There is no level at which drinking improves your health. That is a change from the advice of twenty years ago and it is now the position of the UK guidance, which dropped the cardioprotection claim in 2016.
• There is a level at which the risk is small enough that most people will reasonably accept it. That is what 14 units describes. Drinking below it is lower risk again; there is no floor below which further reduction stops helping.
• Pattern matters as much as volume. Fourteen units across a week is a different exposure from fourteen units on Saturday. Binge patterns carry injury, arrhythmia and blood pressure risks that spreading the same quantity does not.
• Nobody should start drinking for their health, and nobody with an existing reason to stop should be reassured by a J-curve that the genetics do not support.
• If you are cutting down, blood pressure is where you will see it first. A few weeks of reduction shows up on a home monitor, which makes it one of the more satisfying changes to measure. See the blood pressure chapter.
• Alcohol degrades sleep even when it shortens the time taken to fall asleep. It suppresses the restorative early part of the night and fragments the second half. See the sleep chapter.