Alcohol

Alcohol raises cancer risk at every level of drinking anyone has measured, and the famous heart benefit of a daily drink has largely dissolved under better methods. What is left of the case for moderate drinking is one contested observational signal and a heart-attack question that genetic studies cannot yet close.

If you don't drink, don't start for your health; if you do, less is better, and for a healthy adult the lowest-risk amount is at or near zero. Count in grams, avoid binges, keep drinks well away from bedtime, and stop completely if you are pregnant or trying to conceive, flush red when you drink, or have liver disease with significant scarring.

Why the "drink a day" benefit fell apart

For three decades, studies kept finding a J-shaped curve: light drinkers died less often than abstainers, and the heart got the credit, through higher HDL cholesterol and slightly less sticky blood. The flaw was the comparison group. Many "non-drinkers" were former drinkers who had quit because they were ill, which made abstaining look dangerous and light drinking look protective. This is the sick-quitter effect.

Correct for that and the benefit probably goes. A review of 107 cohort studies covering about 4.8 million people reproduced the J-curve in its raw numbers, but once the analysis corrected for studies that counted former drinkers as non-drinkers, and for study quality, occasional and low-volume drinkers lived no longer than lifetime non-drinkers. Risk rose significantly from 45 g a day in men, about 15% higher, and from 25 g a day in women, about 21% higher.[1] Sorting the same studies by quality made the point more bluntly: the better-designed ones found no benefit, while the weaker ones produced the familiar 16% benefit.[2] The correction is not a clean sweep: an independent review found that three of the five reviews that kept former drinkers out still showed a J-shape, though the only one at low risk of bias found risk rising with every level of drinking,[3] and a formal published rebuttal has come from authors with extensive past funding from wine, beer and spirits bodies.[4]

The one serious dissent. A congressionally mandated review by the US National Academies of Sciences, Engineering, and Medicine pooled eight studies that excluded former drinkers and found moderate drinkers had about 16% lower all-cause mortality than never-drinkers, graded moderate certainty.[5] But every study it used was observational; people who say they never drank are often unreliable about their own history and differ from drinkers socioeconomically in ways no cohort fully adjusts away; and drinking more versus less within the moderate range made no measurable difference. A parallel federal review, modelling deaths from conditions alcohol is known to cause, found no significant net benefit at any age, but never went beyond a draft.[6]

Global modelling points the same way, with one wrinkle. The Global Burden of Disease study first put the harm-minimising level at zero.[7] Its 2022 update split that by age: at or near zero for everyone aged 15–39, with injuries the largest single reason, but roughly 1 to 19 g a day for adults 40 and over, depending on the background disease rates where they live. That is under one and a half US drinks at the very top, it is a relative-risk statement about a population rather than a promise to an individual, and it is the single most-quoted line in industry communications.[8]

No trial has settled it. The one large trial designed to test about one drink a day against abstention was shut down in 2018 after only 32 people had been randomised.[9] Roughly two-thirds of its $100 million budget came from five alcohol producers, and an advisory group at the US National Institutes of Health found that contacts between agency officials and industry appeared "to intentionally bias the framing of the scientific premise" towards showing a benefit; reviewers also judged it too small and too short to assess cancer at all.[10]

Meanwhile the harm at the heavy end is not in dispute. Excessive drinking accounted for about one in eight deaths among US adults aged 20 to 64 between 2015 and 2019, and one in five among those aged 20 to 49.[11]

Cancer: the clearest harm

Alcoholic drinks have been classified as a Group 1 carcinogen — the same category as tobacco and asbestos — since 1988. There is sufficient evidence that alcohol causes seven cancers: mouth, pharynx (throat), larynx, oesophagus (the squamous-cell type), liver, bowel and female breast.[12] Stomach cancer is not on the list; a 2026 pooling of two million people found little overall link, though at 30 g a day or more risk ran about 20% higher in never-smokers and in Asian cohorts than in very light drinkers.[13]

About 4% of the world's new cancers — roughly 741,000 cases in 2020 — are attributable to alcohol, and about one in seven of those arose in moderate drinkers taking under 20 g a day.[14] A pooling of 572 studies found risk climbing steadily with dose and no protective dip at the bottom. At about one drink a day, mouth and throat cancer was already about 13% more likely and squamous-cell oesophageal cancer about 26% more likely. At one to four drinks a day both roughly doubled or more, and heavy drinkers had more than five times the risk of mouth and throat cancer.[15] Pancreatic cancer may be joining the list: a pooling of 2.5 million people found risk rising about 3% for each extra 10 g a day, compared with very light drinkers rather than non-drinkers — one analysis, not yet enough to move it onto the official list.[16]

Breast cancer is the cleanest refutation of the J-curve. The risk is a straight line that starts below one drink a day. Against not drinking at all, a review of 23 prospective studies found risk about 5% higher at half a standard drink a day, 10% higher at one, 18% at two and 22% at three, counting a drink as 10 g, with the rise seen both before and after menopause.[17] The US National Academies review likewise found no J-shape and a rise at every amount.[18] This evidence is consistent but observational. In US terms, researchers at the National Cancer Institute estimated that about 18,500 of the roughly 298,000 breast cancers expected in 2023 were attributable to alcohol, about 5,000 of them in women drinking one drink a day or less — estimates that vary several-fold between groups, so trust the direction more than the count.[19]

The main culprit is acetaldehyde, the first product the body makes from alcohol, which damages DNA and is itself a Group 1 carcinogen; alcohol also raises oestrogen, the leading explanation for the breast-cancer line. The best proof is a natural experiment. About 540 million people, mostly of East Asian descent, carry an enzyme variant that clears acetaldehyde poorly, so their face flushes when they drink. Carriers who drink have markedly higher rates of mouth, throat and oesophageal cancer than non-carriers drinking the same amount, and the gap widens with heavier drinking.[20] Flushing is a reliable stand-in for the variant, which makes it a simple warning sign.

Stopping helps, but slowly and site by site. The World Health Organization's cancer agency found sufficient evidence that cutting down or stopping lowers the risk of mouth and oesophageal cancer, limited evidence for larynx, bowel and breast, and inadequate evidence for pharynx and liver — which means not yet shown, not shown to fail.[21] In French hospital records, people with alcohol dependence who had been through rehabilitation or had documented abstinence had about 40% lower risk of alcohol-linked cancer than dependent patients without either.[22] Take the direction from that, not the size: these were dependent drinkers in a retrospective analysis of administrative records, not someone giving up a glass of wine a night.

The heart: what is left of the J-curve

The strongest evidence comes from genetics. Some people inherit gene variants, mostly in the two enzymes that break alcohol down, that make them drink more or less. Because those variants are fixed at conception, they are much less open to confounding by wealth, diet or illness, which makes them a natural experiment in lifetime drinking. This method is called Mendelian randomisation.

  • In a cohort of half a million Chinese adults, the genetic analysis in men found that genetically predicted drinking raised stroke risk with no protective threshold: about 58% more brain haemorrhage and 27% more ischaemic stroke per 280 g a week, roughly 20 US drinks. It found no clear effect on heart attack in either direction.[23] A later analysis of the same men — very few of the women drank — found genetically predicted drinking raised death from any cause by about 7% per 280 g a week and "no genetic evidence of protection from moderate drinking".[24]
  • In UK Biobank, each additional daily drink was causally linked to about 2.65 mmHg higher systolic blood pressure, roughly double the risk of haemorrhagic stroke and about 26% more atrial fibrillation; alcohol "was not protective for any disease outcome with any analysis method, dataset, or strata."[25]
  • A third genetic analysis linked a step up in genetically predicted drinking, on a scale that does not convert to drinks, to about 27% more stroke and roughly three times the risk of peripheral artery disease, the narrowing of the leg arteries.[26]
  • Among US veterans of African, European and Hispanic ancestry, mostly men, the familiar U-shaped curve for heart disease and diabetes appeared in the ordinary analysis and vanished in the genetic one — no benefit, and no harm either.[27] That is the signature of confounding, not protection.

The genetics is not uniformly damning, and the reviews say so. A review of 24 such studies found mostly null results for cardiovascular disease and judged it "not yet possible to draw conclusions" about moderate drinking.[28] A 2026 review found that only three of 20 genetic studies of heart disease had even tested for a J-shape, and judged the genetic evidence "insufficient to refute" one for heart disease specifically.[29] Heart attack, the endpoint the J-curve was built on, is the one place genetics does not show harm. Taken together, the studies rule out a large effect either way: no substantial harm, and no protection of the size the J-curve claimed.

Blood pressure rises in a straight line. Against a reference of 12 g a day, risk of high blood pressure ran about 11% higher at 24 g, 22% at 36 g and 33% at 48 g, while not drinking at all was associated with about 11% lower risk than the reference.[30] Over a single year the change is small: in Japanese adults tracked through annual checkups, giving up one to two drinks a day (10–20 g) was associated with about 1 mmHg lower systolic pressure, and giving up less than one drink a day made no measurable difference to systolic pressure.[31]

Atrial fibrillation, the common irregular heartbeat, is genuinely disputed. One pooling of over 10 million people found risk up about 6% per daily drink.[32] A larger 2026 analysis found 75% higher risk above 60 g a day but 28% lower risk below 12 g a day, compared with no intake — the same non-drinker comparison that manufactured the mortality J-curve.[33] The genetic evidence runs the other way, and the only randomised trial shows that cutting down from heavier drinking helps. In that trial, 140 regular drinkers averaging about 200 g of alcohol a week (about 14 US drinks) who already had the arrhythmia were told either to abstain or to carry on; over six months it came back in 53% of abstainers against 73% of those who kept drinking.[34] It was small, open-label and 85% male, and it tests cutting down, not taking up. Binges can also trigger episodes in healthy hearts, and heavy long-term drinking weakens and enlarges the heart muscle.

Liver and brain

The liver has no safe threshold either. Pooling 44 studies and 5.1 million people, 25 g a day — under two US drinks — was associated with about 1.8 times the risk of cirrhosis and 2.7 times the risk of dying from it, rising steeply from there.[35] If you already have fatty liver linked to weight or metabolism, drinking 140–350 g a week as a woman or 210–420 g as a man, roughly 20–60 g a day, changes the diagnosis to a combined metabolic-and-alcohol liver disease; above that range it is alcohol-related liver disease outright.[36] The two risks multiply rather than add. In screening cohorts in Spain and the US, moderate drinking in people with fatty liver raised the odds of the inflamed, progressive form by about 70%, with "moderate" meaning 10–13 drinks a week for women and 10–20 for men, below that reclassification threshold.[37] See bile and metabolism.

Grey matter is smaller the more people drink, though what that means for thinking is unknown. Alcohol use disorder clearly causes brain atrophy and cognitive decline. In 25,378 UK Biobank participants, grey matter was steadily lower with higher intake; smaller volumes showed up from 56–112 g a week, while anything below that was the study's reference group, which never-drinkers did not differ from.[38] A second analysis of 36,678 people found the changes "already apparent" at one to two UK units a day.[39] Whether a few percent of grey matter matters for how you function is not established; treat it as a reason not to feel reassured, not a diagnosis.

The old idea that moderate drinking protects against dementia has not held up. In over half a million older adults in the US and UK, the U-shaped curve appeared in the ordinary analysis and disappeared in the genetic one, where risk simply climbed with intake. People who went on to develop dementia had already been cutting down, which is what made moderate drinking look protective.[40] The 2024 Lancet Commission lists heavy drinking — over 168 g a week — as accounting for the joint-smallest share of dementia cases, with obesity, among its 14 modifiable risk factors, and states there is no clear evidence that not drinking raises dementia risk.[41] See dementia prevention.

Alcohol may also age cells faster. Genetically predicted alcohol use disorder was linked to shorter telomeres, the protective caps on chromosomes. In the observational data, drinking more than 29 UK units a week matched one to two years of age-related change, and the genetic link appeared only at heavier intakes, above about 17 UK units a week — suggestive only, since telomeres are an imperfect ageing marker.[42]

Sleep and mood: the nightcap trap

Alcohol cuts into REM (dreaming) sleep, and a 2025 pooling of 27 dosing studies found that disruption already present at about two drinks, worsening with dose. Falling asleep faster turned out to need a high dose, about five drinks, which is also the dose that does the most damage to REM later in the night. The effects on total sleep time and time awake were too uncertain to call.[43] An earlier, non-statistical review of the same research read it the other way on both points, so treat this as the better-supported reading rather than settled.[44] Snoring and sleep apnoea also worsen — see sleep-disordered breathing.

If you use a sleep tracker, a nightcap can produce a flattering deep-sleep score on a night that is measurably worse. Whether alcohol actually adds deep sleep is disputed — the claim rests on a 2013 review whose figures other sleep researchers formally challenged[45] — but either way the number goes up while the night gets worse, which shows the score is not a health metric. See sleep architecture.

Moving the last drink earlier helps but is not a clean fix. The "three hours before bed" rule is a rule of thumb. In one experiment, a moderate dose taken six hours before bedtime, when breath alcohol had fallen to zero, still cut sleep efficiency, total sleep and REM, and doubled wakefulness in the second half of the night — in ten healthy men averaging 62.[46]

Drinking and low mood are often said to feed each other, but the genetic evidence supports one direction more firmly: liability to depression raised problem drinking in women, while drinking showed no causal effect on depression in either sex.[47] The hard endpoint is suicide. Across 24 studies, alcohol use disorder carried roughly two and a half times the odds of dying by suicide in both sexes, though the studies varied so much that this is an order of magnitude rather than a precise figure.[48] If drinking has become a problem for you, treat it as urgent.

What's overrated

  • Red wine. A 150 mL glass contains roughly 0.05–0.3 mg of resveratrol.[49] Supplement trials use about 100 to 1,500 mg a day, so even the lowest of those doses would take several hundred to a few thousand glasses a day — and even those doses mostly improve surrogate markers, not health events.[50] Wine drinkers in the cohorts were wealthier, better educated, more active and ate better. No special protective effect of wine survives rigorous methods.
  • The HDL story. Alcohol does raise HDL cholesterol, but genetic variants that raise HDL do not lower heart-attack risk,[51] and a later analysis found any benefit only when HDL starts low, at or below 1.3 mmol/L.[52] HDL-raising drugs have mostly failed in trials. Raising an already-normal HDL, which is what alcohol does, is exactly the case where the evidence is null — see lipid management.
  • Diabetes protection. This is the second-strongest surviving benefit claim, and it deserves a fair hearing. Across three US cohorts of about 200,000 people, those who drank 5–6 days a week had about a quarter lower risk than those drinking 1–2 days, a comparison between drinkers that avoids the abstainer problem.[53] But frequent light drinkers differ from occasional ones in many other ways, and the genetic analyses do not reproduce the effect. Real association; cause unproven.
  • "A drink helps me relax." True in the moment, and a real reason people drink. It is not a health benefit, and the same effect is available from things that are not Group 1 carcinogens.

How much is a drink, and what the guidelines say

Count in grams, because "a drink" differs by country. A US standard drink is 14 g of pure alcohol — about 350 mL of 5% beer, 150 mL of 12% wine or 45 mL of 40% spirits.[54] A UK unit is 8 g;[55] an Australian standard drink is 10 g. Home and bar pours routinely run larger.

GuidanceWeekly amount
Lowest risk0
Canada 2023, low risk2 standard drinks or fewer, about 27 g
Canada 2023, moderate risk3–6 drinks, about 40–81 g
Canada 2023, increasingly high riskmore than 6 drinks, over 81 g
Australia (2020)10 drinks, about 100 g
UK (since 2016)14 units, about 112 g
US (2025–2030)no numeric limit

Canada's bands come from a lifetime-risk model: about two drinks a week carries a 1-in-1,000 risk of dying before 75 from an alcohol-related condition, and six a 1-in-100 risk. Those are model estimates with wide uncertainty, and the risk levels are conventions rather than anything measured about alcohol.[56] A binge is drinking that brings blood alcohol to 0.08 g/dL — roughly four drinks for women or five for men in about two hours, fewer if you are smaller or drink fast. The US National Institute on Alcohol Abuse and Alcoholism, which sets that binge line, counts heavy drinking as 15 or more drinks a week for men and 8 or more for women.[57]

Guidance and evidence have recently moved in opposite directions. The World Health Organization's European office announced in 2023 that "no level of alcohol consumption is safe";[58] the underlying statement says, more carefully, that no safe amount can be established, because no threshold for alcohol's cancer effects has been found.[59] In January 2025 the US Surgeon General named alcohol the third leading preventable cause of cancer, behind tobacco and obesity, and asked for recommended limits to be reassessed in light of the cancer evidence.[60] A year later the 2025–2030 US Dietary Guidelines instead dropped the numeric limits of two drinks a day for men and one for women, and the distinction between the sexes, leaving an unquantified "consume less" and a list of people who should avoid alcohol.[61] The liver specialists' society objected.[62] Ireland, the first country to require cancer warnings on alcohol labels, deferred them from 2026 to 2028, citing economic circumstances rather than any change in the evidence.[63] The science for lower limits strengthened; the policy moved the other way.

Who should be most careful

Pregnancy, trying to conceive or breastfeeding: abstain. No safe level is established, because fetal alcohol harm is dose-dependent and susceptibility varies widely. It is also more common than people assume: active screening of Toronto schoolchildren estimated that about 2 to 3% are affected,[64] though estimates vary enormously between studies depending on how cases are diagnosed, so read that as an order of magnitude.[65] Most cases go unrecognised: only about 0.2% of US children carry a recorded diagnosis.[66]

In recovery from alcohol use disorder, or finding it hard to control how much you drink: avoid alcohol entirely; the US guidelines list both groups.[67]

Women. The same drink produces a higher blood-alcohol level, and the risks broadly follow: mortality rises at lower intakes, blood pressure climbs more steeply at high intake,[68] and cirrhosis develops at lower exposures. Atrial fibrillation is the exception, with no clear rise in women.[69] The evidence for separate advice for women got stronger, on balance, just as the US dropped it.

A family history of breast cancer. The risk line has no floor that anyone has found, so this is one of the few places where the difference between zero and a small glass most nights is worth taking seriously.

Flushing when you drink, or East Asian heritage. Given the cancer risk described above, avoidance is the appropriate response.

Fatty liver or any liver disease. Obesity and alcohol together raise the risk of liver injury, cirrhosis, liver cancer and death from liver disease more than either alone. Ask your fibrosis stage: US liver specialists recommend complete abstinence for anyone with significant scarring, stage F2 or worse.[70]

Older adults on medication, or who have fallen. Around a quarter of older adults (estimates run from 20% to 80%) drink while taking a medicine alcohol interacts with; the serious combinations are excess sedation with benzodiazepines, opioids and antipsychotics, and low blood sugar with sulfonylurea diabetes drugs.[71] Alcohol was involved in about 618,000 US emergency visits for falls among over-65s between 2011 and 2020, with the rate rising 7.5% a year from 2011 to 2019.[72] If you take a sedating medication or have fallen in the past year, this is the strongest single argument for zero.

Anyone with atrial fibrillation, given the trial above, and anyone who binges: a binge adds acute risks to the liver, brain and heart rhythm on top of your average intake.

What cutting back buys you

Blood pressure responds fastest, if you were drinking a lot. Pooling 36 trials, cutting back did nothing measurable for people drinking about 25 g a day or less (two of that review's 12 g drinks), while people drinking six or more such drinks who halved their intake dropped 5.5 mmHg systolic and 4.0 mmHg diastolic.[73] A single alcohol-free month in moderate-to-heavy drinkers cut insulin resistance by about a quarter, systolic pressure by 6.6% and weight by 1.5%, though participants chose whether to abstain, so it is not a randomised result.[74]

Cancer risk falls slowly and only partly. Stopping was associated with about 12% lower risk of oestrogen-receptor-positive breast cancer and no detectable change for the receptor-negative kind — one estimate, consistent with the cancer agency's "limited" grading.[75] Quitting stops the risk you are still accumulating; it does not undo past exposure, and neither does cancer screening, which finds cancers earlier without removing the cause.

A month off works as a test as well as a lever. Finding it hard to stop is the signal, not the inconvenience. In a Dutch month-off challenge of 2,128 people, confidence in refusing a drink was still higher and drinking still lower a year later — uncontrolled and self-reported, so encouragement rather than proof.[76]

Further reading

  • Bagnardi V et al. Alcohol consumption and site-specific cancer risk: a comprehensive dose-response meta-analysis. Br J Cancer 2015.[77]
  • Rumgay H et al. Global burden of cancer in 2020 attributable to alcohol consumption: a population-based study. Lancet Oncol 2021.[78]
  • Millwood IY et al. Conventional and genetic evidence on alcohol and vascular disease aetiology: a prospective study of 500 000 men and women in China. Lancet 2019.[79]
  • Lankester J et al. Alcohol use and cardiometabolic risk in the UK Biobank: A Mendelian randomization study. PLoS One 2021.[80]
  • van de Luitgaarden IAT et al. Alcohol consumption in relation to cardiovascular diseases and mortality: a systematic review of Mendelian randomization studies. Eur J Epidemiol 2022.[81]
  • Carr S et al. A review of the relationship between dimensions of alcohol consumption and the burden of disease: 2026 update including Mendelian randomisation studies. Addiction 2026.[82]
  • Zhao J et al. Association Between Daily Alcohol Intake and Risk of All-Cause Mortality: A Systematic Review and Meta-analyses. JAMA Netw Open 2023.[83]
  • GBD 2020 Alcohol Collaborators. Population-level risks of alcohol consumption by amount, geography, age, sex, and year: a systematic analysis for the Global Burden of Disease Study 2020. Lancet 2022.[84]
  • NASEM. Review of Evidence on Alcohol and Health. National Academies Press 2025.[85]
  • Llamosas-Falcón L et al. How does alcohol use impact morbidity and mortality of liver cirrhosis? A systematic review and dose-response meta-analysis. Hepatol Int 2024.[86]
  • Brooks PJ et al. The alcohol flushing response: an unrecognized risk factor for esophageal cancer from alcohol consumption. PLoS Med 2009.[87]
  • US Surgeon General. Alcohol and Cancer Risk. Advisory, January 2025.[88]

— § —