Alcohol
The idea that a drink or two a day is good for you has not survived better methods: nobody has found a level of drinking below which cancer risk stops rising, and the apparent heart benefit largely dissolves once you account for the fact that many "non-drinkers" quit because they were already ill. For a healthy adult the lowest-risk amount is at or near zero.
The scientific consensus on alcohol has shifted substantially over the past decade. The historical "J-curve" — a few drinks per day appearing protective for cardiovascular events and all-cause mortality — has largely collapsed under more rigorous methods, particularly Mendelian randomization (using inherited genetic differences as a natural experiment) and meta-analyses that correct for "abstainer bias."
What survives is narrower and more contested than the headlines on either side suggest. The cancer evidence is genuinely strong, and rises across every dose anyone has measured. The cardiovascular picture is a live argument between observational data that still shows a benefit and genetic data that does not — and, as the sections below show, the genetic data is not uniformly damning either.
Where official guidance now stands
Evidence rating: Caution — documented harm signal at commonly consumed doses, which is what has driven guidance down. Note that official guidance and the evidence have recently moved in opposite directions.
- World Health Organization, January 2023: a WHO Regional Office for Europe news release announcing a WHO statement in The Lancet Public Health, under the headline "No level of alcohol consumption is safe for our health."[1] The peer-reviewed statement behind it, written by thirteen WHO staff, puts the same point more carefully: there is no threshold at which alcohol's carcinogenic effects begin, so "no safe amount of alcohol consumption for cancers and health can be established." That is an absence of an established safe level rather than a demonstration that every level is unsafe, and the statement itself concedes that studies show cardiovascular effects in middle-aged and older people.[2]
- Canada, 2023: the Canadian Centre on Substance Use and Addiction (CCSA), an arm's-length agency, published guidance intended to replace the 2011 low-risk drinking limits with a continuum of risk — 2 standard drinks per week or fewer is low risk, 3–6 is moderate, and more than 6 is "increasingly high," with every additional drink adding risk. Those cut-points come out of a lifetime-risk model: the report puts the limit associated with a 1-in-1,000 risk of premature death — which it defines as death before age 75 — from an alcohol-related condition at two standard drinks per week, and the 1-in-100 limit at six. Both are point estimates from a model whose 95% uncertainty intervals run from under one drink a week up to somewhere between six and eleven, and the 1-in-1,000 and 1-in-100 levels are borrowed conventions about how much risk people accept from a voluntary activity rather than measurements of anything about alcohol.[3]
- US Surgeon General, January 2025: an advisory on alcohol and cancer identifying alcohol as the third leading preventable cause of cancer in the US, after tobacco and obesity. It attributes nearly 100,000 cancer cases and about 20,000 cancer deaths a year to alcohol — more than the roughly 13,500 alcohol-associated traffic crash fatalities, a count that includes impaired drivers along with the passengers and other road users killed — notes that less than half of Americans know alcohol raises cancer risk, and calls for updated warning labels.[4]
The 2026 reversal
That trend broke in January 2026. The 2025–2030 Dietary Guidelines for Americans dropped the previous numeric limits — no more than two drinks a day for men and one for women — along with the distinction between the sexes, replacing them with an unquantified instruction to consume less alcohol for better overall health. What survives is a qualitative avoid-list — pregnant women, people recovering from alcohol use disorder or unable to control how much they drink, and people on interacting medications or with interacting conditions — so the numbers went, not all of the guidance.[5] The guidelines are also silent on the cancer link that the Surgeon General's advisory had highlighted a year earlier. The American Association for the Study of Liver Diseases objected to the removal of evidence-based guidance,[6] and The Lancet Gastroenterology & Hepatology called it a step backwards.[7]
The retreat was not only American. Ireland was the first country to require a label linking drinking at any level to cancer — alongside a liver-disease warning, a pregnancy warning and the alcohol content in grams — with the requirement due to take effect on 22 May 2026. In July 2025 the government deferred it to 3 September 2028, in the Minister for Health's words "in light of economic circumstances", with exposure of the drinks export trade the reported driver rather than any change in the evidence.[8]
So the honest summary is not "limits keep falling everywhere." A year before the guidelines dropped their numbers, the Surgeon General's advisory had asked federal guidance to move in the opposite direction: the second of its six recommendations was to "reassess recommended limits for alcohol consumption to account for the latest evidence on alcohol consumption and cancer risk."[9] The limits were deleted instead. So the scientific case for lower limits has strengthened while the largest national guidance in the field moved the other way — a policy divergence, not an evidentiary one.
The one surviving dissent: the National Academies review
Evidence rating: Moderate (observational only).
The strongest remaining pro-moderate-drinking signal comes from a congressionally mandated review by the US National Academies of Sciences, Engineering, and Medicine (NASEM). Pooling eight eligible studies, it found that people who drank moderately had about 16% lower all-cause mortality than people who never drank — a relative risk (RR) of 0.84, with a 95% confidence interval of 0.81 to 0.87, graded moderate certainty.[10]
That range is a confidence interval: the span within which the true effect most plausibly lies. When it sits entirely below 1.0, as here, the result is unlikely to be chance alone; when it crosses 1.0, the finding is compatible with no effect at all. Confidence intervals appear throughout this article and are read the same way each time.
Three caveats keep this from overturning the "no safe level" picture. The finding is observational; NASEM reached no conclusion at high certainty, noting that every study it included anywhere was a cohort study, with no randomized trial in the report at all. The comparison rests on a never-drinker reference group. The review did exclude studies that lumped former drinkers in with never-drinkers, so this is not the crude sick-quitter error — it says its results "are not directly comparable to past evidence that does include such abstainer bias." What remains is subtler and not fixable from outside the data: people who report never drinking are unreliable reporters of their own history, and lifetime abstainers differ from drinkers socioeconomically in ways no cohort fully adjusts away. And when the review compared higher versus lower amounts within the moderate range, the mortality difference vanished: a 4% difference well inside the range expected from chance (RR 0.96, 95% CI 0.87 to 1.06), graded low certainty. A parallel federal review published the following month — the Interagency Coordinating Committee on the Prevention of Underage Drinking's draft Scientific Findings of the Alcohol Intake and Health Study — reached the opposite conclusion, that "at no age was there a significant net benefit of alcohol consumption on all-cause mortality."[11] It got there by a different route, modelling deaths from conditions causally tied to alcohol rather than pooling all-cause-mortality cohorts, and stating that the evidence base it modelled from "did not include data from quasi-experimental (e.g., Mendelian randomization) studies or randomized controlled trials" — though it does discuss the genetic findings narratively. Both reports drew conflict-of-interest accusations from opposite directions, and neither carried the day: the ICCPUD report never went beyond its public-comment draft, and none of its findings reached the 2025–2030 guidelines. The live frontier is the tension between these observational signals and the genetic evidence below, which finds no protective threshold.
Cancer: the strongest harm signal
Evidence rating: Strong.
The International Agency for Research on Cancer (IARC) has classified alcoholic beverages as a Group 1 carcinogen since 1988, and added acetaldehyde — the first product the body makes from alcohol — in 2009. Group 1 is the same category as tobacco and asbestos. There is sufficient evidence that alcohol causes seven cancers: mouth, pharynx, larynx, oesophagus (squamous cell), liver, colorectum, and female breast.[12] Stomach cancer is pointedly not among them, and the 2026 evidence keeps it off: pooling 20 cohorts and two million people found little overall association — a 6% difference well inside the range expected from chance (hazard ratio, HR, 1.06, 95% CI 0.96 to 1.16, comparing 30 g a day or more against almost none) — with a modest excess only among people who had never smoked (about 20% higher, HR 1.20, 1.02 to 1.42) and in Asian cohorts (about 21% higher, HR 1.21, 1.02 to 1.42).[13]
About 4% of cancers worldwide — roughly 741,000 cases in 2020 — are attributable to alcohol. Most of that burden comes from heavy drinking (346,400 cases) and risky drinking (291,800), but moderate drinking, defined as under 20 g of alcohol a day, still accounted for 103,100 cases, of which 41,300 arose at 10 g a day or less.[14] In other words, roughly one alcohol-attributable cancer in seven occurs at intakes most people would not call heavy drinking. Those are new cases; the death toll has been counted more recently, and it is rising. A Global Burden of Disease analysis put alcohol-attributable cancer deaths at 343,370 in 2021 — 3.5% of all cancer deaths, and 51% more than in 2000 — even though the age-standardised death rate edged down over the same period, which points to population growth and ageing rather than to worsening per-person risk. Liver cancer accounted for the largest share (27%), then oesophageal (24%) and colorectal (16%).[15] This is disease modelling on a different metric from the incidence counts above, not an update of them.
Dose-response
A landmark dose-response meta-analysis of 572 studies and 486,538 cancer cases found clear monotonic curves — risk climbing steadily with dose, with no protective dip at the bottom.[16] At light drinking (about one drink a day, ~12.5 g of alcohol), risk is already measurably raised: about 13% higher for mouth and pharyngeal cancer (RR 1.13), 26% higher for squamous-cell oesophageal cancer (RR 1.26), and 4% higher for female breast cancer (RR 1.04).
At moderate drinking — in this analysis any intake band whose midpoint fell between 12.5 and 50 g a day, so roughly one to four drinks, a wider and higher range than the under-20 g/day Rumgay used above, because the word is not standardised across studies — the same cancers rise further: oral and pharyngeal risk roughly doubles (RR 1.83, 95% CI 1.62 to 2.07), oesophageal squamous-cell risk more than doubles (RR 2.23, 1.87 to 2.65), breast cancer runs about a quarter higher (RR 1.23, 1.19 to 1.28), and colorectal about a sixth higher (RR 1.17, 1.11 to 1.24) — though the colorectal signal rests on the case-control studies in the pool, and across the cohort studies alone it could not be distinguished from no effect (RR 1.07, 0.99 to 1.16). Among heavy drinkers, oral and pharyngeal cancer risk is more than five times that of non-drinkers (RR 5.13) and oesophageal squamous-cell cancer nearly five times (RR 4.95).
One site sitting just outside the seven has moved. Pancreatic cancer has long been graded limited or inconclusive, but pooling 30 prospective cohorts across four continents — 2.5 million people and 10,067 pancreatic cancers — found risk rising about 3% for every extra 10 g of alcohol a day (HR 1.03, 95% CI 1.02 to 1.04), a link that held in people who had never smoked (about 3% higher, 1.01 to 1.06) and did not differ by sex.[17] The signal was clearest above 15 g a day in women and 30 g a day in men, appeared in European, Australian and North American cohorts but not Asian ones, and showed up for beer and spirits but not wine — differences the authors attribute to drinking habits rather than to the drink. Note the comparison group: these are increments measured against very light drinkers (0.1 to under 5 g a day), not against non-drinkers, and one pooled analysis does not move pancreas onto IARC's sufficient-evidence list.
Breast cancer specifically
Evidence rating: Moderate (consistent, but observational).
The breast-cancer dose-response is the cleanest refutation of the J-curve anywhere in this literature: it is a straight line that starts below one drink a day. NASEM's meta-analysis put the increase at about 5% for every 10–14 g of alcohol per day (RR 1.05, 95% CI 1.04 to 1.06, moderate certainty) and concluded there is no evidence of a J-shaped association — the relationship appears linear, with increased risk at every amount. Comparing moderate drinkers with never-drinkers, the same review found about 10% higher risk (RR 1.10, 95% CI 1.02 to 1.19, moderate certainty).[18]
A 2024 systematic review and meta-analysis of 23 prospective-cohort publications, led from the Centre for Addiction and Mental Health with IARC and WHO Europe staff among the authors, measures the whole curve rather than one increment of it. Against not drinking at all — a different comparison from NASEM's per-drink increment above, so the two 5% figures are not the same quantity — breast-cancer risk ran about 5% higher at half a standard drink a day, 10% higher at one, 18% higher at two and 22% higher at three (RR 1.05, 95% CI 1.04 to 1.06; 1.10, 1.08 to 1.12; 1.18, 1.15 to 1.21; 1.22, 1.19 to 1.25; one standard drink here is 10 g of ethanol). Even below one standard drink a day the increase was measurable, about 4% higher (RR 1.04, 1.01 to 1.07). Each additional daily drink raised both premenopausal (about 3% higher, RR 1.03, 1.01 to 1.06) and postmenopausal (about 10% higher, RR 1.10, 1.08 to 1.12) breast cancer, which closes the gap NASEM left open — its premenopausal estimate could not be distinguished from no effect (RR 1.03, 0.98 to 1.08).[19]
For scale in absolute terms: combining US alcohol-use data through 2019 with the projected number of American breast-cancer diagnoses for 2023, researchers at the National Cancer Institute estimated that about 18,500 of 297,790 expected cases that year would be attributable to alcohol, and that roughly 5,000 of those would arise among women drinking one drink a day or less.[20] Estimates of this burden vary several-fold between research groups depending on how intake is measured, so treat the exact count as approximate and the direction as settled.
Mechanisms
- Acetaldehyde — the first product of alcohol metabolism; it binds to DNA and interferes with repair, and is itself a Group 1 carcinogen.
- Oxidative damage — an alternative alcohol-processing pathway in the liver generates reactive molecules that damage DNA.
- Hormonal effects — alcohol raises circulating oestrogens, the leading explanation for the breast-cancer dose-response.
- Disrupted folate metabolism — impairing the chemical tagging that regulates which genes are switched on.
- Impaired absorption of protective nutrients (vitamins A, D and E, folate, carotenoids).
The natural experiment that closes the causal loop. Roughly 540 million people — about 8% of the world's population, mostly of East Asian descent — carry a variant of the enzyme that clears acetaldehyde; it works poorly, so acetaldehyde accumulates and the face flushes. Carriers who drink have markedly higher rates of cancers of the mouth, throat and oesophagus than non-carriers drinking the same amount, and the gap widens sharply with heavier drinking. Because the variant is inherited at random and affects nothing except acetaldehyde clearance, this is about as close to a controlled experiment as the field gets — and the flushing reaction is a reliable stand-in for the genotype, making it a simple and actionable warning sign.[21]
What happens if you stop
Evidence rating: Moderate (sufficient for two cancer sites, limited or inadequate for the rest).
Quitting helps, but the evidence is site-specific and thinner than the evidence for harm. An IARC Working Group reviewed exactly this question in Lyon in 2023 and reported in 2024 that there is sufficient evidence that reducing or stopping drinking lowers the risk of mouth cancer and oesophageal cancer; limited evidence for larynx, colorectum and breast; and inadequate evidence for pharynx and liver. Inadequate does not mean quitting was shown to fail — for the pharynx, long-term cessation was not associated with lower risk in the most influential pooled analysis and results disagreed between subsites; for the liver, reverse causation and competing causes of death could not be ruled out. The same review found sufficient mechanistic evidence that stopping reverses alcohol-driven carcinogenesis: local acetaldehyde exposure in the mouth and upper gut falls immediately, acetaldehyde-driven DNA damage decreases, and the increased gut permeability of chronic heavy drinking recovers.[22] A short summary of the volume was published in [23].
The largest empirical test so far is a French national hospital cohort of 24 million adults discharged between 2018 and 2021, among whom about 865,000 carried a hospital-recorded alcohol dependence. Patients with a record of rehabilitation treatment or a documented history of abstinence had roughly 40% lower risk of an alcohol-associated cancer than dependent patients with neither (adjusted hazard ratio 0.58, 99.89% CI 0.56 to 0.60 in men; 0.62, 0.57 to 0.66 in women), and the reduction held at every alcohol-associated cancer site in both sexes.[24] Take the direction from this and not the number: these are people with alcohol dependence rather than moderate drinkers, and it is a retrospective analysis of administrative hospital records, so the size of the drop will not transfer to someone cutting back from a glass of wine a night. Only two in five dependent patients had any record of rehabilitation or abstinence at all.
Cardiovascular: the J-curve has mostly collapsed
Evidence rating: Strong (for harm to blood pressure and stroke); Moderate (for the absence of coronary benefit).
For three decades observational studies showed a J-shape: light drinkers appeared healthier than abstainers. The protective hypothesis rested on alcohol raising HDL ("good") cholesterol and mildly reducing blood clotting.
What the genetic evidence shows
Mendelian randomization exploits the fact that people inherit, at random, gene variants that strongly influence how much they drink — most importantly in the two enzymes that metabolise alcohol. 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 for lifetime alcohol exposure.
- The China Kadoorie Biobank enrolled 512,715 adults and genotyped 161,498 men. Genetically predicted alcohol intake had a continuously positive association with stroke — about 58% higher risk of brain haemorrhage and 27% higher risk of ischaemic stroke per 280 g a week, about 20 US standard drinks (RR 1.58, 95% CI 1.36–1.84; RR 1.27, 95% CI 1.13–1.43) — and raised blood pressure, with no protective threshold anywhere. Notably, it found no significant effect on heart attack (RR 0.96, 95% CI 0.78–1.18), a result compatible with no effect at all; the authors concluded alcohol "appears in this one study to have little net effect on the risk of myocardial infarction."[25]
- Two later analyses of the same Chinese cohort carried the picture well beyond stroke, and both are in men — only about 2% of the cohort's women drank at all. Genetically predicted drinking raised death from any cause by about 7% per 280 g a week and death from cardiovascular disease by 15% (HR 1.07, 95% CI 1.05–1.10; HR 1.15, 95% CI 1.10–1.19), driven by stroke (HR 1.18, 95% CI 1.12–1.24) rather than ischaemic heart disease (HR 1.06, 95% CI 0.99–1.14); the authors found "no genetic evidence of protection from moderate drinking for all-cause and cause-specific mortality, including CVD."[26] A companion scan of 207 diseases in the same men put genetically predicted intake at 38% higher stroke risk and more than double the risk of liver cirrhosis per 280 g a week (HR 1.38, 95% CI 1.27–1.49; HR 2.30, 95% CI 1.58–3.35), while ischaemic heart disease stayed flat across 8,408 cases — roughly three times as many heart-disease events as the 2019 analysis had (HR 1.04, 95% CI 0.94–1.14).[27]
- A 94-variant analysis found roughly 27% higher stroke risk and a three-fold higher risk of peripheral artery disease among people whose genes predispose them to drink substantially more (odds ratios, OR, of 1.27, 95% CI 1.12–1.45, and 3.05, 95% CI 1.92–4.85, per standard-deviation increase in predicted drinking).[28]
- In UK Biobank, each additional daily drink was causally associated with +2.65 mmHg systolic blood pressure, roughly double the risk of haemorrhagic stroke (OR 2.25) and about 26% higher atrial fibrillation risk (OR 1.26). The authors' summary is blunt: alcohol "was not protective for any disease outcome with any analysis method, dataset, or strata."[29]
- A Million Veteran Program analysis of 33,053 coronary heart disease cases and 28,278 diabetes cases — across African American, European American and Hispanic American veterans, the only large genetic study of alcohol to cover all three, though the cohort is predominantly male — reproduced the U-shaped observational curve and then found no causal association in either direction in the genetic analysis: no benefit at moderate intake, and no harm either, with or without adjustment for blood pressure and smoking.[30] That is the signature of confounding, not of protection.
A systematic review of 24 Mendelian randomization studies found null associations between genetically predicted alcohol and cardiovascular disease in 67% of studies and diabetes in 75%, with the single study of all-cause mortality reporting harm. Its authors were careful, though: given how much the methodological quality varied, they judged it "not yet possible to draw conclusions on the causal role of moderate alcohol consumption on cardiometabolic health." (Moderate)[31] That is weaker than "genetics proves harm" — but it is fatal to "genetics proves benefit," which is what the J-curve needed.
A 2026 systematic review from a World Health Organization collaborating centre sharpened the same caution. Reviewing 20 Mendelian randomization studies of ischaemic heart disease specifically, it found they "largely suggested null or harmful relationships" — but only three had tested for a non-linear effect, and it judged the genetic evidence so far "insufficient to refute a J-shaped relationship" at that one endpoint, recommending that heart-disease risk estimates still lean on cohort data. In the cohort literature the same review covers, J-shapes survive for ischaemic heart disease, ischaemic stroke and type 2 diabetes — generally only among people who avoid heavy drinking sessions — while risk rises steadily with dose for cardiovascular disease overall, cancers, and infectious and digestive disease. (Moderate)[32] That is not a rehabilitation of moderate drinking. It is a statement about what genetic studies are currently powered to rule out.
Specific outcomes
- Hypertension — an almost perfectly linear dose-response with no safe floor. Against a 12 g/day reference, risk runs about 11% higher at 24 g/day, 22% at 36 g/day and 33% at 48 g/day; drinking nothing at all is associated with 11% lower risk than the 12 g/day reference (RR 0.89, 95% CI 0.84–0.94). (Strong)[33] Within-person data give a more modest short-run figure than the genetic estimate above. Among 58,943 Japanese adults tracked across 359,717 annual checkups, giving up one to two drinks a day — 10 to 20 g of alcohol — was associated with about 1 mmHg lower systolic pressure (−1.03 mmHg, 95% CI −1.70 to −0.35 in men; −0.78 mmHg, 95% CI −1.53 to −0.04 in women), while giving up less than one drink a day produced no measurable systolic change in either sex; starting to drink raised pressure by the same dose-dependent amounts, and the type of drink made no difference. (Moderate)[34] The genetic figure is for a lifetime of exposure; this one is for a year of change, which is why they differ so much.
- Atrial fibrillation — the dose-response here is genuinely disputed. Pooling 13 studies and over 10 million participants, each additional daily drink raised risk about 6% (RR 1.06, 95% CI 1.03–1.08), linearly in men (about 8% per drink) but J-shaped in women, whose pooled estimate did not reach significance (RR 1.05, 95% CI 0.96–1.14). (Moderate)[35] A larger 2026 network meta-analysis of 26 studies and nearly 15 million people found a threshold instead: compared with people reporting no intake, risk was 75% higher above 60 g a day (RR 1.75, 95% CI 1.25–2.44) and 28% lower below 12 g a day (RR 0.72, 95% CI 0.63–0.83), and its authors suggested that a more liberal approach at low intakes "could be reasonable," while adding that further prospective studies are needed before any of it is called causal. (Moderate)[36] Note what that low-dose protection rests on: pooled observational cohorts with a no-intake reference group, the same construction that manufactured the mortality J-curve. The genetic evidence runs the other way — 26% higher risk per additional daily drink in UK Biobank (OR 1.26, 95% CI 1.07–1.48)[37] and, on a separate scale that does not convert to drinks, 17% higher per standard-deviation increase in genetically predicted drinking, borderline with an interval touching no effect (OR 1.17, 95% CI 1.00–1.37)[38] — as does a randomised trial described below, in which giving up alcohol reduced recurrences in drinkers who already had the arrhythmia. Binge drinking can also trigger acute "holiday heart" episodes in otherwise healthy hearts.
- Stroke — rises with intake, and the genetic evidence supports causation, most strongly for bleeding strokes.
- Cardiomyopathy — heavy chronic drinking causes dilated cardiomyopathy, a weakening and enlargement of the heart muscle.
The residual protective signal in some cohorts appears to be confounded by the socioeconomic and lifestyle advantages that travel with moderate drinking in Western populations. The one endpoint where the genetic evidence still does not show harm is heart attack — which is exactly the endpoint the J-curve was built on, and exactly why this argument has not ended. It is worth being precise about what "does not show harm" means. The Chinese cohort's later analyses have not reproduced the 0.96. On roughly three times as many events, genetically predicted drinking left ischaemic heart disease incidence flat — a 4% difference well inside the range expected from chance (HR 1.04, 95% CI 0.94–1.14, 8,408 cases);[39] and on the separate endpoint of dying from it, the estimate sat just above no effect (HR 1.06, 95% CI 0.99–1.14, 2,363 deaths).[40] Those are three different endpoints, not one rising series; the 94-variant analysis put coronary artery disease 16% higher on its own, non-convertible scale, again with the interval touching no effect (OR 1.16, 95% CI 1.00–1.36);[41] and the largest US genetic analysis, spanning African American, European American and Hispanic American veterans, found nothing in either direction, as noted above. Together they rule out any large effect either way — no substantial harm, and no protection of the size the J-curve claimed. And almost none of these studies tested for a J-shape at all, which is the specific thing they would have to disprove: of 20 genetic studies of heart disease reviewed in 2026, three did.[42] Genetics has removed the evidence for cardioprotection here without yet being able to close the question.
All-cause mortality and global burden
Evidence rating: Strong (harm above roughly 45 g/day); Moderate (absence of benefit at low intakes).
Bias-corrected meta-analysis. A systematic review of 107 cohort studies — 4,838,825 people and 425,564 deaths — reproduced the classic J-curve in naïve estimates. But once it corrected for the sick-quitter problem (former drinkers, many of whom quit because they were ill, landing in the abstainer reference group) and for study quality, the apparent low-dose protection disappeared. Occasional drinkers showed no meaningful mortality difference from lifetime non-drinkers — a 4% gap well within the range expected from chance (RR 0.96, 95% CI 0.86–1.06) — and low-volume drinkers (1.3–24 g/day) likewise fell short of significance. Risk turned significantly upward at 45–64 g/day (about 19% higher) and at 65 g/day or more (about 35% higher). The threshold at which risk became significant was lower for women than men: mortality rose significantly in women drinking 25 g a day or more — about 21% higher than female lifetime non-drinkers in the 25–44 g/day band (RR 1.21, 95% CI 1.08–1.36) — against 45 g a day or more in men, about 15% higher than male lifetime non-drinkers (RR 1.15, 95% CI 1.03–1.28). Pooled across every level of intake, female drinkers as a whole ran about 22% above female lifetime non-drinkers (RR 1.22, 95% CI 1.02–1.46). (Strong)[43]
That correction has since been pressed three ways. A follow-up on the same 107 studies made the point more directly by sorting the studies on quality rather than adjusting for it: those that enrolled younger cohorts, followed them past 55, and kept former and occasional drinkers out of the abstainer group found low-volume drinkers dying at essentially the same rate as abstainers — a 2% difference well inside the range expected from chance (RR 0.98, 95% CI 0.87–1.11) — while the studies failing those criteria produced the familiar 16% benefit (RR 0.84, 95% CI 0.79–0.89).[44] An independent Australian group reached the same place from outside that research team: of 25 systematic reviews of alcohol and all-cause mortality, only five kept former drinkers out of the reference group, and among the 20 that did not, 17 reported lower mortality at low intakes — though three of those five retained reviews still found a J-shape, so this is the best-conducted review finding no benefit rather than every bias-corrected review doing so.[45] And the correction has been argued over in print rather than quietly accepted: a defence of moderate drinking published as a formal comment on the review disputes the bias adjustments, and its authors disclose extensive past funding and consultancies from wine, beer and spirits bodies.[46]
Global burden modelling. The Global Burden of Disease study found that "the level of alcohol consumption that minimised harm across health outcomes was zero" standard drinks per week. Alcohol was the seventh leading risk factor for death worldwide in 2016 and the leading risk factor for people aged 15–49, where it accounted for 12.2% of male and 3.8% of female deaths. (Moderate — modelled estimate, now superseded)[47]
The same collaboration replaced that single global figure in 2022 with age- and region-specific thresholds. For people aged 15–39 the harm-minimising level is at or near zero in every region — the range across regions runs from 0 to 0.6 standard drinks a day, so 6 g at the very top — and injuries carry the largest single share of the burden — 66% of alcohol-related healthy years lost among young men and 48% among young women. For adults 40 and over the burden-weighted risk curve was J-shaped in every region, with a harm-minimising level of roughly 1 to 19 g of alcohol a day depending on background disease rates. The paper reports that as 0.114 to 1.87 standard drinks, and its standard drink is 10 g rather than the 14 g of a US one, so the top of the range is under one and a half US drinks. That "older adults may benefit" element is a relative-risk statement inside an age band, says nothing about an individual's absolute benefit, and is the single most-quoted line in industry communications. (Moderate — modelled estimate; supersedes the 2018 "zero" figure)[48]
The US toll. Excessive drinking accounted for an estimated 12.9% of all deaths among US adults aged 20 to 64 between 2015 and 2019 — about 89,700 deaths a year, or one in eight — rising to 20.3%, one in five, among adults aged 20 to 49.[49] Counting all ages, the toll has since climbed steeply: an average of 178,307 deaths a year in 2020–2021 against 137,927 in 2016–2017, a 29.3% rise, taking the age-standardised rate from 38.1 to 47.6 per 100,000. The female count rose faster than the male one — up 34.7% against 26.8%.[50]
One trial of cutting down, none of taking up
One randomised trial of alcohol with a heart endpoint has been done, but it asked a different question. Australian researchers randomised 140 regular drinkers who already had paroxysmal or persistent atrial fibrillation — averaging about 200 g of alcohol a week, some 17 drinks at that trial's 12 g standard drink — either to abstain or to carry on. Over six months the abstinence group cut intake by 88% and the control group by 20%, and the arrhythmia came back in 53% of abstainers against 73% of controls (hazard ratio 0.55, 95% CI 0.36–0.84). Abstainers also spent less of the six months in the arrhythmia than controls did: a median 0.5% of the time against 1.2%. (Moderate)[51] That is a trial of cutting down among people who already drink and already have an arrhythmia — small, open-label, six months long and 85% male. It is not a trial of taking up drinking, which is what the J-curve claim actually needs.
The one large randomized trial designed to settle the question — the Moderate Alcohol and Cardiovascular Health Trial, testing about one drink a day against abstention — collapsed in 2018. Of 390 people who consented, only 32 had been randomized before the National Institutes of Health withdrew funding.[52][53] The trial had aimed to enrol 7,800 people; enrolment opened on 5 February 2018 and was suspended on 10 May with 105 signed up. The reason NIH gave names the gap the next paragraph describes — doubts about "whether the study would effectively address other significant consequences of moderate alcohol intake, such as cancer."
The reason it ended matters as much as the fact. Roughly two-thirds of the trial's $100 million budget had been raised from five alcohol producers — Anheuser-Busch InBev, Carlsberg, Diageo, Heineken and Pernod Ricard — and an NIH advisory working group found that the collaborations between agency officials and industry actors appeared "to intentionally bias the framing of the scientific premise in the direction of demonstrating a beneficial health effect of moderate alcohol consumption," while peer reviewers of the protocol separately judged that the trial had too few patients and too little follow-up to assess cancer endpoints at all.[54] This is why genetic and bias-corrected observational evidence, rather than a trial, carry the weight here — and why conflict-of-interest flags recur throughout this field, on the advocacy side as well as the industry side.
Liver
Evidence rating: Strong (heavy drinking); Moderate (interaction with fatty liver disease).
Alcohol produces a direct dose-response for fatty liver, alcoholic hepatitis and cirrhosis, and is a dominant cause of cirrhosis worldwide alongside the hepatitis viruses. Pooling 44 studies and 5.1 million people, drinking 25 g a day — under two US standard drinks — was associated with roughly 1.8 times the risk of developing cirrhosis and 2.7 times the risk of dying from it, rising to about 8-fold and 16-fold at 100 g a day. The curve has no threshold below which risk stops rising, and mortality accelerates faster than illness does.[55]
The clinically useful numbers now come from the 2023 renaming of fatty liver disease. Metabolic dysfunction–associated steatotic liver disease (MASLD, formerly non-alcoholic fatty liver disease) plus drinking of 140–350 g per week for women or 210–420 g per week for men — roughly 20–50 and 30–60 g a day — is reclassified as MetALD, a combined metabolic-and-alcohol disease; above that range it becomes alcohol-related liver disease outright.[56] The practical reading: if you already have fatty liver, amounts well inside what most guidelines call moderate are enough to change your diagnosis. And the two risks do not simply add up. In population screening cohorts in Spain and the US measured with liver elastography, moderate drinking among people who already had MASLD raised the odds of steatohepatitis at risk of progression by about 70% (odds ratio 1.69, 95% CI 1.06 to 2.71) — in a US cohort where 13% of people with MASLD had at-risk steatohepatitis and 17% had significant fibrosis to begin with. Significant fibrosis became steadily more common as drinks per week and the number of cardiometabolic risk factors rose together, an interaction the authors call supra-additive, concluding that "there are no safe limits of daily alcohol intake in patients with unhealthy metabolic status and MASLD".[57] Note the doses involved: "moderate" there is 10–13 drinks a week for women and 10–20 for men — below the MetALD threshold, and inside what the sentence above describes as a naming change. See bile and metabolism for the wider liver picture.
Brain
Evidence rating: Strong (heavy drinking); Moderate (structural changes at moderate intake); Weak (clinical significance of those changes).
- Alcohol use disorder clearly causes brain atrophy and cognitive decline.
- Even moderate drinking tracks with reduced grey-matter volume. In 25,378 UK Biobank participants, alcohol intake was negatively and linearly associated with global grey matter across the range the study measured — which is not the same as there being no floor. The lowest exposure category, under 7 UK units (56 g) a week, is the study's own reference group, so nothing below that was tested; the lowest band showing smaller grey-matter volumes was 7–14 units (56–112 g) a week. To scale it, the authors calculated that one year of aging cost the same grey matter as a difference of about 13 UK units (roughly 100 g) a week. Never-drinkers were indistinguishable from the lightest drinkers while former drinkers had less grey matter than the lightest drinkers, so the sick-quitter pattern shows up in the imaging data too.[58] A separate analysis of 36,678 people found the negative associations with both grey-matter volume and white-matter structure "already apparent in individuals consuming an average of only one to two daily alcohol units," strengthening with intake.[59]
- The 2024 Lancet standing Commission on dementia lists excessive alcohol — which it defines as more than 21 UK units (168 g) a week — among its 14 modifiable risk factors, though it is the weakest of them: a population attributable fraction of 1%, tied with obesity for the smallest, resting on about 20% higher risk whose interval only just clears no effect (relative risk 1.2, 95% CI 1.0 to 1.5). The Commission is careful in the other direction too, stating that "a lack of clear evidence exists that not drinking alcohol increases the risk of dementia".[60] See dementia prevention.
- The older "moderate alcohol protects against dementia" finding has not held up. Among 559,559 adults aged 56–72 in the US Million Veteran Program and UK Biobank, followed to 14,540 dementia diagnoses, the familiar U-shape appeared in the ordinary observational analysis and disappeared in the genetic one: risk climbed with intake, with no protective range anywhere. A standard-deviation increase in genetically predicted log drinks per week raised dementia odds about 15% (OR 1.15, 95% CI 1.03 to 1.27), and a doubling of alcohol-use-disorder prevalence raised them about 16% (OR 1.16, 95% CI 1.03 to 1.30). The mechanism behind the illusion is visible in the same data: people who went on to develop dementia had already been cutting their drinking, and when intake was measured closest to diagnosis, moderate drinking looked protective again.[61]
What these imaging studies do not establish is how much a few percent of grey-matter volume matters for how you think and function. Treat the structural signal as a reason not to be reassured, not as a diagnosis.
Biological aging
Evidence rating: Moderate (telomeres); Weak (gut–liver axis).
- Telomeres. (Moderate) Alcohol appears to accelerate one measurable marker of cellular aging. A UK Biobank study of 245,354 people paired with Mendelian randomization found genetically predicted alcohol use disorder causally associated with shorter telomeres — the protective caps on chromosome ends. To scale it: in the observational data, drinking more than 29 UK units a week — a UK unit is 8 g, so roughly 17 US standard drinks — versus fewer than 6 was equivalent to 1–2 years of age-related telomere change, while the genetic estimate came out larger, at about 3 years per standard-deviation increase in predicted drinking. The genetic association appeared only in current drinkers and not in never- or former-drinkers, which is the test the authors used to confirm the gene variants act on telomeres through alcohol rather than by some other route. Two honest caveats: the non-linear analysis suggested a threshold rather than a straight line — the genetic association reached significance only in the top two intake quintiles, above roughly 17 UK units (136 g) a week, and the top quintile's estimate did not survive excluding related participants — and telomere length is an imperfect aging marker, so treat this as suggestive.[62]
- Gut–liver axis. (Weak) Chronic drinking disrupts the gut microbiome and increases intestinal permeability, letting bacterial fragments cross into the blood draining the gut, where they activate liver immune cells and drive inflammation and scarring. A plausible mechanistic bridge to the low-grade systemic inflammation associated with biological aging, but not demonstrated as a pathway in humans at moderate intakes.
Sleep
Evidence rating: Moderate — consistent laboratory findings, but on sleep-architecture measures rather than hard endpoints.
Alcohol delays the onset of the first REM (rapid eye movement, or dreaming) period and shortens REM across the night, and a 2025 systematic review and meta-analysis of 27 dosing studies in healthy adults found the REM disruption already present at a low dose — 0.50 g of alcohol per kilogram of body weight or less, roughly two standard drinks — worsening steadily as the dose rises. Falling asleep faster turns out not to be a general effect: sleep-onset latency shortened only at a high dose, 0.85 g/kg or more, about five standard drinks, which is also the dose that does the most damage to REM later in the night. The pooled effect on total sleep time, sleep efficiency and time spent awake after falling asleep was too uncertain to call in either direction.[63] An earlier qualitative review of the same literature reported the opposite pattern on both counts — faster sleep onset at every dose, and no clear REM trend at low ones — and also described the second half of the night becoming fragmented, which the newer pooling could not resolve.[64] Snoring and sleep apnoea worsen — see sleep-disordered breathing.
The trap for anyone using a sleep tracker: a nightcap can produce a flattering deep-sleep score while degrading the night in every way that matters. Treat the size of that effect as unsettled. The claim that alcohol increases slow-wave sleep in the first half of the night across doses, ages and sexes comes from a 2013 qualitative review;[65] eight sleep researchers formally challenged that review's slow-wave statistics in the same journal,[66] and the original authors disputed the challenge in reply.[67] The 2025 meta-analysis reports only that a high dose shortens the time taken to reach deep sleep.[68] The tracker point survives that dispute either way: whether alcohol adds deep sleep or merely brings it forward, the number goes up on a night that is measurably worse — which is the demonstration that the deep-sleep number is not a health metric — see sleep architecture.
Practical: moving the last drink earlier helps, but do not treat any gap as a clean fix. The common advice to stop three hours before bed is a rule of thumb rather than a tested interval. In the one controlled experiment that used a long gap, a moderate dose — 0.55 g per kilogram of body weight, taken six hours before bedtime, by which point breath alcohol had fallen to zero in every subject — still cut sleep efficiency, total sleep time and REM sleep, and doubled wakefulness in the second half of the night, in ten healthy men averaging 62 years old.[69] Earlier is better than later; none at all is better than either, on nights when sleep matters.
Mental health
Evidence rating: Moderate.
The relationship between alcohol and both depression and anxiety is usually described as bidirectional and self-reinforcing: distress prompts drinking, and drinking worsens distress through disrupted sleep, rebound anxiety the following day, and stress-hormone effects. The genetic evidence supports one of those arrows more firmly than the other. Sex-stratified Mendelian randomization found that liability to major depression raised the risk of problematic alcohol use in women, while no measure of drinking showed a causal effect on depression in either sex — and adjusting for socioeconomic position weakened several of the remaining links.[70] So depression-drives-drinking is the better-supported direction; drinking-drives-distress rests on short-term pharmacology rather than on genetics. Alcohol is pharmacologically a depressant, and even moderate use degrades sleep quality enough to feed the loop.
The hardest endpoint here is suicide. Across 24 studies covering 37.9 million people, alcohol use disorder was associated with roughly two and a half times the odds of dying by suicide in the longitudinal studies, and similarly in both sexes — odds ratio 2.68 (95% CI 1.86 to 3.87) pooled from 14 male estimates, and 2.39 (1.50 to 3.81) from 11 female estimates. The apparent sex gap in older work turns out to be an artefact of study design: among the cross-sectional studies the same comparison gave roughly 3.7 times the odds among men (odds ratio 3.67, 95% CI 2.47 to 5.48) and nearly 7 times among women (6.88). Differences between the pooled studies were extreme, so read these as an order of magnitude rather than precise numbers.[71] If drinking has become something you would describe as a problem, this is the reason to treat it as urgent rather than gradual.
Sex differences
Evidence rating: Moderate.
Women reach a higher blood-alcohol concentration than men from the same drink — the standard explanation is lower total body water and less first-pass breakdown of alcohol in the stomach — and the outcome data broadly track that. Women crossed into significantly raised all-cause mortality at 25 g a day against 45 g a day in men, as above,[72] the hypertension dose-response was steeper in women at high intake,[73] cirrhosis develops at lower exposures, and breast cancer is a large alcohol-attributable burden with no threshold that applies almost entirely to women.
The pattern is not universal, which is worth stating plainly: the atrial-fibrillation dose-response was linear in men but J-shaped and non-significant in women.[74] The 2025–2030 US Dietary Guidelines removed the sex distinction from national advice; the evidence for keeping it, on balance, got stronger rather than weaker.
Pregnancy
Evidence rating: Strong (harm); Caution (no threshold established).
No safe level is established. Fetal alcohol spectrum disorders are dose-dependent, and individual susceptibility varies widely enough that no threshold can be identified as safe. These conditions are also more common than most people assume: an active-screening study of 2,555 children aged 7 to 9 in the Greater Toronto Area estimated a prevalence of 1.8% (18.1 per 1,000, 95% CI 10.8 to 26.3 per 1,000), rising to 2.9% under less conservative assumptions — a band its authors summarised as 2 to 3%, and higher than Down syndrome, spina bifida, trisomy 18 or autism spectrum disorder in Canada.[75] Modelling from pooled Australian prenatal-exposure data reaches similar territory by a different route, at 3.64% (95% CI 2.91 to 4.41).[76] Treat any single figure as an order-of-magnitude claim, though: across 41 studies in high-income countries the estimates span 5.8 to 170,000 per million, almost entirely because of which diagnostic criteria and sampling methods were used.[77] What is not in doubt is the diagnostic gap: only 0.21% of US children carry a recorded diagnosis (95% CI 0.13 to 0.29), roughly a tenth of what screening finds, and seven times higher among children on public insurance than private (0.49% versus 0.07%).[78] The recommendation across every major authority is complete avoidance during pregnancy, and ideally from the point of trying to conceive.
What's overrated
Evidence rating: Moderate — these are claims the evidence has moved away from, not claims never tested.
- Red wine and resveratrol. A 150 mL (5 oz) glass of red wine contains roughly 0.05–0.3 mg of resveratrol: typical red wines measure 0.36–1.97 mg/L, and a French adult's entire wine consumption works out at about 0.2 mg a day, some 5,000 times less than the 1 g a day proposed for therapeutic effect.[79] Supplement trials run from about 100 to 1,500 mg a day, so reaching even the bottom of a studied dose would take several hundred to a few thousand glasses — in one day. And at those doses the pooled evidence still tops out at surrogate markers rather than events: across 45 meta-analyses and 129 associations, only four reached high-certainty evidence, and doses of 500 mg a day or more performed worse rather than better.[80] The "wine drinkers live longer" cohorts are overwhelmingly explained by confounding — wine drinkers in those studies were wealthier, better educated, more active and ate better. There is no special protective effect of wine over other alcohol that survives rigorous methods.
- The HDL story. Alcohol does causally raise HDL cholesterol. But a Mendelian randomization study found that genetic variants which raise HDL do not lower heart-attack risk, concluding that these data "challenge the concept that raising of plasma HDL cholesterol will uniformly translate into reductions in risk of myocardial infarction".[81] Later work refines that rather than reversing it: in 348,636 UK Biobank participants a causal inverse association survived only in the stratum with HDL cholesterol at or below 1.3 mmol/L (50 mg/dL) (odds ratio 0.86 per unit increase, 95% CI 0.79 to 0.94) and was absent in every stratum above it.[82] Raising HDL from an already-normal level — which is what alcohol does — is precisely the case where the evidence is null. HDL-raising drugs have mostly failed in trials, and the one that did cut events makes the point rather than undermining it: in 30,449 patients, anacetrapib doubled HDL cholesterol and reduced major coronary events by about 9% (rate ratio 0.91, 95% CI 0.85 to 0.97), but it also lowered non-HDL cholesterol by 18%, which is where the benefit is attributed.[83] That removes the mechanism the J-curve was resting on — see lipid management.
- Type 2 diabetes protection. This is the second-strongest surviving benefit claim and it is real in observational data: across three US cohorts totalling 200,969 people and 20,551 diabetes cases over three decades, light-to-moderate drinking was associated with lower risk, driven mainly by drinking frequency rather than amount — those drinking 5–6 days a week had roughly a quarter lower risk than those drinking 1–2 days (hazard ratios, HR, of 0.73, 95% CI 0.65 to 0.83 and 0.73, 0.62 to 0.86 in the two nurses' cohorts, and 0.76, 0.67 to 0.86 in the male health professionals).[84] This one deserves a fair hearing. The frequency comparison sets drinkers against drinkers, so no abstainer enters it, and former regular drinkers were excluded from the non-drinking reference group used in the total-intake analyses — the standard control for abstainer bias. What it cannot rule out is that people who drink a little most evenings differ from occasional drinkers in a hundred other ways, and the genetic analyses do not reproduce the effect. (Moderate for the association; Weak for causation.)
- "A drink helps me relax." True acutely, and it is a real reason people drink. It is also not a health benefit, and the same effect is available from things that do not carry a Group 1 carcinogen classification.
Quantifying "a drink"
Standard drinks differ by country, so grams are the only stable unit. A US standard drink is 14 g of pure alcohol, or 0.6 fluid ounces — about 350 mL (12 oz) of 5% beer, 150 mL (5 oz) of 12% wine, or 45 mL (1.5 oz) of 40% spirits.[85] A UK unit is 10 mL, or 8 g;[86] a Canadian standard drink is 17.05 mL, or 13.45 g;[87] an Australian one is 12.5 mL, or 10 g. Most of these are officially defined by volume, with grams as the conversion — which is why the same 150 mL glass of 12% wine counts as one drink in Washington and about 1.4 in Sydney. When comparing studies, always convert to grams.
Where the thresholds sit now:
| Level | Amount |
|---|---|
| Lowest risk | 0 |
| Low risk (Canada 2023) | 2 standard drinks/week or fewer ≈ 27 g/week |
| Moderate risk (Canada 2023) | 3–6/week ≈ 40–81 g/week |
| Increasingly high risk (Canada 2023) | more than 6/week ≈ over 81 g/week |
| Australia (NHMRC 2020) | 10 drinks/week ≈ 100 g/week |
| UK (Chief Medical Officers, unchanged since 2016) | 14 units/week ≈ 112 g/week |
| US (2025–2030 Dietary Guidelines) | no numeric limit |
| Binge (single occasion) | 4+ for women, 5+ for men in about 2 hours |
The binge threshold is not really a count. The US National Institute on Alcohol Abuse and Alcoholism (NIAAA) defines binge drinking as a pattern that brings blood alcohol concentration to 0.08 g/dL or higher, and four or five drinks in about two hours is what that corresponds to for a typical adult.[88] Smaller people, women and anyone drinking fast get there on fewer. The table's risk bands are Canada's, while NIAAA's own heavy-drinking line sits far higher, at 15 or more drinks a week for men and 8 or more for women.
The shift from "moderate is fine" to "less is better, ideally zero" reflects better methods, not new puritanism — and note that the two are separable: the case for lower limits strengthened in the same period that US federal guidance dropped its numbers.
What cutting back buys you
Evidence rating: Strong (blood pressure, randomised); Moderate (cancer, observational).
Some things recover quickly, and one does not fully recover at all. Blood pressure responds dose-dependently, with a clear threshold: pooling 36 trials and 2,865 participants, cutting back did nothing measurable for people already drinking two or fewer drinks a day, while people drinking six or more who halved their intake dropped 5.5 mmHg systolic (95% CI 4.3 to 6.7) and 4.0 mmHg diastolic (3.3 to 4.7).[89] A drink in that review is 12 g, so the threshold sits near 25 g a day. A single alcohol-free month in moderate-to-heavy drinkers moved insulin resistance down 25.9%, systolic pressure down 6.6% and weight down 1.5%, with no matching change in a comparison group that kept drinking — though participants chose whether to abstain, so this is not a randomised result.[90] Cancer is the slow one, and only partly reversible: compared with continuing to drink, stopping is associated with about 12% lower risk of oestrogen-receptor-positive breast cancer (relative risk 0.88, 95% CI 0.79 to 0.98) and no detectable change for receptor-negative disease (1.23, 0.98 to 1.55).[91] IARC graded the cessation evidence for breast cancer as limited, so read this single estimate as consistent with that grading rather than stronger than it. Quitting stops the risk you are still accumulating; it does not undo the exposure you already have.
Older adults: medications and falls
Evidence rating: Moderate.
Two alcohol risks get sharply worse with age, and both are acute rather than a lifetime hazard ratio. Around a quarter of older adults — published estimates range from about 20% to 80% depending on the population studied — drink while taking a medication alcohol can interact with; the consequential interactions are excess sedation with benzodiazepines, opioids and antipsychotics, and low blood sugar with sulfonylureas.[92] The review is candid that most of the underlying pharmacology comes from studies not conducted in older adults. Falls are the other: alcohol was involved in 2.2% of US emergency-department visits for falls among people aged 65 and over between 2011 and 2020 — a weighted 618,099 visits — with the head and face the commonest injury site, and the annual rate rose 7.5% a year (95% CI 6.1 to 8.9) from 2011 to 2019, a sustained rise not seen below age 55.[93] If you take a sedating medication or have fallen in the past year, this is the strongest single argument for zero on this page.
Practical guidance
- If you don't drink, don't start for health benefits — none survive rigorous methods.
- If you drink, less is better. Treat each drink as an exposure with a dose, not a category you are either in or out of.
- Count in grams, not glasses. Home and bar pours routinely exceed a standard drink.
- Avoid binge drinking (defined in the table above). It adds acute risks for the liver, brain and heart rhythm on top of your average intake.
- Move the last drink as early as you can if sleep matters, and expect your tracker to disagree with you.
- Family history of breast cancer: the dose-response is a straight line from the lowest intakes that have been measured, with no threshold below which it flattens, so this is one of the few places where the difference between zero and "a small glass most nights" is worth taking seriously.
- Flushing reaction, or East Asian heritage: flushing marks the poor-acetaldehyde-clearing variant, and drinking with it carries sharply raised risk of mouth, throat and oesophageal cancer. Avoidance is the appropriate response.
- Fatty liver, hepatitis, or any liver disease: amounts inside conventional "moderate" limits are enough to reclassify your diagnosis, and obesity and alcohol together raise the risk of liver injury, cirrhosis, liver cancer and death from liver disease by more than either does on its own. Ask which fibrosis stage you are at, because the answer changes the advice: the American Association for the Study of Liver Diseases recommends complete abstinence for anyone with clinically significant fibrosis — stage F2 or worse — rather than for everyone with a fatty liver.[94]
- Pregnancy, trying to conceive, or breastfeeding: abstain.
- Use periods of abstinence as both a diagnostic and a lever. Finding it hard to stop is the signal, not the inconvenience — and among 2,128 participants in a Dutch month-off challenge, drink-refusal confidence was still higher and craving and consumption still lower a year later, with the largest changes among those who completed the month.[95] The study was uncontrolled and self-reported, so read it as encouragement rather than proof.
- Screening doesn't undo the exposure. Screening finds cancers earlier; it does not remove the cause. Reducing intake and screening are complements, not substitutes — see what cutting back buys you.
Further reading
The most load-bearing sources for the page's argument; everything else is cited inline.
- Bagnardi V et al. Alcohol consumption and site-specific cancer risk: a comprehensive dose-response meta-analysis. Br J Cancer 2015.[96]
- Rumgay H et al. Global burden of cancer in 2020 attributable to alcohol consumption: a population-based study. Lancet Oncol 2021.[97]
- 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.[98]
- Lankester J et al. Alcohol use and cardiometabolic risk in the UK Biobank: A Mendelian randomization study. PLoS One 2021.[99]
- 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.[100]
- 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.[101]
- 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.[102]
- 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.[103]
- Esser MB et al. Estimated deaths attributable to excessive alcohol use among US adults aged 20 to 64 years, 2015 to 2019. JAMA Netw Open 2022.[104]
- NASEM. Review of Evidence on Alcohol and Health. National Academies Press 2025.[105]
- 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.[106]
- Topiwala A et al. Alcohol consumption and telomere length: Mendelian randomization clarifies alcohol's effects. Mol Psychiatry 2022.[107]
- Brooks PJ et al. The alcohol flushing response: an unrecognized risk factor for esophageal cancer from alcohol consumption. PLoS Med 2009.[108]
- US Surgeon General. Alcohol and Cancer Risk. Advisory, January 2025.[109]