Was moderate drinking ever actually protective?

The finding was real in the data but the interpretation appears to have been wrong. Decades of cohort studies showed lower cardiovascular mortality in moderate drinkers than in non-drinkers, and this became one of the most widely believed results in public health. The problem lay in who the non-drinkers were: the comparison group included people who had stopped drinking because they had become ill — with liver disease, cancer, heart failure or cognitive decline — or had been told to stop by a doctor. That group has high mortality for reasons unconnected to alcohol, which makes every drinking category look good by comparison. When studies separate lifelong abstainers from former drinkers and adjust properly, the apparent protection shrinks substantially or vanishes.

What is the strongest evidence on this?

Mendelian randomisation, which sidesteps the problem entirely. Certain genetic variants — particularly ALDH2, common in East Asian populations, which causes the alcohol flushing reaction — substantially reduce how much people drink over a lifetime. Those variants are allocated randomly at conception, long before any illness, and are unrelated to income, education, smoking or diet. Using them as a natural randomisation for alcohol exposure removes the confounding and reverse causation that make ordinary cohort studies unreliable here. These studies consistently find that blood pressure, stroke and cardiovascular risk rise continuously with genetically predicted alcohol intake, with no protective range at any level. It is the closest thing to a randomised trial that can ethically be conducted on lifetime drinking.

How strong is the cancer evidence?

It has never been in doubt, and it is considerably stronger than the cardiovascular evidence that received all the attention. Alcohol is classified by IARC as a Group 1 carcinogen — the same evidential category as tobacco and asbestos — and is causally linked to cancers of the mouth, throat, voice box, oesophagus, liver, bowel and female breast. The mechanism is understood: acetaldehyde, the first product of alcohol metabolism, is itself a Group 1 carcinogen, damages DNA and impairs its repair, with additional effects through oxidative stress, disrupted folate metabolism and raised oestrogen levels. For breast cancer, risk begins to rise at low levels of drinking with no identifiable threshold below which there is no increase.

Why do so few people know about the cancer link?

Partly because the cardiovascular story was more appealing and was heavily promoted, and partly because of documented industry involvement. Analyses of alcohol industry-funded information organisations have found systematic misrepresentation or omission of the cancer evidence in material aimed at the public, particularly regarding breast cancer. In 2018 the US National Institutes of Health terminated a large randomised trial of moderate drinking after an internal review found that staff had solicited industry funding and that the study design had been shaped in ways likely to favour a positive result. None of this invalidates individual studies, but it explains why the public conversation for thirty years centred on a contested cardiovascular benefit rather than an undisputed carcinogen — and why cancer warning labelling is now being contested through the same channels used against tobacco labelling.

So what is a safe amount to drink?

On the evidence, there is no level at which cancer risk is not increased, which is why WHO states that no amount is safe for health. That framing is unhelpful if it is heard as an instruction rather than as information, so it is worth being precise: the relationship is a continuum, not a cliff. Less is better and none is best, and the increase in risk at genuinely low intake is small in absolute terms. Canada now describes it exactly this way, as a continuum of risk that begins to rise noticeably beyond about two standard drinks per week, replacing the older idea of a threshold below which drinking is safe. Alcohol also has real social and cultural value, and people are entitled to weigh a quantified risk against that — provided they are actually given the quantity, which historically they have not been.

How does GMJ News cover alcohol and the safe level question?

GMJ News covers the collapse of the cardioprotection hypothesis, sick-quitter bias and how correcting it removes the J-curve, Mendelian randomisation using ALDH2 and ADH1B variants, the never-disputed Group 1 carcinogenicity and the acetaldehyde mechanism, the absence of a threshold for breast cancer, low public awareness and the documented industry role including the terminated MACH trial, revised national guidelines in Canada and the UK and the shift from threshold to continuum framing, alcohol use disorder and withdrawal safety, and the WHO best buy policy measures including minimum unit pricing. Part of PHIG's contested science, cancer prevention and alcohol policy programmes.

Home › Topics › Alcohol and the Safe Level

Alcohol and the Safe Level

GMJ News knowledge hub · last reviewed September 2026 · Georgian Medical Journal

Summary Statistics Glossary GMJ news FAQ Documents Organizations Research

The proposition that moderate drinking protects the heart was among the most widely believed findings in public health and has been substantially overturned, in one of the more important evidence reversals of the past decade: the apparent J-shaped curve is now largely attributed to sick-quitter bias — abstainer groups containing people who stopped drinking because of illness — together with confounding by socioeconomic position, and studies that separate lifetime abstainers from former drinkers find the cardioprotective effect shrinks or disappears (WHO). Mendelian randomisation, which uses genetic variants affecting alcohol metabolism to avoid these confounders entirely, finds no protective threshold and a monotonic relationship between alcohol and cardiovascular risk. Meanwhile the carcinogenicity is not disputed at all: alcohol is an IARC Group 1 carcinogen causally linked to cancers of the breast, colorectum, liver, oesophagus, oropharynx and larynx, with breast cancer risk rising from low levels of consumption — which is why WHO now states there is no safe level for cancer risk, and why several countries have revised guidelines sharply downward.

Key messages

THE REVERSAL: moderate drinking is no longer considered cardioprotective

For decades the J-shaped curve — apparently lower cardiovascular mortality in moderate drinkers than in abstainers — was among the most widely believed findings in public health, and it has substantially collapsed. The principal explanation is methodological rather than a new discovery about alcohol: abstainer comparison groups in these studies contained people who had stopped drinking because of illness, along with lifelong abstainers who differ systematically in health, income and social circumstances. When these groups are separated properly, the apparent protection shrinks markedly or disappears.

SICK-QUITTER BIAS EXPLAINS MOST OF THE J-CURVE

People stop drinking for reasons — developing liver disease, cancer, heart failure, cognitive decline, or being advised to stop by a clinician. Classifying them as non-drinkers creates a comparison group enriched with people who are already ill and about to die, which makes any drinking category look protective by comparison. A systematic review and meta-analysis of 107 cohort studies found that studies with the least bias, using lifetime abstainers as the reference and adjusting appropriately, showed no significant mortality protection at any level of drinking.

MENDELIAN RANDOMISATION REMOVES THE CONFOUNDING ENTIRELY

Genetic variants affecting alcohol metabolism — principally ALDH2 and ADH1B — are randomly allocated at conception, precede any disease, and are unrelated to income, diet, smoking or social circumstances. Using them as instruments for lifetime alcohol exposure eliminates the confounding and reverse causation that plague conventional epidemiology. These studies consistently find a monotonic relationship: cardiovascular risk and blood pressure rise with genetically predicted alcohol intake, with no protective threshold. This is the strongest available evidence and it points in one direction.

NEVER DISPUTED: alcohol is an IARC Group 1 carcinogen

The carcinogenicity has never been in question, and it receives strikingly little public attention relative to the cardiovascular debate. Alcohol is causally linked to cancers of the oral cavity, pharynx, larynx, oesophagus, liver, colorectum and female breast. Breast cancer risk rises from low levels of consumption without a threshold, meaning there is no drinking level at which cancer risk is not increased. The mechanism is established: acetaldehyde, the primary metabolite, is genotoxic, forms DNA adducts and is itself Group 1 classified, with additional pathways through oxidative stress, altered folate metabolism and raised oestrogen.

PUBLIC AWARENESS OF THE CANCER LINK IS STRIKINGLY LOW

Survey after survey finds that a minority of the public associates alcohol with cancer, and awareness of the breast cancer link is particularly poor even among women. This is not accidental: it reflects decades of industry-funded health messaging emphasising cardiovascular benefit, and industry-funded information bodies have been documented misrepresenting the cancer evidence. Cancer warning labelling, implemented in Ireland and South Korea and under consideration elsewhere, is the policy response and is being contested through the same trade and legal mechanisms used against tobacco labelling.

WHAT FOLLOWS: guidelines are moving down, and the framing has changed

Canada moved from a weekly limit to a continuum of risk, describing any amount above roughly two standard drinks per week as carrying increasing risk. The UK reduced its guideline to 14 units weekly for both sexes and abandoned the previous distinction. WHO states that no level of alcohol consumption is safe for health. The reframing matters: rather than a threshold below which drinking is safe, the evidence supports a dose-response in which less is better and none is best — while acknowledging that alcohol has social and cultural value that individuals may reasonably weigh against a quantified risk.

Key statistics

107 cohorts

meta-analysis found no significant mortality protection once bias was addressed

JAMA Netw Open 2023

Monotonic

Mendelian randomisation finds risk rising with alcohol exposure and no protective threshold

JAMA Netw Open/BMJ

IARC Group 1

alcohol and acetaldehyde both classified as carcinogenic to humans

IARC

No threshold

breast cancer risk increases from low levels of consumption without a safe lower limit

IARC/WCRF

Low awareness

a minority of the public associates alcohol with cancer risk in repeated surveys

WHO Europe

~2 drinks/week

Canadian guidance describes risk as increasing beyond this, replacing threshold framing

CCSA 2023

Alcohol and health — where the disagreement actually lies

Source: Bars show strength of supporting evidence. The cardioprotection claim has largely collapsed; the carcinogenicity was never disputed.

Glossary of key terms

Sick-quitter bias
EpidemiologyThe systematic error arising when former drinkers who stopped because of ill health are included in the abstainer reference group. Because people commonly stop drinking after developing liver disease, cancer, heart failure, cognitive impairment or on medical advice, this group carries elevated mortality that has nothing to do with not drinking. Comparing any drinking category against it manufactures an appearance of protection. The correction is to use lifetime abstainers as the reference, exclude former drinkers or analyse them separately, and account for the fact that even lifetime abstainers differ systematically — they are more likely to be older, less affluent, in poorer health and from specific religious or cultural groups. Studies applying these corrections progressively lose the J-curve.

Mendelian randomisation in alcohol research
Genetic epidemiologyThe ALDH2 rs671 variant, common in East Asian populations, impairs acetaldehyde clearance and causes the flushing reaction, substantially reducing alcohol consumption in carriers. ADH1B variants alter ethanol oxidation rate with similar consequences. Because these alleles are randomly assigned at conception and precede any exposure or disease, they function as natural randomisation for lifetime alcohol intake, removing confounding by socioeconomic position, diet and smoking, and removing reverse causation entirely. Large studies including the China Kadoorie Biobank have used this design and found blood pressure and stroke risk increasing continuously with genetically predicted intake, with no protective range. The principal limitations are pleiotropy and the concentration of informative variants in specific populations.

Acetaldehyde and the mechanisms of carcinogenicity
CarcinogenesisEthanol is oxidised by alcohol dehydrogenase to acetaldehyde, itself IARC Group 1 classified, which forms DNA adducts, causes double-strand breaks and interferes with repair. It is then metabolised by aldehyde dehydrogenase 2 to acetate; impaired ALDH2 function raises acetaldehyde exposure and, in people who drink despite flushing, substantially increases oesophageal cancer risk. Additional mechanisms: CYP2E1 induction generating reactive oxygen species; interference with folate absorption and one-carbon metabolism affecting DNA methylation and synthesis; increased circulating oestrogen, relevant to breast cancer; and enhanced penetration of other carcinogens through mucosa, which explains the multiplicative interaction between alcohol and tobacco in upper aerodigestive tract cancer.

Industry involvement in alcohol research and messaging
Conflicts of interestThe cardioprotection narrative was substantially amplified by industry-funded research and communication, and this history is documented rather than alleged. The Moderate Alcohol and Cardiovascular Health trial, a large randomised study of moderate drinking, was terminated by the US National Institutes of Health in 2018 after an internal review found that institute staff had solicited industry funding and that the design had been shaped in ways favourable to a positive result. Separately, analyses of industry-funded information bodies have documented systematic misrepresentation or omission of the alcohol-cancer relationship in materials directed at the public. This does not by itself invalidate any individual study, but it is essential context for why the evidence base developed as it did and why public awareness of cancer risk remains low.

Alcohol use disorder and dependence
ClinicalDistinct from the population risk question and requiring different clinical action. Screening with AUDIT or AUDIT-C identifies hazardous drinking efficiently in primary care, and brief interventions have reasonable evidence for reducing consumption in hazardous but non-dependent drinkers. Dependence requires assessment for withdrawal risk: unplanned cessation in physically dependent individuals can produce seizures and delirium tremens, which carries meaningful mortality, so advising abrupt cessation without assessment is dangerous. Pharmacotherapy is under-prescribed relative to its evidence base: acamprosate and naltrexone both reduce relapse, with disulfiram effective under supervision. Thiamine supplementation is essential to prevent Wernicke encephalopathy, and should be given parenterally where deficiency is suspected.

Population-level policy: the WHO best buys
Public health policyWHO identifies three interventions as the most cost-effective for reducing alcohol harm: increasing excise taxes, restricting availability through licensing hours and outlet density, and comprehensive bans or restrictions on advertising, sponsorship and promotion. Minimum unit pricing, implemented in Scotland and subsequently elsewhere, targets the cheapest high-strength products consumed disproportionately by the heaviest drinkers, and evaluations have reported reductions in consumption and in alcohol-attributable deaths. Each measure faces sustained industry opposition through trade challenges, legal action and voluntary-agreement alternatives — a pattern closely paralleling tobacco control history, including the same arguments about personal responsibility, unintended consequences and the primacy of education.