The headline finding the public-health field reversed on
For most of the last three decades, the "1-2 drinks a day is good for you" claim was treated as established public-health science. It appeared in the US Dietary Guidelines for Americans. It was the basis for the polite-society defence of moderate drinking. The graph behind it - the so-called "J-curve" - showed mortality lowest at 1-2 standard drinks per day, rising at higher doses but also rising at zero.
The Griswold et al. 2018 analysis in The Lancet rewrote that consensus [1]. Pooling 694 studies covering 195 countries and 28 million participants, with corrections for the methodological flaws that had driven the J-curve, the conclusion was unambiguous: the safest level of alcohol consumption for overall health is zero drinks per week. The dose-response was roughly linear above zero - any drink incrementally raised your overall health risk. The earlier J-curve was an artifact, not a finding.
Three years later, the WHO's 2023 statement made it official policy: "no level of alcohol consumption is safe for our health" [2]. The US Dietary Guidelines have been slower to formally retract but the scientific advisory committee's 2025 evidence review reached the same conclusion in private even as the public guideline language stays cautious.
Why the J-curve was wrong
Three methodological problems drove the older J-curve. Each is now broadly acknowledged.
Abstainer bias. The "non-drinker" comparison group in older studies bundled lifelong abstainers with people who had quit drinking due to ill health (recovering alcoholics, people stopping for liver disease, people stopping after a cardiac event, etc.). The latter group's elevated mortality dragged the non-drinker baseline up - making 1-2 drinks per day look protective in comparison [3]. The Stockwell et al. 2016 reanalysis re-classified studies that distinguished "former drinker" from "lifelong abstainer" and found the J-curve disappeared.
Healthy-drinker confounding. Moderate drinkers in observational studies skew higher socioeconomic status, higher educational attainment, more health-conscious behaviour, more regular medical care. Even after adjusting for the obvious confounders (income, exercise, smoking), residual confounding from unmeasured lifestyle correlates persisted. Mendelian randomisation studies (which exploit genetic variants in alcohol-metabolism genes as instrumental variables, bypassing the confounding) consistently show no protective effect of moderate drinking [4].
Reverse causation in the elderly. Older adults who stop drinking often stop because they're already getting sicker. Lumping them with the "never drank" group in old-age cohorts produced spurious protective effects for the "still drinking" group.
When you correct all three: the J-curve flattens out and turns into a monotonically rising line from zero.
What moderate drinking actually does to your sleep
The most under-appreciated near-term cost of moderate drinking shows up in sleep architecture. Most people experience alcohol as "I fall asleep faster", and that's true - alcohol is acutely sedating. But the same molecule that produces the sedation suppresses REM and slow-wave (deep) sleep in the first half of the night, then produces rebound wakefulness in the second half as it metabolises and the body releases the suppression [5].
Pietila et al. 2018 measured this in the field using wearable HRV data across 4,098 nights from 4,026 individuals [6]. The dose-response on overnight HRV was striking:
- Low-dose (≤2 drinks for men, ≤1 for women): HRV dropped 6-9% vs sober nights
- Moderate-dose (3-6 men, 2-4 women): 14% drop
- High-dose (≥7 men, ≥5 women): 19% drop
HRV is the autonomic-nervous-system marker most strongly linked to overnight cardiovascular recovery + next-day cognitive performance. A 9% drop on an "I only had a glass with dinner" night is real. The subjective "I sleep better after a drink" sensation is the front-end sedation; the actual restorative architecture is worse than a sober night, not better.
If you wear an Oura, Whoop, Apple Watch, or Garmin, you can test this on yourself in a week. Track HRV nightly. Have a glass of wine with dinner on Tuesday + Saturday, nothing else. Watch the trend. The data is reproducible enough to convince most people in 7 nights.
For the broader picture of what sleep does for biological age, our piece on sleep stages and longevity walks through the deep / REM / glymphatic clearance biology.
The cancer dose-response
The International Agency for Research on Cancer classifies alcohol as a Group 1 carcinogen - the highest category, alongside tobacco + asbestos + ionising radiation [7]. The mechanism: alcohol is metabolised to acetaldehyde, which is directly genotoxic + interferes with DNA repair.
The dose-response is steepest for cancers of tissues the alcohol directly contacts (oral cavity, pharynx, larynx, oesophagus) but extends to breast, liver, colorectal, and stomach cancers. Bagnardi et al. 2015 meta-analysis quantified the dose-response across all alcohol-associated cancers, pooling 572 studies [8]:
- Oral / pharyngeal cancer: RR ~1.42 per 10g/day (one drink)
- Oesophageal squamous cell carcinoma: RR ~1.30 per 10g/day
- Liver cancer: RR ~1.05 per 10g/day, climbing steeply above 50g/day
- Breast cancer (female): RR ~1.07 per 10g/day, statistically significant even at "light" intake
- Colorectal: RR ~1.07 per 10g/day
The breast-cancer finding is the one that surprises most moderate drinkers: every 10g of ethanol (one standard drink) per day raises lifetime breast cancer risk ~7-10% [9]. There's no safe threshold; the effect is linear from zero.
What about cardiovascular disease?
The historical case for moderate drinking was specifically about cardiovascular protection. The mechanism cited: HDL cholesterol elevation + fibrinogen reduction + slight platelet inhibition.
Recent Mendelian randomisation studies have largely dismantled this case [4]. Using genetic variants in alcohol-metabolism enzymes as instrumental variables (which bypass the lifestyle confounding of observational studies), genetic predisposition to drink less is associated with lower cardiovascular disease risk + lower all-cause mortality. The historical "moderate drinking protects the heart" finding doesn't survive the instrumental-variable correction.
Even where some protective effect on coronary heart disease can still be demonstrated, the cancer-risk increase from the same alcohol intake dominates the net-effect calculation. The Lancet 2018 analysis explicitly modelled the cardiovascular + cancer combined effect and found zero as the optimum across all considered outcomes.
"What about red wine?"
The red-wine-is-good-for-you belief started with the 1980s "French paradox" observation (the French had lower CV disease than expected for their saturated-fat intake; one explanation pointed at red wine) and was amplified by early in vitro work on resveratrol [10].
Subsequent human evidence has been consistently disappointing. Resveratrol concentrations in red wine are far too low to replicate the lifespan effects seen in mice - you'd need to drink ~100 L of wine per day to match the typical experimental dose. The polyphenol content of red wine is positive on the margins; the alcohol content drives all the same negative outcomes (sleep disruption, cancer risk, BP elevation, HRV suppression) regardless. The net effect is the same as any other alcoholic beverage with the same ethanol content.
If you want the resveratrol or polyphenol mechanism: eat the grapes, or take a polyphenol-rich diet (berries, dark chocolate, green tea, olive oil) where the dose-response is real and the ethanol confounder isn't there. Don't drink wine for the polyphenols.
If you do drink: harm-reduction levers that actually work
The framing "abstinence or nothing" doesn't suit most people's actual lives. If you do drink, the harm-reduction levers are real and the dose-response is your friend in reverse - less is meaningfully less harmful.
Total weekly volume matters most. The cancer + cardiovascular dose-response is roughly per-gram-of-ethanol. Cutting from 14 drinks/week to 7 cuts the elevated risk roughly in half. Going from 7 to 3 cuts it further. You don't have to hit zero to materially move the dial.
Binge episodes are worst per-gram. Five drinks Saturday is biologically worse than one drink each day, despite the same weekly total. Binge-pattern drinking concentrates the acetaldehyde load + cardiovascular strain into a single window that the body's clearance systems can't handle gracefully.
Never within four hours of sleep. The HRV + sleep-architecture cost is worst when alcohol metabolises during the second half of the night. Last drink with dinner, not after dinner.
Eat with it + drink slower. Slows the absorption + reduces peak blood alcohol. The total ethanol load is the same but the dose-rate matters for the acute organ stress.
Skip nights when your HRV is already low. If your morning HRV is below baseline, your autonomic system is already under recovery load. Adding alcohol stacks the cost. The wearable data turns this from "willpower" into "obvious choice".
Don't combine with sleep medication. The respiratory-depression interaction is non-trivial. Same logic for benzodiazepines, opioids, and Z-drugs.
Track abstinence-day count, not abstinence-percentage. "5 dry nights this week" is a clearer target than "70% sober" and lands the same arithmetic.
What the evidence does not support
Three claims that don't survive a careful read.
- "A daily glass of red wine is heart-healthy." The Lancet 2018 + Mendelian randomisation literature buries this. The net effect of any ethanol intake on overall mortality is non-negative - zero is the optimum.
- "Light drinking improves your social/mental health enough to outweigh the physical cost." Possibly true for some individuals' subjective experience, but unprovable at the population level. The published mental-health evidence shows correlation, not causation, and reverse causation (anxious people self-medicate) is the dominant interpretation.
- "Tolerance means you handle alcohol better." Behavioural tolerance is real; physiological tolerance is mostly fictional. Your liver doesn't get better at clearing acetaldehyde with practice. Your judgment about your own drunkenness does get worse with practice.
The takeaway
The J-curve is dead. The safest level of alcohol consumption for overall health is zero, with a roughly linear dose-response above that. The strongest near-term cost of moderate drinking shows up in sleep architecture + HRV (measurable on consumer wearables in a week). The strongest long-term cost is cancer risk, with breast cancer being the most surprising finding for moderate drinkers. Red wine isn't protected by its polyphenol content; the ethanol fraction drives the same harms as any other drink. If you do drink, total weekly volume + binge-pattern avoidance + protecting your last 4 hours before sleep are the real harm-reduction levers.
If you want overnight HRV, sleep stages, and the rest of your bio-age markers tracked together so you can see what alcohol actually does to your numbers, have a look at Thier.
Frequently asked questions
Is moderate drinking really bad for you?
Yes. The Lancet 2018 Global Burden of Disease analysis (Griswold et al., pooling 694 studies across 195 countries) concluded that the safest level of alcohol consumption for overall health is zero drinks per week. Every additional drink incrementally raises overall health risk. The earlier "J-curve" belief (that 1-2 drinks reduced mortality vs zero) was driven by methodological confounds - chiefly the abstainer-bias problem where ex-drinkers who quit due to ill health were lumped with lifelong non-drinkers, dragging the non-drinker mortality up artificially.
Does alcohol affect sleep quality?
Yes, significantly. Alcohol is sedating in the first half of the night but suppresses both REM and slow-wave (deep) sleep architecture, producing rebound wakefulness in the second half as it metabolises. A 2018 meta-analysis (Pietila et al., wearable validation across 4,098 nights) measured HRV-during-sleep dropping 6-19% on nights with any alcohol vs sober nights, with dose-dependent severity. Even one drink with dinner measurably degrades the sleep stages most associated with cellular repair and emotional consolidation. The subjective "I sleep better after a drink" sensation is the sedation; the actual restorative architecture is worse, not better.
Is red wine actually good for you?
No. The "red wine is good for you" belief came from the French paradox observation in the 1980s + early in vitro work on resveratrol. Subsequent human evidence has been consistently disappointing. Resveratrol concentrations in red wine are far too low to replicate the lifespan effects seen in mice (you'd need to drink ~100L of wine per day to match the experimental dose), and the alcohol fraction of the wine drives all the same negative outcomes (sleep disruption, cancer risk, BP elevation) regardless of polyphenol content. If you want resveratrol's mechanism, eat the grapes.
Does alcohol increase cancer risk meaningfully?
Yes. The International Agency for Research on Cancer classifies alcohol as a Group 1 carcinogen (the highest category, alongside tobacco + asbestos). The dose-response for breast cancer specifically is steep and well-established - every additional 10g of ethanol per day (roughly one drink) increases breast cancer risk by 7-10% (Bagnardi et al. 2015). For oral, oesophageal, liver, and colorectal cancers, the dose-response is similarly clear. Cancer is the dominant longevity-relevant risk of moderate drinking; the cardiovascular effects are smaller in either direction.
Should I never drink at all?
That's a personal choice, not a strictly health-driven one. The honest answer the published evidence supports: zero is healthier than any positive amount, the dose-response is roughly linear (no safe threshold), and the social/cultural value of moderate drinking is the only legitimate counter-argument. If you do drink, the harm reduction levers are real: drink less, drink slower, eat with it, never within 4 hours of sleep, never with sleep medication, and skip nights when your HRV is already low. Total weekly volume matters most; binge episodes are the worst-per-gram pattern.
References
- GBD 2016 Alcohol Collaborators (Griswold MG, et al.). Alcohol use and burden for 195 countries and territories, 1990-2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet. 2018;392(10152):1015-1035. PubMed
- WHO Europe. No level of alcohol consumption is safe for our health. WHO Statement. 4 January 2023. WHO
- Stockwell T, Zhao J, Panwar S, et al. Do "moderate" drinkers have reduced mortality risk? A systematic review and meta-analysis of alcohol consumption and all-cause mortality. Journal of Studies on Alcohol and Drugs. 2016;77(2):185-198. PubMed
- Holmes MV, Dale CE, Zuccolo L, et al. Association between alcohol and cardiovascular disease: Mendelian randomisation analysis based on individual participant data. BMJ. 2014;349:g4164. PubMed
- Ebrahim IO, Shapiro CM, Williams AJ, Fenwick PB. Alcohol and sleep I: effects on normal sleep. Alcoholism: Clinical and Experimental Research. 2013;37(4):539-549. PubMed
- Pietila J, Helander E, Korhonen I, et al. Acute Effect of Alcohol Intake on Cardiovascular Autonomic Regulation During the First Hours of Sleep in a Large Real-World Sample. JMIR Mental Health. 2018;5(1):e23. PubMed
- IARC Working Group on the Evaluation of Carcinogenic Risks to Humans. Alcohol consumption and ethyl carbamate. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans. 2010;96:3-1383. PubMed
- Bagnardi V, Rota M, Botteri E, et al. Alcohol consumption and site-specific cancer risk: a comprehensive dose-response meta-analysis. British Journal of Cancer. 2015;112(3):580-593. PubMed
- Seitz HK, Pelucchi C, Bagnardi V, La Vecchia C. Epidemiology and pathophysiology of alcohol and breast cancer: Update 2012. Alcohol and Alcoholism. 2012;47(3):204-212. PubMed
- Baur JA, Sinclair DA. Therapeutic potential of resveratrol: the in vivo evidence. Nature Reviews Drug Discovery. 2006;5(6):493-506. PubMed