Alcohol and healthy aging, what the reassessed evidence shows

For decades, a single shape dominated public understanding of alcohol and health: the J-curve, where light-to-moderate drinkers appeared to live longer than people who did not drink at all. That shape came from real data. It has not held up well as the studies behind it have been re-examined for a specific design flaw, and the current evidence base, including two federally commissioned reviews completed in 2025, no longer points in one clean direction. This page sets out what changed, why, and what the current evidence actually supports.

Why the J-curve looks different under scrutiny

The original cardioprotection finding largely came from cohort studies comparing current drinkers with a single “non-drinker” reference group. That reference group is not as clean as it sounds. It typically includes former drinkers who quit because of illness, and people who never drank because of preexisting health problems, alongside genuinely healthy lifelong abstainers. If sick people are disproportionately counted among the “non-drinkers,” moderate drinkers will look healthier than abstainers by comparison even if alcohol itself does nothing protective. This is called abstainer bias, and it has been a known concern in this literature for years.

Stockwell, Zhao, and colleagues (Journal of Studies on Alcohol and Drugs, 2024) tested this directly, pooling 107 cohort studies covering more than 4.8 million participants and 425,564 deaths. Studies that used younger cohorts and excluded former and occasional drinkers from the abstainer reference group found low-volume drinkers had essentially the same mortality risk as abstainers (relative risk 0.98, 95% CI 0.87 to 1.11). Studies that did not meet those quality criteria found a significantly lower risk for low-volume drinkers (relative risk 0.84, 95% CI 0.79 to 0.89). The size of the apparent protective effect tracked the quality of the study design, not just the alcohol.

A related analysis by the same research group, Zhao and colleagues (JAMA Network Open, 2023), reanalyzed the same 107-study pool with adjustments for sampling variation, former-drinker bias, and other study-level quality criteria. In the fully adjusted model, occasional drinkers (relative risk 0.96) and low-volume drinkers (relative risk 0.93) showed no statistically significant reduction in all-cause mortality compared with lifetime nondrinkers, while risk rose significantly at 45 grams of alcohol per day and higher (relative risk 1.19 to 1.35), and rose at a lower threshold for women than men.

What genetic evidence adds

A cohort study, however carefully adjusted, cannot fully rule out confounding by factors researchers did not measure. Millwood and colleagues (Lancet, 2019) took a different approach in the China Kadoorie Biobank, a prospective study of over 500,000 adults. Because two common genetic variants strongly affect how much men in this population drink, largely independent of their other health behaviors, the variants function as a natural randomized assignment, a method called Mendelian randomization. Conventional self-reported drinking data in this cohort showed the familiar U-shaped curve, with moderate drinkers (about 100 grams a week) appearing to have lower stroke and heart attack risk than non-drinkers. Genotype-predicted alcohol intake showed no such U-shape. Instead, it had a continuously positive association with intracerebral hemorrhage (relative risk 1.58 per 280 g/week) and ischemic stroke (relative risk 1.27 per 280 g/week), and no significant association with heart attack. The study’s own conclusion: “the apparently protective effects of moderate alcohol intake against stroke are largely non-causal.”

All-cause mortality and cancer risk specifically

Two large syntheses give the clearest population-level answer to how alcohol relates to overall risk. The GBD 2016 Alcohol Collaborators (Lancet, 2018), analyzing 694 data sources and 592 studies across 195 countries, developed a method to estimate the level of alcohol consumption that minimizes total health risk across 23 alcohol-related outcomes combined. That level was zero standard drinks per week (95% uncertainty interval 0.0 to 0.8). The same analysis found that among people aged 50 and older, cancers accounted for 27.1% of alcohol-attributable deaths in women and 18.9% in men, the single largest cause of alcohol-attributable death in that age group.

On cancer specifically, the National Cancer Institute states there is “strong scientific evidence that alcohol drinking can cause cancer,” classifying alcohol as a Group 1 carcinogen (the same category as tobacco) with established links to cancers of the oral cavity, throat, larynx, esophagus, liver, breast, and colorectum. The NCI fact sheet notes that women who have one drink a day already show a higher breast cancer risk than women who drink less than one drink a week, a dose-response relationship that starts at low intake rather than beginning at some higher threshold. The WHO put the same point plainly in January 2023: “there is no safe amount that does not affect health,” and roughly half of alcohol-attributable cancers in Europe were estimated to come from “light” and “moderate” drinking, not heavy drinking.

Where U.S. federal guidance actually stands

The picture at the institutional level is genuinely more contested than either “moderate drinking is protective” or “any drinking is clearly harmful” suggests, and that contest is itself part of the honest record. Congress directed the National Academies of Sciences, Engineering, and Medicine to review the evidence as one input to the Dietary Guidelines for Americans, 2025-2030; reporting on that 2025 committee report indicates it concluded, with moderate certainty, that drinking within the study’s defined moderate range was associated with lower all-cause and cardiovascular mortality than not drinking. A separate 2025 review, produced by the Interagency Coordinating Committee on the Prevention of Underage Drinking for HHS, evaluated a substantially overlapping body of observational literature and reported that mortality risk begins rising at low average levels of intake, without identifying a protective threshold. Two federally commissioned reviews, drawing on much of the same underlying cohort literature, reached different emphases, which is consistent with the methodological point demonstrated directly above: how a review handles the abstainer-comparison problem changes the answer it produces.

What eventually shipped in the Dietary Guidelines for Americans, 2025-2030 moved away from the prior edition’s specific numeric ceiling (up to 1 drink a day for women, up to 2 for men) toward guidance to drink less for better health, without stating that any specific nonzero amount is protective.

Where claims outrun the evidence, in both directions

“A daily glass of wine protects the heart” states as settled a finding that the best-designed evidence available, genetic and bias-corrected, does not support; the raw cohort association it rests on has a documented, demonstrated source of distortion. The opposite overreach also happens: treating population-level findings like “the mortality-minimizing level is zero” as proof that a specific already-established individual habit carries a specific, quantified personal risk goes beyond what a population dose-response curve, built from aggregated cohorts, can say about any one person. The cancer dose-response relationship is real and documented at low intake levels; it does not by itself specify what any individual’s absolute risk increase is.

Weighing this evidence against an individual situation

Alcohol interacts with numerous medications, and the risk calculus differs by sex, age, liver function, family history of cancer or alcohol use disorder, and other conditions, none of which a population-level relative risk can resolve for a specific person. How this reassessed evidence bears on an existing drinking pattern is a question for a clinician who knows the relevant history, not something this evidence base, built from group averages, is designed to answer on its own.