A study commissioned by the Biden administration and published this week in the Journal of Studies on Alcohol and Drugs carries a headline conclusion its authors wanted written into American dietary guidelines: alcohol has no protective effect on mortality at any level, and health risks begin at one drink a day. The Trump administration declined to feature the findings in the 2025-2030 Dietary Guidelines for Americans, advising instead that people have “less alcohol for better overall health”.
According to the analysis, men consuming 14 standard drinks a week — the upper limit the previous guidelines permitted — face a 1-in-25 lifetime risk of an alcohol-attributable death. Risk begins to register at roughly seven drinks a week for both sexes. Even levels considered “moderate” raise the risk of premature death and more than 200 diseases, including heart disease and cancer, researchers found.
But a closer look at the charts and tables shows the conclusions are not as straightforward as they appear to be.
At one drink per week, the study’s central estimate — the model’s single most probable value — is not a positive number representing harm. It is minus 16.30 alcohol-attributable deaths per 1,000 men. At three drinks a week, minus 10.56. Negative numbers here imply protection: that drinking at these levels is associated with fewer deaths than not drinking at all.
The authors note that the confidence interval — the statistical range within which the true value plausibly falls — crosses zero at these doses, rendering the result uncertain. A confidence interval running from minus 47 to plus 11 deaths per thousand, as it does for men at one drink per week, is consistent with strong protection, mild harm, and everything in between — all at once.
{{/usCountry}}The authors note that the confidence interval — the statistical range within which the true value plausibly falls — crosses zero at these doses, rendering the result uncertain. A confidence interval running from minus 47 to plus 11 deaths per thousand, as it does for men at one drink per week, is consistent with strong protection, mild harm, and everything in between — all at once.
{{/usCountry}}But that is a simplistic reading of what is a statistical expression of bad study design.
Behind it is the study’s architecture. It is a modelling exercise built on relative risks drawn from observational research — studies recording health outcomes in people who drink versus those who don’t, without random assignment. Drinkers and non-drinkers differ in ways that resist full statistical correction: income, diet, physical activity, healthcare access.
The researchers were not unaware of these traps. They chose lifetime abstainers — people who never drank — as their reference group rather than all non-drinkers, a deliberate attempt to exclude former drinkers who may have quit because they were already ill, and whose presence would make abstention look artificially unhealthy. They also adjusted self-reported consumption upward against national sales data, correcting for the tendency of respondents to undercount their drinking.
These are genuine methodological choices, made in good faith and documented transparently.
The authors acknowledge in their own limitations section, however, that the definition of lifetime abstention varies across the source studies from which the relative risks are drawn — which means the correction, however carefully constructed, rests on inputs it cannot fully standardise.
And that shows in the numbers, which ultimately undermines the very point it was trying to make.
Vinay Prasad, an American epidemiologist who until recently served as a senior FDA official under the Trump administration, rightly picked on some of the unusual implications. For instance, among women at modest consumption levels, the disability-adjusted life year calculations — measuring healthy life after accounting for time lost to illness and premature death — show a marginal gain rather than a loss. That is not a finding on which any public health policy can rest.
Alcohol is too socially embedded and commercially powerful for this debate to resolve itself on scientific grounds alone, much less if the scientific grounds appear shaky.
After the study’s draft findings were released, trade associations called it “irretrievably flawed.” The House oversight committee called it “fraught with bias,” accusing the authors of predetermined conclusions. Robert Vincent, the government official who led the study, told the Associated Press he was “asked to kill it” within the Trump administration, refused, and was later laid off.
Two of the five officials tasked with recommending tighter guidelines — including the CDC’s alcohol policy lead — were fired in the spring staff reductions. A rival study commissioned by Congress at $1.3 million was itself partly the product of industry lobbying, according to two former alcohol lobbyists who exposed it to media.
The industry’s campaign has targeted the study’s methodology — but not because of the flaws that, inadvertently, end up showing alcohol to be protective in low doses; but because of its headline recommendation that – by all indications – appears to be an editorial choice.
This does not minimise alcohol’s documented harm at higher doses determined by more robust epidemiological studies. The scientific consensus goes: ethanol is metabolised to acetaldehyde, a compound that forms chemical bonds with DNA — adducts — that disrupt normal replication and can become permanent mutations.
Alcohol also impairs folate metabolism and elevates circulating oestrogen, both relevant to hormone-sensitive cancers.
For liver disease, the progression from fatty liver to hepatitis to fibrosis to cirrhosis to hepatocellular carcinoma is dose-responsive and mechanistically established.
For injuries, the relationship with blood alcohol concentration is steep: each 0.02% rise in BAC raises the odds of a fatal road crash by roughly 74%.
On cardiovascular disease, Mendelian randomisation studies — which use naturally occurring genetic variants that alter alcohol metabolism as a proxy for lifetime drinking, bypassing the confounding that distorts observational data — have generally found no protective effect.
The most powerful, the China Kadoorie Biobank study published in The Lancet in 2019, followed 512,715 people for ten years. Where conventional analysis suggested cardiovascular protection, the genetic analysis found none.
The new study, in some ways, harks back to the early days of the pandemic – the time since when trust in science has eroded, instead of improving.
At the time, in the spring of 2020 as lockdowns spread, some public health guidance framed masks primarily as protection for the wearer. The stronger evidence was for source control: masks reduce what an infected person spreads more reliably than what an uninfected person catches. When that gap became visible, it was used to discredit all mask guidance — including the source-control evidence, which had never been seriously in doubt.
The cancer and liver findings are not in dispute — the evidence is established, the mechanisms are understood, and no serious scientific body contests them. The low-dose mortality question is different: the data is uncertain, the methodology is contested, and this study does not resolve it. For the average person looking to science and public health messaging to make sense of what they put in their bodies, that distinction matters. Presenting both findings under the same headline, as though the uncertainty of one were as settled as the solidity of the other, ties the credible to the contested. It hands anyone with an interest in the outcome — industry, ideology, or otherwise — a way in.