The study, published in the European Journal of Epidemiology, drew on the Whitehall-1 Study, a legendary cohort of London civil servants first recruited between 1967 and 1970. Of the original 19,019 men, 8,537 survivors were resurveyed in 1997 and 1998, when their average age was 77. Researchers led by Sodiq Yusuff of the Clinical Trial Service Unit and Epidemiological Studies Unit at the University of Oxford focused on 4,582 men who had no prior diagnosis of cardiovascular disease or cancer, and tracked their deaths through national mortality registers until mid-2020. The result was an extraordinarily complete picture of mortality at the very end of life: during an average follow-up of 11.8 years, 4,194 of the participants, a staggering 91 percent, died, with more than 99 percent of those deaths occurring after age 70.

Participants reported their drinking habits through postal questionnaires. The researchers classified them as lifelong abstainers, ex-drinkers who had previously consumed alcohol but had stopped, or current regular drinkers, the latter divided into five bands of weekly intake ranging from fewer than 7 units per week to 42 or more. A single unit corresponds to 8 grams of pure alcohol, roughly half a standard glass of wine. Light drinkers consuming fewer than 7 units weekly served as the reference group, a deliberate choice designed to avoid the abstainer bias that has plagued earlier studies, where comparisons against non-drinkers can exaggerate apparent benefits of modest consumption.

The headline finding was stark. Men who drank 42 or more units per week, equivalent to roughly 21 pints of beer or more than two bottles of wine, faced a 38 percent higher risk of death from any cause compared with light drinkers, with a hazard ratio of 1.38 and a 95 percent confidence interval of 1.15 to 1.67. Ex-drinkers also fared poorly, showing a 21 percent elevated risk. Meanwhile, lifelong abstainers and those drinking 7 to 27 units per week showed no statistically significant difference from the light-drinking reference group. Among current drinkers, each additional 10 units of weekly alcohol was associated with a 5 percent increase in all-cause mortality risk.

When the researchers dissected deaths by cause, the excess risk concentrated overwhelmingly in non-vascular mortality. Heavy drinkers showed a 39 percent higher risk of non-vascular death, and the dose-response analysis pointed chiefly to alcohol-related cancers, where each additional 10 units per week carried a 19 percent higher risk. This aligns with long-standing evidence from the International Agency for Research on Cancer that alcohol is a carcinogen for multiple tumour types. By contrast, the association with vascular mortality was weaker and statistically uncertain, with a hazard ratio of 1.04 per 10 units per week and confidence intervals that crossed unity, leaving open the question of whether alcohol has any meaningful effect, protective or harmful, on heart disease and stroke deaths in this age group.

What makes this study particularly compelling is the methodological machinery deployed to dismantle the biases that have historically inflated the apparent benefits of moderate drinking. First, the team separated ex-drinkers from lifelong abstainers, a critical distinction because people who quit alcohol often do so precisely because illness has made drinking untenable, the so-called sick-quitter effect. Second, they excluded anyone with pre-existing vascular disease or cancer, removing those whose disease might have driven both their abstinence and their mortality. Third, they used light drinkers rather than abstainers as the reference category. Fourth, sensitivity analyses excluded deaths in the first five years of follow-up, and the core findings held firm. Finally, the researchers calculated E-values, a quantitative measure of how strong an unmeasured confounder would need to be to explain away the results. For heavy drinkers the E-value was 2.11, meaning only a very substantial hidden factor could account for the observed risk.

Even so, the authors are candid about the limits of their evidence. When deaths in the first ten years of follow-up were excluded, the associations became modest and non-significant, though the researchers note it is difficult to determine whether this reflects reverse causation or simply a loss of statistical power. When the analysis was restricted to never smokers, the associations attenuated, with wider confidence intervals, hinting at possible residual confounding by smoking or reduced precision in the smaller subgroup. The E-values for some categories, such as 1.30 for abstainers, suggest that relatively modest unmeasured confounding could account for those particular estimates. Alcohol intake was also measured only once, at the 1997-98 resurvey, so drinking patterns in middle age could not be accounted for, and self-reported consumption is inherently imprecise.

The study’s population also imposes constraints on generalisability. These were retired male civil servants, 94 percent of whom had held high employment grades at retirement, a strikingly privileged group compared with the broader British population. Survivor bias is another concern, since men alive at resurvey in their late seventies may represent a healthier subset of the original cohort, potentially over-representing light-to-moderate drinkers. The researchers could not assess binge drinking, beverage type, or drinking context, such as whether alcohol was consumed with meals. And the sample size, while sufficient for all-cause mortality, limited precision for vascular deaths and rarer cause-specific categories like dementia, pneumonia and infection, where the exploratory analyses were likely underpowered.

Nevertheless, the broader picture painted by the study is consistent with the direction of modern alcohol epidemiology. Two previous meta-analyses encompassing more than four million participants across all adult ages, which carefully accounted for reverse causation, found no statistically significant mortality difference between abstainers and light-to-moderate drinkers. Mendelian randomisation studies, which use genetic variants as natural instruments to probe causality, have likewise shown strong linear associations between genetically predicted alcohol intake and all-cause mortality, with no protective signal for moderate consumption. The Whitehall-1 analysis now extends that convergence of evidence into the oldest age brackets, where the J-shaped curve has been most persistent and most seductive.

The public health implications are considerable. Populations across high-income countries are ageing rapidly, and regular alcohol consumption among older people is rising, with alcohol-related deaths in England recently reaching record highs. If the apparent cardioprotective halo around moderate drinking in old age is largely an artefact of sick-quitter bias, healthy-user bias and residual confounding, then there is no epidemiological basis for recommending alcohol as a longevity aid to elderly patients. The authors conclude that their findings support public health strategies to reduce population-level alcohol consumption in older people, and to prevent alcohol-related deaths in old age just as in middle age. For the millions of older adults who have clung to the idea that a daily drink is a prescription for a longer life, this study delivers an uncomfortable but rigorously argued correction: when it comes to alcohol and mortality in later life, more is worse, and less is probably not protective at all.

Subject of Research: Associations between alcohol consumption and cause-specific mortality in older men from the Whitehall-1 cohort

Article Title: Alcohol consumption and cause-specific mortality in older men: Whitehall-1 study of London civil servants

Article References: Alcohol consumption and cause-specific mortality in older men: Whitehall-1 study of London civil servants. (n.d.). https://doi.org/10.1007/s10654-026-01465-4

Image Credits: AI Generated

DOI: 10.1007/s10654-026-01465-4

Keywords: alcohol consumption, mortality, older adults, Whitehall study, epidemiology, cardiovascular disease, cancer, sick-quitter bias, cohort study, public health, hazard ratio, ageing