I don’t think we have seen this one before even though it was published in July. I am currently tracking a period without alcohol, but I am intending to start drinking again probably in about a week’s time. It may be a little longer simply to have two more blood tests. My GGT started out in the normal range, but has reduced by about 20% during this.
chatGPT(6max):
This paper strengthens the observational evidence linking higher alcohol consumption to increased mortality. Its apparent protective association for wine is much less secure. The novelty is mainly greater detail, and several inconsistencies in the published tables warrant clarification.
Li and colleagues: s12937-026-01361-3.pdfTotal and beverage-specific alcohol use at mid-life and all-cause and cause-specific mortality: a prospective cohort study, Nutrition Journal, 2026.
Summary. The researchers analysed 340,924 UK Biobank participants, initially aged approximately 40 to 70 years, over an average of 13.4 years. There were 22,381 deaths, including 4,288 cardiovascular deaths and 11,063 cancer deaths.
Alcohol consumption was measured by questionnaire at baseline. The authors examined total consumption, different beverages, and drinking with meals, using statistical models adjusted for demographic, lifestyle and health characteristics.
Their terminology needs attention: “moderate” consumption meant above 20 to 40 grams daily for men and above 10 to 20 grams daily for women. For men, the upper limit corresponds to 35 UK units weekly.
The principal results were:
| Consumption category | All-cause mortality HR | Cancer mortality HR |
|---|---|---|
| Moderate total alcohol | 1.02 | 1.11 |
| High total alcohol | 1.24 | 1.36 |
| Low beer/cider | 1.01 | 1.07 |
| Low spirits | 1.08 | 1.07 |
| Low wine | 0.89 | 0.92 |
| Moderate wine | 0.92 | 1.02 |
| High wine | 1.02 | 1.10 |
HR means hazard ratio. A value of 1.24 indicates a 24% higher estimated death hazard, rather than a 24 percentage point increase in the probability of dying. Each comparison uses the corresponding never or occasional consumption category.
The strongest findings were:
- High total consumption: higher all-cause mortality, HR 1.24, with a 95% confidence interval of 1.17 to 1.31; and higher cancer mortality, HR 1.36, interval 1.26 to 1.47.
- Moderate total consumption: higher cancer mortality, without a statistically clear association with all-cause mortality.
- Spirits: adverse associations were already present in the low consumption category.
- Wine: low and moderate consumption were associated with lower all-cause mortality. Moderate wine consumption was also associated with lower cardiovascular mortality, HR 0.79, interval 0.70 to 0.88. High wine consumption was associated with higher cancer mortality.
- Drinking with meals: generally associated with lower mortality than drinking outside meals.
The authors also examined 20 cancer mortality outcomes, reporting associations involving head and neck, respiratory, gastrointestinal, liver, blood, breast and genital cancers. These findings varied substantially by beverage and consumption category.
Novelty. The broad finding that wine drinkers have different mortality patterns from beer or spirits drinkers is already well established in observational research, including earlier analyses of UK Biobank.
Three particularly relevant predecessors are:
- Jani and colleagues, 2021: studied 309,123 UK Biobank participants and already examined beverage type, drinking frequency and drinking with food in relation to mortality and other outcomes. Springer Nature Link
- Schaefer and colleagues, published online in 2022: studied 354,386 participants and reported a curved association between wine consumption and mortality, alongside generally adverse associations for other alcoholic beverages. PMC
- Ortola and colleagues, 2024: examined 135,103 older UK Biobank drinkers using essentially the same alcohol categories, while considering frailty, deprivation, wine preference and drinking with meals. JAMA Network
The present paper adds finer separation of red, white and fortified wine, a broader age range than the older-adult study, longer follow-up than some predecessors, and a wider assessment of cancer mortality outcomes.
I would describe this as an incremental epidemiological contribution. Its more unusual cancer associations are potentially interesting, but require independent confirmation. Because these studies reuse UK Biobank, their agreement does not constitute replication in independent populations.
Strengths. The study has a large sample, substantial follow-up, linked death records, detailed beverage information and extensive adjustment. Excluding participants with baseline cardiovascular disease or cancer helps reduce some reverse causation. The authors also report sensitivity analyses excluding early deaths and lifetime abstainers, and apply false discovery rate correction to the cancer analyses.
The main criticisms are as follows.
-
The wine association cannot establish that wine protects health.
Beverage choice is associated with education, deprivation, diet, social circumstances and drinking patterns. The paper itself shows differences between beverage groups.
Adjustment helps, but smoking was represented by broad categories rather than detailed lifetime exposure. This matters particularly because low wine consumption was associated with lower lung cancer mortality: residual differences in smoking intensity and duration are a plausible explanation.
The study cannot determine how much of the wine association comes from the beverage and how much comes from the characteristics of its drinkers.
-
The beverage comparisons do not directly test switching drinks at equal alcohol intake.
The reference group for wine consists of people consuming little or no wine. Some may consume substantial beer or spirits. The equivalent problem applies to the other beverage analyses.
Although the models adjust for other beverages, they do not directly establish what happens when a person replaces a fixed amount of beer-derived ethanol with the same amount from wine.
A formal substitution analysis would better address claims about beverage choice.
-
There are concrete reporting inconsistencies.
Comparing the paper with the journal’s supplementary material reveals several problems:
Location Concern Main Table 1 versus Supplementary Table 2 College-level education is reported as 11.6% overall in the main table, but approximately 36% in the supplement, despite the same stated cohort size. Main Table 1 Weighting the alcohol category means by their participant counts gives approximately 144.9 grams weekly, rather than the reported overall mean of 148.6 grams. Rounding cannot explain this difference. Supplementary Table 1 “Female genital cancers” includes code C50, which the same table identifies as breast cancer. The last issue is especially relevant to the cancer conclusions. If that definition was used in the analysis, the female genital cancer endpoint includes breast cancer and overlaps with a separately reported outcome.
These could be errors in presentation rather than analysis. The published material does not allow that distinction to be resolved; corrected tables and analysis code are needed.
-
The abstract generalises beyond some individual results.
Low beer/cider consumption was associated with higher cancer mortality, but its all-cause and cardiovascular estimates were essentially neutral: HR 1.01 and 1.00, respectively. Moderate beer/cider consumption also lacked a clear cardiovascular association.
Similarly, moderate wine consumption was not associated with lower cancer mortality.
The broad wording about low to moderate beverage consumption therefore needs more outcome-specific qualification.
-
Exposure measurement and participant selection limit interpretation.
One baseline questionnaire cannot capture changes in drinking over 13 years, previous heavy consumption or detailed binge patterns. The authors suggest measurement error probably weakened the associations, but its direction is not guaranteed across multiple exposure categories and beverage types.
More than 109,000 participants were excluded for incomplete alcohol data. The paper should explain more clearly how questionnaire routing, missingness and abstainer classification contributed to this exclusion.
Excluding early deaths reduces some reverse causation, but cannot eliminate the effects of longstanding illness or earlier drinking behaviour.
-
Some adjustments may remove part of alcohol’s effect.
Blood pressure, cholesterol and frailty can reflect consequences of previous alcohol exposure as well as factors influencing current drinking.
Adjusting for them may therefore obscure part of the total effect being investigated. Presenting clearly justified models with and without potential intermediate variables would improve interpretation.
-
The cancer findings need more cautious presentation.
The 20 outcomes include overlapping categories, such as gastrointestinal cancers and their component cancers. They are not 20 independent discoveries. Some estimates also have wide confidence intervals, indicating limited precision.
Furthermore, cancer mortality combines cancer development, diagnosis, treatment and survival. An association with cancer death does not identify which process alcohol affected.
False discovery rate correction is useful, but cannot correct confounding, coding errors or unstable estimates.
-
The proposed biological explanations remain speculative.
Suggestions involving wine antioxidants, resveratrol or beer purines were not tested. There were no measurements connecting these compounds to the mortality associations.
In particular, the claim that antioxidants may mainly explain the apparent wine benefit goes beyond the evidence presented.
The most useful next steps would be correction of the reporting discrepancies, publication of reproducible code and cancer event counts, and an independent analysis incorporating repeated alcohol measurements, detailed smoking and drinking patterns, and explicit comparisons at equal ethanol intake.