ISFAR critique 306 – Reevaluating the Alcohol–Cancer Link: Long-Term Cancer Mortality Outcomes in the REGARDS Study

Pinheiro, L.C., Jumonville, G., Ringel, J., Berg, K.Y., Rosenberg, S., Tonorezos, E. Chandwani, R., Tamimi, R.M., Phillips, E.

Journal of General Internal Medicine (2026) Online ahead of print https://doi.org/10.1007/s11606-026-10479-3

Abstract

Background: Cancer prevention strategies often focus on behaviour change such as increased screening, smoking cessation, and healthy diets. Recently, enthusiasm for addressing alcohol as a significant cancer risk factor has gained attention. Using a large, population-based cohort, we sought to determine independent associations between alcohol consumption and cancer mortality.

Methods: We used data from the REGARDS cohort study, which enrolled 30,239 adults 45+ years between 2003 and 2007 and follows them today. At enrolment, participants self-reported alcohol consumption as none, light (≤ 3 drinks/week), moderate (4–7 drinks/week for women, 4–14 drinks/week for men), and heavy (≥ 8 drinks/week for women and ≥ 15 drinks/week for men). We estimated Cox models to determine associations between alcohol consumption and cancer mortality.

Results: Among 26,694 participants, mean age was 64.4 (SD 9.4) years, 44% were male, 42% were Non-Hispanic Black, 63% reported no alcohol, 22% light, 11% moderate, and 4% heavy alcohol consumption. Over a median 13.3-year follow-up, we observed 2306 cancer deaths. After full adjustment for covariates, compared to abstainers, heavy drinkers had an increased risk of cancer death (1.21; 95% CI 1.01–1.45), and light drinkers had a decreased risk of cancer mortality (0.87; 95% CI 0.78–0.98), and there was no association between moderate drinking and cancer mortality.

Conclusions: Our findings contribute to the growing evidence that heavy alcohol use is consistently linked with higher cancer mortality. We encourage cancer prevention strategies to step away from isolating single health behaviours and consider holistic perspectives of an individual’s lifestyle including physical activity, smoking, diet, and alcohol consumption.

ISFAR Summary

Although the REGARDS (Pinheiro et al., 2026) and Burden of Proof (Dai et al., 2026) studies employ markedly different methodological approaches, both reinforce a consistent public health message that heavy alcohol consumption is associated with increased cancer mortality and incidence. The REGARDS study also suggests that light-to-moderate drinking may appear neutral or even protective against cancer mortality in certain analyses. These findings may have been sensitive to how drinking categories are defined, to adjustment for important confounders such as smoking and to non-adjustment for other important lifestyle factors. The authors emphasise the need for careful interpretation of epidemiological evidence and consideration of the broader lifestyle context.

Background

Cancer accounted for one in five deaths across OECD countries – particularly cancers of the lungs, colon and rectum, pancreas, breast, and prostate (OECD, 2025). A considerable proportion of cancers may be prevented through lifestyle changes. Various risk factors have been identified and ranked by their contribution to cancer incidence. Key lifestyle factors include smoking, excess body weight, UV radiation, poor diet, pathogenic infections, and physical inactivity.[1] Alcohol consumption has been estimated to contribute 5.4% to all cancer cases.[2]

Alcohol consumption and cancer risk have been discussed in previous ISFAR critiques. Some recent examples are ISFAR critiques #303, #300 and #290, to indicate that the alcohol-cancer association is topical. Critique #303 discussed findings by Arecco et al. (2026), confirming alcohol consumption as a modifiable risk factor, particularly for oestrogen receptor-positive breast cancer. However, alcohol appeared to have a limited influence on breast cancer prognosis. ISFAR critique #300 discussed a study by Larsson et al. (2025), which did not find an overall positive association between alcohol consumption and the incidence of any cancer type, but unfortunately, these authors defined their alcohol consumption using genetic information. ISFAR critique #290 discussed a brief feature article by Pearson (2025) that elaborated on the 2025 US Surgeon General’s Advisory on Alcohol and Cancer Risk.

Overall, the aspects derived from these critiques of the alcohol-cancer association may be summarised as follows. Firstly, a unilateral approach is usually applied to the alcohol-cancer association. Positive associations between alcohol consumption and cancer risk are emphasised, whereas negative associations are neglected. Secondly, alcohol consumption has been positively associated with only a limited number of cancers. Thirdly, the level of alcohol consumption is usually not considered, which is unfortunate, since only heavy alcohol consumption is associated with a number of cancer types affected by alcohol consumption. Fourthly, cancer risk is presented as the most important aspect of alcohol consumption, whereas other, more impactful and beneficial effects on cardiovascular health are not taken into account. This is a pity, because many prospective studies report a J-shaped association with overall mortality (death from all diseases and other causes combined), indicating that light and moderate alcohol consumption has an overall beneficial effect on mortality, including cancer deaths. Fifthly, alcohol consumption is not the only lifestyle factor affecting cancer risk; other lifestyle factors, such as smoking, overweight and obesity, and physical inactivity, contribute to the overall lifestyle-associated risk. Some interactions between lifestyle factors may exist and are not always accounted for. Sixthly, no context is provided in terms of the magnitude of all risks arising from lifestyle and non-lifestyle factors, such as environmental factors. 

The paper by Pinheiro et al. (2026) takes a different approach. The authors did not estimate risks using publicly available data but used a longitudinal cohort to determine the association between self-reported alcohol consumption and cancer mortality in a large, diverse US cohort. Cancer mortality reflects not only cancer incidence but also the effects of cancer treatment. The author concluded that “Our findings contribute to the growing evidence that heavy alcohol use is consistently linked with higher cancer mortality.”

A more common approach, inventorying risks for various diseases by alcohol-drinking category, was used by Dai et al. (2026) in a recent study from the Institute of Health Metrics and Evaluation. Selected reviews of health outcomes were used to estimate the burden of proof for the health effects associated with alcohol consumption. The authors arrived at very similar, already well-described conclusions; while acknowledging various health benefits of light and moderate drinking, they state, “While potential health impacts at low-to-moderate levels varied by outcome, high levels of alcohol consumption were associated with increased risk across all outcomes.”   

Critique

The REGARDS (REasons for Geographic and Racial Differences in Stroke) study enrolled some 25,000 participants aged 45+ and followed this diverse US population for almost 15 years. The study covered some 300,000 person-years, which may be considered relatively modest, especially given the population’s great diversity. The cohort also included a large proportion of non-drinkers (63%), which may have affected the precision of estimates for the higher-drinking groups, where only approximately 4% of the cohort were heavy drinkers. In addition, alcohol consumption was assessed only once at baseline. Any changes in drinking during the subsequent 13 years could not be captured, introducing exposure misclassification that would tend to attenuate true associations. This limitation is particularly relevant for alcohol consumption, as drinking habits commonly change over time in older adults because of ageing, illness, medication use, or changes in lifestyle.

Characteristics of the participants varied significantly across drinking categories, so multiple models were used to adjust for these differences. Non-drinkers were relatively unhealthy; they had higher rates of diabetes and a history of stroke, and were more likely to be living with obesity, although they were less likely to be smokers. Although multivariable adjustment was undertaken, reverse causation cannot be completely excluded, as some participants may have reduced or stopped drinking because of underlying ill health. After full adjustment, a small but significantly lower hazard ratio was observed among light drinkers, whereas no significant association was found for moderate drinkers. Heavy drinkers consistently exhibited higher hazard ratios across all models. Overall, the findings support the authors’ conclusion that heavy alcohol consumption is associated with increased cancer mortality. However, cancer mortality should not be interpreted as equivalent to cancer incidence, because mortality is influenced not only by the occurrence of cancer but also by tumour aggressiveness, stage at diagnosis, access to treatment, treatment effectiveness, and competing causes of death. Furthermore, given the relatively old age of the cohort at enrolment (64 years), competing risks from cardiovascular disease and other causes should also be considered when interpreting cancer mortality outcomes.

Relatively little information has been provided on the possible drinking histories of the non-drinkers. Drinking categories were constructed based on the NIAAA guidelines and a recent meta-analysis by Stockwell (Stockwell et al., 2024), but the paper does not demonstrate the effect of this guidance. The paper indicates that light and moderate drinkers may not benefit substantially in terms of cancer mortality, but it also shows that heavy drinking increases the risk of cancer mortality. The only significant decrease in cancer mortality was observed in light drinkers after adjustment for smoking, which may indicate that smoking is an important contributor to cancer incidence and mortality and that some residual confounding may still exist among current and former smokers. Unfortunately, other important lifestyle factors, such as diet quality and physical activity, were not included in the analysis. In addition, factors such as drinking pattern (e.g. binge versus regular consumption), beverage type, changes in socioeconomic status during follow-up, healthcare utilisation, and participation in cancer screening programmes may also influence cancer mortality and could have contributed to residual confounding. Interestingly, the pattern of hazard ratios does not suggest a simple linear dose-response relationship. While light drinking was associated with a significantly lower hazard ratio and heavy drinking with a higher hazard ratio, the absence of an association among moderate drinkers was not explored and may warrant further investigation. Supplementary analyses using the conventional three-category NIAAA classification (none, moderate and heavy) are also informative. When light and moderate drinkers were combined into a single moderate drinking category (up to 7 drinks per week for women and up to 14 for men), the fully adjusted hazard ratio was 0.89 (95% CI 0.81–0.97), compared with 1.21 (95% CI 1.01–1.45) for heavy drinkers. These findings suggest that the interpretation of the association at lower levels of alcohol consumption is influenced by the categorisation of alcohol consumption, reinforcing the need for caution when interpreting apparent differences between light and moderate drinking categories.

Unlike the REGARDS study, which reflects the experience of a single prospective cohort, Dai et al. synthesised evidence from numerous cohort and case-control studies, thereby providing a broader overview of the available epidemiological evidence across multiple populations and health outcomes. Rather than relying solely on pooled effect estimates, the Burden of Proof framework incorporates assessments of study quality, potential bias and uncertainty to derive more conservative estimates of the strength of the evidence. Following a systematic review and study selection, dose-response relationships were estimated, potential bias was adjusted for, and uncertainty was incorporated using the Burden-of-Proof Risk Function (BPRF). These estimates and adjustments yield a Risk Outcome Score (ROS). The ROS is converted into a star rating that reflects the strength of the evidence. However, the resulting estimates depend on a series of modelling assumptions about exposure-response relationships, bias adjustment and uncertainty estimation. Consequently, they are more sensitive to modelling assumptions than estimates derived directly from an individual prospective cohort. Accordingly, the REGARDS and Burden of Proof studies should be viewed as complementary rather than competing approaches, each with distinct strengths and limitations. The results indicate that J- or U-shaped relationships exist between alcohol consumption and various health outcomes, each rated 1 or 2 stars. Only pharyngeal cancer has a high ROS and an accompanying five-star rating; other cancers associated with high alcohol consumption, as well as cirrhosis and chronic liver disease, have three stars. Converting continuous risk estimates into a one- to five-star evidence rating provides a useful summary for readers but may oversimplify complex and sometimes heterogeneous epidemiological evidence. Overall, the study likewise indicates that high or heavy alcohol consumption is associated with substantially increased and well-documented risks of disease and mortality.

The authors also state in their conclusions that their study should not be interpreted as endorsing alcohol consumption for health benefits. While the results may suggest benefits, these observations are uncertain under the scoring system used. Despite markedly different methodologies, both studies reach broadly similar conclusions: the strongest and most consistent adverse health effects are observed among heavy drinkers, whereas the evidence on light-to-moderate alcohol consumption is less certain and varies by the health outcome examined.

In summary, these findings suggest that, within this cohort, heavy alcohol consumption was consistently associated with increased cancer mortality, whereas the interpretation of findings for low-risk drinking was influenced by analytical choices, including the categorisation of alcohol consumption and adjustment for important confounders such as smoking. These results should not be interpreted as evidence on cancer incidence, nor as establishing the safety of alcohol consumption. It is important to distinguish cancer mortality from cancer incidence. Cancer mortality is influenced not only by the occurrence of cancer but also by tumour characteristics, stage at diagnosis, treatment, and competing causes of death. Consequently, associations with cancer mortality cannot be equated with associations with cancer incidence.

Although the REGARDS and Burden of Proof studies employ markedly different methodological approaches, both reinforce a consistent public health message that heavy alcohol consumption is associated with adverse health outcomes. At the same time, the REGARDS study also shows that light-to-moderate drinking may be neutral or even protective against cancer mortality in certain analyses. The supplementary analyses presented by Pinheiro et al. (2026) further suggest that the interpretation of lower levels of alcohol consumption may be sensitive to how drinking categories are defined, emphasising the need for careful interpretation of epidemiological evidence and for consideration of the broader lifestyle context.

Specific comments

Forum member Ellison appreciated the opportunity to review this report, which is designed to estimate the cancer death rates that may be associated with alcohol consumption. However, he had some concerns about the methodology:

“(1) A majority of subjects were from the Southeastern USA, an area where alcohol use is strongly condemned by religious and social norms (an area with a high percentage of Protestants where any alcohol is considered “sinful”).  We know that social and religious pressures would be expected to have many participants not want to disclose any alcohol drinking or the amount, so the usual finding of “under-reporting” would be high.  This suggests to me that the commonly used three-level classification (none, light to moderate, heavy) may be a more appropriate way to judge the effects of increasing amounts.  Further, the baseline characteristics, point estimates, and 95% confidence intervals in the main results, as reported by the authors in the text and in the supplementary data, are very similar between the two classification groups. allows comparisons with results from most earlier studies.  In addition, the number of subjects reporting “heavy drinking” was very small, making it difficult to judge the degree of effect on mortality, especially impairing the ability to separate casual drinkers with values slightly above moderate levels from true heavy alcohol abusers.

(2)   I sought in vain for a report of total mortality in the paper, but only “cancer mortality” was given.  While the approach they used for this information may have been slightly improved over using death certificate data alone, there is still considerable uncertainty in determining whether the cause of death was due to cancer or simply death in someone with cancer.  I was not given the total mortality rates in this study.

(3)   The authors point out that the non-drinking group in this study had higher rates of diabetes and obesity, both strong risk factors for cardiovascular disease (CVD) deaths. Since moderate alcohol intake clearly reduces the risk of diabetes and, in many studies, lowers the prevalence of obesity, part of the increased risk in the abstainer group might be due to abstinence from alcohol, rather than being treated as a true confounder of effect.

(4)   As with many studies, only baseline alcohol intake data were available, so changes over long periods of observation could not be used as a potentially more accurate measure of exposure. Also, the authors had no data on drinking patterns (type of beverage, with or without food, regularly or in binges, etc.), which provide a better estimate of exposure than the number of drinks consumed. Further, in most other studies, concurrent smoking, which is always more frequent among heavier drinkers, is a stronger factor in calculating the risk of death than that reported in this paper; smoking must always be included as a confounder.” 

Forum member Waterhouse comments that “the results strongly suggest that the first drop of alcohol does not increase risk, at least from a cancer perspective. The light drinkers had the lowest rate of mortality, significantly lower than non-drinkers.  The data demonstrates a “J” shaped curve for cancer mortality, somewhat similar to heart disease mortality.” 

Forum member Ursini muses that “recent epidemiological analyses have renewed the debate on the relationship between alcohol consumption and health. Although alcohol is an established cause of several cancers, these studies also illustrate the complexity of its biological effects: the dose-response relationship is not linear, and several health outcomes—notably cardiovascular disease, and in some studies even cancer—display the familiar J-shaped curve. Rather than dismissing such observations as paradoxical, they should prompt a more fundamental question: why should living systems respond linearly to environmental challenges?

Since Aristotle, moderation has been regarded as a fundamental principle of health. The concept of the golden mean (Aurea Mediocritas), later echoed by Paracelsus in the famous statement that “the dose makes the poison,” anticipated a central principle of modern biology: the biological effect of any exposure depends not only on its intrinsic properties but also on its dose. Glucose perhaps provides the clearest example. At physiological concentrations, it is indispensable for life; in excess, it becomes a major driver of metabolic disease. This is not a paradox but the natural consequence of a nonlinear biological response. There is no compelling reason why the same principle should not apply to alcohol, phytochemicals, or even some truly toxic compounds.

This concept underlies hormesis and, more specifically, parahormesis. A mild challenge does not simply produce less damage than a larger one; it activates endogenous adaptive programs that increase cellular resilience. During ethanol metabolism, the transient generation of electrophilic species, including acetaldehyde, may initially act as a stress signal that modifies the Keap1–Nrf2 system, promotes glutathione synthesis, and enhances the cellular nucleophilic tone. As exposure increases, however, acetaldehyde accumulates, DNA damage and mutagenesis predominate, and the same molecule behaves as the carcinogen it is unequivocally known to be.

The beneficial limb of the J-shaped curve, therefore, reflects not the intrinsic benefit of ethanol, but the predominance of adaptive responses over direct toxicity. Once adaptive capacity is exceeded, molecular injury progressively overwhelms cellular defences, and disease risk rises. The observed dose–response relationship is thus the result of two competing processes: direct toxicity and adaptive protection.

This interpretation has implications far beyond alcohol. If a mild biological challenge can activate adaptive pathways, operating on nucleophilic tone, that are sufficient to reduce cardiovascular risk, there is no obvious biological reason why the same mechanisms should not also contribute to protection against neurodegeneration and, at least during the initiation phase of carcinogenesis, against cancer. The nucleophilic tone is now recognised as a major determinant of resistance to oxidative, electrophilic, and inflammatory stress.

The J-shaped curve should, therefore, no longer be regarded as an epidemiological paradox, but as the epidemiological signature of biological adaptation. The real challenge is not to explain the paradox, but to understand the nonlinear biology that generates it.”

Forum member McIntosh comments that “there are a number of statistical problems with the paper. First, in choosing cancer mortality due to alcohol use as the variable to be explained, there is a selection problem. In their duration model, the selection of cancer is related to alcohol because the probability of having cancer is related to alcohol use. But alcohol use is also a regressor in the model so the parameter estimates associated with alcohol use will be biased. There are many simple examples that show this.

 There is another source of bias, and that is caused by the presence of competing risks. In a competing risk model of two diseases: heart disease and all other diseases, as outlined in McIntosh (2014). In their model, the probability of having died of cancer, PC, will rise if the probability of the other diseases is negatively related to alcohol use. In this framework, individuals die of cancer because they don’t die of other diseases.

This paper also showed that there were parameter biases associated with the variable of interest (cancer) when the other diseases which censor these durations are related to alcohol use.

Forum member Skovenborg endorses the critique and the later comments from ISFAR members. The results from a large cohort study are always an important addition to the increasingly popular meta-analyses; however,

•       The abstainers were participants who, at enrollment, reported no alcohol use in the previous 7 days. The risk of potential misclassification in drinking categories is obvious; however, the study adjusted for a history of hypertension, high cholesterol, diabetes, heart disease, stroke and obesity. In view of the high percentage of abstainers in this US population, it is plausible that the majority of the abstainers are lifelong teetotalers;

•       It is always difficult to separate the drink from the drinkers, as illustrated by the observation that some of the excess risk among heavy drinkers is attributable to their higher prevalence of smoking;

•       The non-significance of moderate alcohol consumption and cancer mortality [0.91 (0,79-1.04)] may be compared with the borderline significance of heavy consumption [1.21 (1.01-1.45)]; and

•       Data on the association between alcohol consumption and all-cause mortality should have been included in the report.

Forum member Romano writes that “the study benefits from a large sample and long follow-up; however, several methodological issues limit the strength of its conclusions. From a sampling perspective, the REGARDS cohort was designed to investigate stroke rather than cancer and enrolled only English-speaking U.S. adults aged 45 years or older. Therefore, the cohort is not representative of the general population, limiting external validity. Exclusion of participants with previous cancer and missing alcohol data may also have introduced selection bias. Furthermore, alcohol intake was assessed only once at baseline through self-report, assuming stable drinking habits over more than a decade despite the likelihood of substantial changes over time, thereby increasing exposure misclassification.

The statistical approach using Cox proportional hazards models and multiple imputation is appropriate, but important concerns remain. Marked baseline differences in smoking, obesity, diabetes, socioeconomic status, and functional health strongly suggest residual confounding. Particularly noteworthy is that the complete-case analysis lost statistical significance after full adjustment, indicating that the findings depend, at least in part, on the imputation strategy. In addition, the observed association for heavy drinking was modest (HR=1.21; 95% CI 1.01–1.45), with a borderline confidence interval. Such small effect sizes are especially susceptible to unmeasured confounding. The study also failed to evaluate drinking patterns (e.g., binge drinking), beverage type, cumulative lifetime alcohol exposure, or time-varying consumption.

The interpretation of the apparent protective effect of light drinking also deserves caution. Although the authors acknowledge residual confounding, they do not adequately address the well-recognised ‘sick quitter’ and ‘healthy drinker’ biases, whereby abstainers frequently include former drinkers with poorer health, while light drinkers often have healthier lifestyles and better access to healthcare. Moreover, all cancer deaths were analysed together despite substantial biological heterogeneity across alcohol-related cancers, potentially masking site-specific associations. Important determinants of cancer mortality, including cancer stage, treatment, screening practices, dietary quality, and healthcare utilisation, were unavailable.

Consequently, the conclusions should be more conservative. While the study supports an association between heavy alcohol consumption and cancer mortality, its observational design does not establish causality. The evidence presented is insufficient, by itself, to reinforce public health recommendations regarding alcohol and cancer without corroboration from studies incorporating repeated exposure measurements, cancer-specific analyses, and more comprehensive control of time-varying and residual confounding.”

Forum member Mattivi “fully supports the critique and the suggestions of Raquel Romano; in particular, it is a severe limit that ‘alcohol intake was assessed only once at baseline and without any detail ‘.”

Forum member Kaplan considers that, “like many issues in epidemiology, this one is more complicated than it first appears. I did some work on this using the REGARDS dataset (see attached). In this large cohort of more than 30,000 participants, educational attainment was among the strongest predictors of survival. Adjustment for multiple covariates, including alcohol use, attenuated the association but did not eliminate it. We concluded that educational attainment is a powerful predictor because it serves as a proxy for what Pinheiro et al. (2026) describe as a holistic lifestyle. In general, people with more education tend to engage in healthier behaviours.

REGARDS revealed one notable exception to that pattern. Participants with lower educational attainment generally reported less potentially hazardous alcohol consumption, whereas those with more education tended to drink more. This relationship appears to be even stronger in the UK, where several studies have found that higher educational attainment is associated with significantly greater alcohol consumption (Beard et al. 2029, Rosoff et al. 2021).

One limitation of the Pinheiro et al. analysis is that alcohol consumption was not clearly disentangled from demographic factors because the covariates were entered in large blocks. As a result, it is difficult to determine the independent contribution of specific covariates.   One advantage of the REGARDS study is that the investigators have made their data available, allowing others to explore these relationships in greater detail.

Many questions remain unanswered. As expected, the available evidence supports an association between high-risk drinking and several cancers. However, the authors appear to discount the possibility—suggested by their own data—that low levels of alcohol consumption may be associated with a lower risk of some cancers.

The unresolved question is whether alcohol itself has a protective effect at low levels of consumption, or whether the apparent association reflects a broader constellation of healthy lifestyle factors, with educational attainment serving as a marker for a collection of those characteristics. Access to the underlying data may help answer that question.

Forum member Harding also supports the critique, and shares the reservations expressed by other Forum members.  A few specific points also occur to him as follows:

1.    Heart disease and stroke remain the leading causes of death in the USA. According to the CDC, in 2024 heart disease was the cause of 683,037 deaths, stroke of 163,000 deaths, and cancer of 619,812 deaths. Given the strong protective effect of light to moderate alcohol consumption on heart disease and stroke, an increase in cancer deaths among light to moderate drinkers is expected simply because they live long enough to die from it.

2.    Even so, cancer risks declined among light to moderate drinkers. The authors argue that ‘the apparent protective effects of light to moderate consumption are less stable and may be explained by residual confounders’ (Discussion, last paragraph). Could the same not be true of heavy drinkers?

3.    Most research on alcohol and cancer is site-specific and shows both increases and decreases in risk, depending on the site. Therefore, it is hard to see the justification, or indeed the value, of lumping all cancers together.

Forum member Lanzmann agrees with all the Forum member comments, “especially those who point out the absence of analysis of different alcoholic beverages, but the size of the study and the number of cancer deaths are too small, probably also the number of wine drinkers. I continue to think that the relation between wine drinking and cancer mortality risk for most of the cancers is very different compared to beer and spirits drinking, due to the beneficial “fruit” effect of the wine that is higher than the detrimental alcohol effect for moderate drinkers, but this topic is not and cannot be treated in this article. Any large observational study of cancer mortality risk and alcohol consumption that does not differentiate alcoholic beverages (and drinking pattern) is exposed to see nothing except the dose effect of alcohol for heavy drinkers.”

References

Arecco, L., Cacilhas, P. M., Bobato Lara Gismondi, C., Bruzzone, M., Gentile, G., Gerosa, R., Blondeaux, E., Agostinetto, E., Dauccia, C., Lobo-Martins, S., Grochot, R., Saini, K. S., Azim, H. A., Debiasi, M., De Caluwé, A., Buisseret, L., Del Mastro, L., Lambertini, M., & de Azambuja, E. (2026). Association between alcohol consumption and breast cancer incidence and prognosis: A systematic review and meta-analysis. Breast, 86. https://doi.org/10.1016/j.breast.2026.104719

Beard, E., Brown, J., West, R., Kaner, E., Meier, P., & Michie, S. (2019) Associations between socio-economic factors and alcohol consumption: A population survey of adults in England. PLoS One, 14(2):e0209442. doi: 10.1371/journal.pone.0209442. Erratum in: PLoS One. 2019 Apr 30;14(4):e0216378.

Dai, X., Nicholson, S. I., Lawlor, H. R., Carr, S., Steinmetz, J. D., Chen, N. M., McLaughlin, S. A., Hay, S. I., Ong, K. L., Aravkin, A. Y., Zheng, P., Sorensen, R. J. D., Gilbertson, N. M., Mullany, E. C., O’Connell, E. M., Murray, C. J. L., & Gakidou, E. (2026). Health effects associated with alcohol consumption: a Burden of Proof study. Nature Health 2026, 1–21. https://doi.org/10.1038/s44360-026-00139-5

Larsson, S. C., Mason, A. M., Cronjé, H. T., Bassett, E., Horta, G., Kar, S., & Burgess, S. (2025). Alcohol consumption and risk of cancer: a Mendelian randomization analysis of four biobanks and consortium data. BMC Medicine, 23(1). https://doi.org/10.1186/s12916-025-04543-8

McIntosh, J. (2014). Inference Problems in the Analysis of the Relationship between Alcohol Consumption and Coronary Heart Disease. Communications in Statistics-Simulation and Computation, 43.10, 2275-2282.

OECD. (2025). Health at a Glance 2025. OECD Publishing. https://doi.org/10.1787/8f9e3f98-en

Pearson, H. (2025). Alcohol and cancer risk: what you need to know. Nature, 639(8054), 290–292. https://doi.org/10.1038/D41586-025-00729-5

Pinheiro, L. C., Jumonville, G., Ringel, J., Berg, K. Y., Rosenberg, S., Tonorezos, E., Chandwani, R., Tamimi, R. M., & Phillips, E. (2026). Reevaluating the alcohol-cancer link: long-term cancer mortality outcomes in the REGARDS Study. J Gen Intern Med. https://doi.org/10.1007/S11606-026-10479-3

Rosoff, D.B., Clarke, T.K., Adams, M.J., McIntosh, A.M., Davey Smith, G., Jung, J., & Lohoff, F.W. (2021) Educational attainment impacts drinking behaviors and risk for alcohol dependence: results from a two-sample Mendelian randomization study with ~780,000 participants. Mol Psychiatry, 26(4), 1119-1132.

Stockwell, T., Zhao, J., Clay, J., Levesque, C., Sanger, N., Sherk, A., & Naimi, T. (2024). Why do only some cohort studies find health benefits from low-volume alcohol use? A systematic review and meta-analysis of study characteristics that may bias mortality risk estimates. J Stud Alcohol Drugs, 85(4), 441–452. https://doi.org/10.15288/JSAD.23-00283

Comments on this critique by the International Scientific Forum on Alcohol Research were provided by the following members:

Henk Hendriks, PhD, Independent consultant and partner of the Nutrition Consultants Cooperative, Netherlands

Creina Stockley, PhD, MBA, Independent consultant and Adjunct Senior Lecturer in the School of Agriculture, Food and Wine at the University of Adelaide, Australia

R. Curtis Ellison, MD, Section of Preventive Medicine/Epidemiology, Boston University School of Medicine, Boston, MA, USA

Andrew Waterhouse, PhD, Professor Emeritus of Enology, Department of Viticulture and Enology, University of California, Davis, CA, USA

Fulvio Ursini, MD, Emeritus Professor of Biochemistry, University of Padova, Padova, Italy

James McIntosh, PhD, formerly Professor of Economics, Concordia University, Montreal, Canada

Raquel Romano, PhD, Independent consultant and Professor of Applied Technology at the University of Aconcagua, Argentina

Fulvio Mattivi, MSc, formerly Professor of Food Chemistry at the University of Trento, Italy

Robert Kaplan, PhD, Distinguished Professor and faculty member, Stanford School of Medicine Clinical Excellence Research Center, California, USA

Erik Skovenborg, MD, specialized in family medicine, member of the Scandinavian Medical Alcohol Board, Aarhus, Denmark

Richard Harding, PhD, Formerly Head of Consumer Choice, Food Standards and Special Projects Division, Food Standards Agency, UK

Dominique Lanzmann, MD, PhD, Nutrition/Cardiology, Praticien Hospitalier Hôpital Emile Roux, Paris, France

Giovanni de Gaetano, MD, PhD, Department of Epidemiology and Prevention, IRCCS Istituto Neurologico Mediterraneo NEUROMED, Pozzilli, Italy

---

[1] https://cancerprogressreport.aacr.org/progress/cpr20-contents/cpr20-preventing-cancer-identifying-risk-factors/

[2] https://www.aacr.org/blog/2025/01/08/how-does-alcohol-consumption-impact-cancer-risk/