Claude Gilois, former Senior Chief Scientific Officer, Haematology dept, The National Hospital for Nervous Diseases[1]

Epidemiology makes unrestrained use of statistics. The following quotes help put this discipline in context.


Alcohol consumption and its effects on health is the most studied subject in scientific literature. This is also the subject on which controversy is the most acute, because no double-blind randomized study (the subjects of the study as well as the subjects of the control group are known, nor are the subjects of the study themselves or observers) has ever been authorised by any government. We cannot take the risk of putting the health of subjects at risk during such studies.

It is therefore necessary to carry out indirect, less precise studies[2] which contain statistical analyses which are sometimes chosen to demonstrate the conclusion that researchers want to reach.

For studies on alcohol consumption to be meaningful, they must be carried out on a large number of subjects (tens of thousands) and over a long period of time (several years).


In the early 1990s, the World Health Organization set limits for safe alcohol consumption at 21 units per week for men and 14 units for women[3].

These recommendations were based on the largest study ever carried out on alcohol consumption by the American Cancer Institute which followed 276,000 men for 12 years. This study[i], the results of which are summarized graphically by Professor Serge Renaud[ii] below:  

The results of epidemiological studies are usually interpreted using an index called RR (RR=Relative Risk). An index of one (1) indicates no risk (alcohol consumption has no effect on health). An index of less than one (1) indicates that the risk is reduced (alcohol consumption could be beneficial for health). An index of more than 1 indicates that the risk is increased (alcohol consumption is harmful to health). The further the value is from 1, the greater the risk and the greater the harm. It is customary to express your results in the form of a graph. A graph has two axes. The horizontal axis (x-axis,) is used to represent the independent variable (health risk), while the vertical axis (y-axis,) is used to represent the dependent variable (alcohol consumption expressed in g/l; a glass of wine =10 g/l). On this graph, the RR (Relative Risk) of 1 (zero risk) has been plotted at the zero point (0) and the RR converted to a percentage. This has no effect on the curves. The RRs are reproduced on the graph and the best curve is then obtained. At the time of the study, in 1990, the average daily consumption of a French person was 36 g/l (3.6 units); today it is 27 g/l (2.7 units). The average daily consumption of an American was 0,6 g/l; today it is 2.0 g/l (2 units

The results indicated that the mortality of alcohol drinkers, from all causes, was less than or equal to that of non-drinkers if their consumption did not exceed 36 grams (3.6 units) of alcohol per day or approximately half a bottle. The J curves indicate that the observed effects are dose dependent.

We find these J curves quite often in scientific literature[iii]. They indicate that the negative effect on health does not begin with the first drink and that there is even a threshold above which consumption could be beneficial.

The various  risks are summarised graphically below:

This study was corroborated by another published in 1995[iv] which concluded that mortality amongst moderate drinkers was lower than  those who did not drink.


However, this consensus on the relative harmfulness of alcohol, rather well accepted for two decades, was to be called into question by two studies. One published in 2016[v] and the other in 2022[vi]. The first study concluded: « The results highlight the need to review alcohol control policies and health programs, and to consider recommendations in favour of abstention ». Faced with outcry from the wine industry and disapproval from some scientists, the publication was later corrected: « Our results show that the safest level of alcohol consumption is to not consume alcohol. » This level was in conflict with most health measures advocating the health benefits associated with drinking up to two units per day. In 2022, another article, also published in The Lancet, tempered the first article by stating that this could have some benefit for certain populations, but others were at risk[4].

NB: It is interesting to note that these two publications, emanating essentially from the same group of scientists, concluded with different results, which proves, if necessary, that the statistical interpretation of the results of non-randomized double-blind studies can lead to subjective interpretations.

However, in both studies, we find that there is a dose effect on the harmfulness of consumption (the more you consume, the greater the risk).  This was the case for certain diseases which presented curves in J form (ischemic heart disease, and diabetes in particular) and more generally in the second study which is summarized graphically below:

This curve shows that a consumption of alcohol of up to 20 mg/l (2 units) is beneficial compared to non-drinkers.

Only the dose effect threshold varies (at what dose consumption becomes dangerous for health). While the most recent research places it at 2 units per day (20 g of alcohol per day), the conclusions of the American Cancer Institute study places it at around 3.6 units and cardiologists at around 4 units (40 g/l).

Suggesting abstinence, as the 2016 study does, is a form of alcohol neo-prohibitionism. It shows a worrying drift in certain research. What undoubtedly explains this drift is that the research was not publicly funded, but financed privately (Melinda and Bill Gates Foundation). We know Bill’s penchant for Hamburgers and Coca Cola! Not much of a party animal, geek Bill. Unless this is the return of an old puritanism, as old as the arrival of the first settlers.


1.         That the range or minimum risk consumption is between 2 and 4 units (20-40 g/l) per day, but on the low range for women (see previous article[5]).

2.         That any consumption above this threshold is dependent on the dose (the more you drink, the more you put your health at risk).

3.         Moderate consumption is probably good for the cardiovascular system.

4.         Consumption, even moderate, could induce certain forms of rare cancers (esophagus, respiratory tract, nostrils, nasal cavity, mouth, throat, pharynx).

NB: drinkers are also often smokers and it is difficult to disentangle the effects of these two substances in studies, even if there are sophisticated statistical analyses to try to do so.

5.         The association between breast cancer and moderate alcohol consumption is not currently proven. In women consuming 35 to 44 g per day (3.5-4.5 units, the RR (Relative Risk) is 1.32 (95% confidence interval (CI): 1.19 to 1.45). (Source: National Institute of Health by the British government).

6.         Moderate consumption will not have any effect on morbidity (disease) or mortality (age of death) in the vast majority of subjects.

7.         If you think that moderate consumption makes your life better, you would probably be wrong to deprive yourself.  If this is not the case, it is better to switch to water or non-alcoholic drinks.

[1] Claude Gilois is also a consultant for Vins du Monde (which he created in 1995) and for Valade & Transandine. He is also the owner of Terroirs du Monde and

[2] Cohort studies, case controls and meta-analyses. To know more :

[3] One unit corresponds to 10 cl of wine (10 g of alcohol) at 12.5 o vol or 25 cl of beer at 5o or 3 cl of whiskey at 40 o or a glass of pastis diluted in 3 volumes of water

[4] See article :

[5] See article :

[i] Bofeta P ET Garfinkel L. Alcohol drinking and mortality among men enrolled in an American Cancer Society prospective study. 1990. Epidemiol. 1.342-348.

[ii] Renaud Serge Pr. Le régime crétois.2004. Edition Odile Jacob. ISBN :2-7381-1471-7

[iii] Alcohol and Health: Praise of the J Curves. Giovanni de Gaetano  and Simona Costanzo. J Am Coll Cardiol. 2017 Aug, 70 (8) 923–925

[iv] Gronbaek M, Deis A, Sorensen TIA, Becker U, Schnohr P, Jensen G. Mortality associated with moderate intakes of wine, beer, or spirits. Br Med J 1995;310:1165-1169.

[v] Alcohol use and burden for 195 countries and territories, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016. GBD 2016 Alcohol Collaborators.

[vi] Population-level risks of alcohol consumption by amount, geography, age, sex, and year: a systematic analysis for the Global Burden of Disease Study 2020. GBD 2020 Alcohol Collaborators