Link to the Lancet article shown above. All images in this post are from this article.

Many people have taken comfort from news reports suggesting that moderate drinking is healthy. The top three graphs in the panel below, from a recent study of hundreds of thousands of Chinese drinkers and non-drinkers, confirm the kind of stylized facts that have led people to believe this. As the authors of “Conventional and genetic evidence on alcohol and vascular disease aetiology: a prospective study of 500 000 men and women in China” describe these results,

… conventional epidemiology showed that self-reported alcohol intake had U-shaped associations with the incidence of ischaemic stroke (n=14 930), intracerebral haemorrhage (n=3496), and acute myocardial infarction (n=2958); men who reported drinking about 100 g of alcohol per week (one to two drinks per day) had lower risks of all three diseases than non-drinkers or heavier drinkers.

But, one should be concerned that the U-shape in the top three graphs is generated by some combination of reverse causality and “Cousin Causality.” In the words of the authors (click here for the full paper, then click on the link to show all authors),

… poor health might affect alcohol consumption (reverse causality), and other systematic differences might exist between people with different drinking patterns that were not fully allowed for (residual confounding).

Fortunately for statistical identification, there are a pair of genes that have a big effect on drinking through well-understood biological pathways. Alcohol itself is mostly pleasant, but is ultimately turned into acetaldehyde, which is unpleasant until the acetaldehyde is in turn broken down into acetate, which is OK. Two gene variants in East Asia lead to a lot of acetaldehyde: one slows down the process of turning the unpleasant acetaldehyde (which causes flushing) into acetate; the other turns alcohol into acetaldehyde faster. Both make drinking less pleasant. Just below is the top of the explanatory box in the paper for this causal pathway:

Assuming these genes have no other substantial affect besides making alcohol consumption unpleasant, they can be used as instruments for alcohol consumption, allowing one to see the effects of alcohol without confounding from reverse causality or “Cousin Causality.” The relevance of these genes as statistical instruments can be seen from the graph below:

The authors have relative risk on the left-hand-side of the equation, put region of China as a control variable on the right-hand side of the regression, then effectively use region and genes as instruments for region and alcohol consumption. They do the analysis separately for men and for women. Since Chinese women in this study do not drink much regardless of their genes, the analysis for women is a good test of whether the genes affect health through pathways other than alcohol consumption. After correcting for multiple hypothesis testing (See “Who Leaves Mormonism?”), no effects of the genes are seen for women.

The results for six levels of alcohol consumption as predicted by region and genes can be seen below for men and women:

This evidence is quite persuasive. It seems unlikely that tinkering with the analysis in any appropriate way would yield a different result. Translating the outcome descriptions into simpler words, the analysis says that higher alcohol consumption in men leads causally to:

• higher blood pressure

• higher “good cholesterol”

• a marker for liver disease

• stroke

• brain bleeding

There is no evidence of heart attack danger from drinking, but no evidence of protection from heart attack either.

Note that, while women help to verify that the gene variants don’t have big effects unrelated to alcohol, the low variation in alcohol consumption among these Chinese women (because few of them drink very much) means that there is little direct evidence here on the effects of alcohol on women. But many of the health effects of alcohol on women are likely to be similar to the health effects of alcohol on men.

The only shred of evidence for a physical health benefit from alcohol consumption is its effect in raising “good cholesterol.” But in their appendix, the authors give the first-stage regression of many health-related measures on the genes, and the news there is not good. For example, gene variant GG, which is associated with higher alcohol consumption than gene variant AG, fairly precisely predicts a body-mass index about .24 higher. Here is their full table for that first-stage regression on the genes:

Let me mention in closing that alcohol doesn’t seem to do its damage by causing an insulin spike. The insulin index data I discuss in “Forget Calorie Counting; It's the Insulin Index, Stupid” gives a very low insulin index (less than five) to white wine and gin, and even beer only has a measured insulin index of 20. So the health harm from alcohol seems to come from other pathways, not from causing an insulin spike that would make you feel hungry afterwards.

For annotated links to other posts on diet and health, see “Miles Kimball on Diet and Health: A Reader's Guide.”