A new study of alcohol consumption has gotten a lot of headlines claiming moderate drinking, net net, is harmful. In this post I’ll try to parse through some more precise claims about what the paper does and doesn’t show. The paper has hundreds of pages of appendices and lots of details you’d normally find in the main text are pushed there. I’ll aim to make a second post when I’ve gone through some of the key details missing from the main text.
Their data say moderate drinking is safe
Most importantly, the paper actually shows exactly the opposite of the headline grabbing claim that “No amount of alcohol is safe,” (CNBC headline). In CNBC’s defense, the sloppily written conclusion of the main article does say, “the safest level of drinking is none.” An editorial accompanying the research article has the more carefully chosen title, “No level of alcohol consumption improves health.” The study’s final summary figure, above, in fact clearly shows no harm for people who drink up to 0.8 drinks per day. That’s not due to a lack of data for people who drink very little. Rather, their data clearly show no harm for drinking up to that level. This result does differ from the previously accepted view that women benefit from one drink per day and men benefit from two drinks per day, but I’ll argue that was due to some unusual methodological choices that make this paper relevant for policy makers but not for doctors or individuals deciding whether up to 1-2 drinks per day is safe.
N of 0
Some papers get published with an “n of 1” when they find some unusual anecdote in a single patient.
This paper actually does not link any outcome to any single person. Instead, the primary and valuable contribution of this paper is to make detailed estimates of the distribution of alcohol consumption in every country on earth. It turns out that in nearly all countries there are a meaningful number of alcoholics and in some countries the average level of drinking qualifies as alcoholism. This fact is very important to know if you’re a legislator deciding whether to support new alcohol taxes. The authors very persuasively argue in favor of such policies. The bulk of the paper is essentially a census of all the damage alcoholism does to the world. However, the paper only links specific levels of drinking to specific health outcomes by comparing the drinking distribution to the percentage of people who die from each condition in every country. They have no idea who did the drinking and who did the dying. This choice allows the authors to cover every country, but it prevents them from saying anything remotely precise about how helpful or harmful it is to have 1-2 drinks per day.
[From the first appendix, the authors actually do take data from prior studies that look at individuals. This paper doesn’t pool the individual’s outcomes though. Somehow they reduce each prior paper to a single point on their dose response curve even if that study covered a range of drinking. I’m not clear yet on how this works.]
You’re not going to get tuberculosis
The figure that shows no benefit to drinking up to 0.8 drinks per day is the dose response curve in the average country. But nobody lives in the average country. It turns out that the leading cause of death from drinking too much is tuberculosis. The authors find countries with high levels of moderate drinking have higher levels of tuberculosis mortality. Besides the many possible confounders that could nullify this result, it is simply not relevant if you live in a country where nobody dies from TB.
The authors do replicate the widely accepted link between moderate alcohol consumption and decreased cardiovascular mortality. I expect that if you used their data to make a dose response curve for wealthy countries with non-existent TB and widespread heart disease you’d recover the previous result that moderate drinking is healthy. I’ll try to find this plot somewhere in the appendix for a follow-up post.
Other Experts Respond
My favorite evidence-based medicine writer is Aaron Carroll. I was careful to read this paper myself before reading his response. He is far more negative than I am. His complaints are 1) my first complaint that their data show the opposite of what they claim. 2) Although they find that one drink per day is harmful, the number needed to harm (that is, the number of people who would need to be exposed to the hazard before one person is harmed who otherwise wouldn’t have been) is 25,000. Higher up the drinking spectrum, he tweets, “At five alcoholic drinks per day, WHICH NO ONE THINKS IS A GOOD IDEA, the number needed to harm each year is 296.”
I’ll add that their main figure runs up to 15 drinks per day, so 13/15ths of the figure covers a range that everyone has always thought was unhealthy for men. The most surprising thing to me is how small the effect sizes are at 15 drinks/day. They actually use a low definition of one drink, so their 15 is more like 11 standard drinks. Still, eyeballing this on my phone while riding BART it looks like the number needed to harm at 11 standard drinks per day is about 50. That means you can be completely wasted every day of your adult life and there’s about a 98% chance
you’ll die of something unrelated to your alcoholism you won’t get any of these conditions. [Updated because this relative risk curve isn’t for all-cause mortality.]
Time Magazine’s coverage quotes a professor from the Harvard School of Public Health saying that the strong evidence linking moderate drinking to hearth health are more important for Americans than the link to tuberculosis.
To be clear, I’m not recommending anybody start drinking. There’s no telling which non-drinkers would become alcoholics if they started drinking. But if you’re already a moderate drinker, this paper is no reason to change. This paper adds to every paper ever saying alcoholism is bad for you and convinces me we should raise alcohol taxes to reduce it.