2,335 Pages of SI Say Moderate Drinking In America Is Safe

In this post, I’ll attempt to calculate the correlation between consuming two alcoholic drinks per day and total mortality based on the recent paper that claimed any level of drinking is unsafe. Their main result was a dose response curve where the “response” was diagnosis with any one of 28 conditions that have been found to have some relation, positive or negative, to alcohol. To create a single dose response curve, they have to come up with weights for each one of the 28 conditions. They chose weights based on global averages for how common these conditions are. As an American, the dose response curve relevant to my health is the one weighted based on prevalence in America. Unfortunately, their whopping 2,335 pages of supplementary information (SI) never calculates dose response curves for any single country or group of countries. What they do have are the separate dose response curves for each of the 28 conditions and the total effect of alcohol in the United States.

Screen Shot 2018-09-16 at 5.48.47 PM

This figure is the summary for the US. It needs to be interpreted carefully. The horizontal dashed line is the border between benefit and harm. Everything above the line is the harm from various conditions attributable to alcohol. Everything below the line is harm prevented by alcohol. As you go left-to-right, you move up age groups. It’s important to note that the vertical scale is not an absolute level of harm/benefit, but the relative fraction of harm/benefit for that age group. Most health problems accrue later in life, so if this plot were converted to an absolute scale, the bars towards the right end would grow much larger than the bars towards the left end. That’s important because the benefits of alcohol consumption accrue later in life when we’re protected from heart disease.

The second important note about this plot is that it does not separate moderate and heavy alcohol consumption. We will see that most of the harms accrue to heavy drinkers and most of the benefits accrue to moderate drinkers. Even including heavy drinkers, it looks like for women the total population-level effect of alcohol across a lifetime is quite positive. The protection against heart disease (purple) dwarfs everything else for women, notably including breast cancer (red). The case for men will require closer inspection.

Let’s look at the leading causes of death across all age groups from the CDC and compare them to the plot above.

As everyone knows, heart disease and cancer are the big ones. Below is the dose response curve for ischaemic heart disease. I’m having a hard time finding data separating ischaemic heart disease from all heart disease, but based on the giant purple bars in the plot above, I’m guessing most heart disease deaths are related to ischaemic heart disease. Any expert input here would be much appreciated. It looks like the benefit is roughly a 20% decrease in risk for one or two drinks per day. If about 20% of deaths are due to ischaemic heart disease, then moderate drinking is correlated with about a 4% lower rate of all cause mortality mediated by heart disease.



Before we consider the very broad category of malignant neoplasms (aka cancer) we need to break it down into the many kinds of cancer. I’ll use this chart from cancer.org.


The paper we’re covering today, which was quite exhaustive, linked drinking to the following 7 cancers. In parentheses, I include the fraction of all cancer deaths attributable to that cancer and after the dash I include the approximate relative risk for moderate drinkers.

  1. Breast (0.1% for men, 14% for women) – about 25% at 2 drinks/day
  2. Colorectal (8.3%) – about 20% at 2 drinks/day
  3. Esophageal (2.6%) – about 40% at 2 drinks/day
  4. Larynx (0.6%) – about 30% at 2 drinks/day
  5. Lip and oral cavity (1.6%, bundled with pharynx and nasopharynx) – about 75% at 2 drinks/day
  6. Liver (4.9%) – about 20% at 2 drinks/day
  7. Pharynx and nasopharynx – about 100% at 2 drinks/day

The total percentages of all cancer are 18% for men and 32% for women. I won’t bother posting dose response curves for all seven cancers. Instead, I’ll read off the effect sizes at 2 drinks/day as best I can and weight the cancers by their prevalence. I get a weighted average risk increase of about 30% for men and 27% for women. The total increase in cancer risk is then 5.4% for men and 8.6% for women. For total mortality we get a 1.2% increase for men and a 1.9% increase for women. Obviously, this is very rough and involves me guessing some numbers off of plots. The point is, these effect sizes are half or less of the protection against heart disease.  

We’ve now covered 45% of deaths in the United States. The third leading cause of death, chronic lower respiratory disease, is not alleged to have a link to alcohol (this effect is distinct from pneumonia and lower respiratory infections, which are linked to alcohol).

The CDC has one category for “accidents (unintentional injuries)” whereas this paper separates “unintentional injuries” from “transport injuries.” Both have an effect size of about 20% at 2 drinks/day, so I’ll merge them and multiply by 5.4% to get a 1.1% increase in total mortality.

Hemorrhagic stroke has a roughly 10% increase at 2 drinks/day (200k US cases per year). Ischaemic stroke decrease about 5% for men and 20% for women. According to the CDC, 87% of all strokes are ischemic strokes. I’ll treat the effect on hemorrhagic stroke as a rounding error, then. Assuming the baseline risks of stroke death are the same for men and women, 2 drinks per day is correlated with a total mortality decrease of 0.2% for men and 0.9% for women, mediated by stroke risk.

The paper makes no claim that alcohol has an impact on Alzheimer’s or dementia.

Next up is diabetes at 2.9% of all deaths. At 2 drinks/day, women have about a 25% decreased risk and men have about a 5% decreased risk. That gives a total mortality decrease of 0.7% for women and a 0.1% decrease for men, mediated by diabetes.

At 2 drinks/day, alcohol has essentially no impact on lower respiratory infections (which I’ll assume we can connect to “influenza and pneumonia” in the CDC table).

There’s no claimed on nephrosis.

Self-harm increases about 25% at 2 drinks/day, for a 0.4% increase in total mortality.

There is essentially no risk of cirrhosis at 2 drinks/day.

We’ve now covered the 15 leading causes of death. If I sum up the effects, I get that consuming 2 drinks/day correlates with a decreased total mortality risk of 1.6% for men and 2.2% for women. This calculation is very much back of the envelope. I’ve matched CDC mortality tables with the slightly different definitions in the paper as best I can and have haphazardly considered differences between men and women. The point is, the data in this paper still suggest that 2 drinks/day for Americans is correlated with slightly lower all cause mortality. In my next post, I’ll discuss other meta-studies that find overall health benefits for moderate drinking.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s