The risks and benefits of moderate drinking and the journalistic hazards of epidemiology

The NYT has a piece (gift link) on the controversy surrounding whether and to what extent the American public health establishment ought to revise its current recommendations in regard to drinking alcoholic beverages.

The most interesting thing to me about this issue is how it’s framed:

A report that is intended to shape the next edition of the U.S. Dietary Guidelines has broken sharply with an emerging scientific consensus that alcohol has no health benefits.

The evidence review, by the National Academies of Sciences, Engineering, and Medicine in December, revived a once-dominant hypothesis that moderate drinking is linked to fewer heart attack and stroke deaths, and fewer deaths overall, compared with never drinking.

Many scientists now take issue with that view. And some fear that, based on the new analysis, the influential dietary guidelines may fail to address recent research into the harms of drinking.

It would be altogether remarkable if such a widespread social practice as drinking alcohol didn’t have both beneficial and detrimental effects on those who engage in it, whether one chooses to label those as “health” effects or not, and there’s every reason to believe that it does.

“This report is a thinly veiled effort to undo the growing evidence that alcohol causes cancer and is increasingly associated with serious health outcomes,” said Diane Riibe, who co-founded the U.S. Alcohol Policy Alliance, a nonprofit focused on the harms of alcohol.

The report did note a small but significantly heightened risk of breast cancer associated with moderate drinking, but said there wasn’t enough evidence to link moderate consumption to other cancers. The National Cancer Institute, among other scientific bodies, disagrees.

Hmmm. Experts are disagreeing, but the key thing here is to understand exactly what they’re disagreeing about. What they’re disagreeing about is this: Do the relative health risks associated with moderate drinking — there’s no doubt that heavy drinking is bad for peoples’ health — outweigh the relative health benefits associated with it?

This is a difficult question to answer, but why it’s difficult to answer is much more significant, in terms of social policy, than what the precise answer might be.

The way the Times’s reporter discusses this issue is revealing:

After a year of study, the National Academies panel found that there was sufficient evidence to draw only three conclusions with some certainty.

They were that a pattern of moderate drinking was associated with 18 percent fewer cardiovascular disease deaths, with a 16 percent lower risk of all-cause mortality, and with a 10 percent heightened risk of breast cancer for women.

Are there reasons to be skeptical that the small positive statistical changes in relative risk from observational studies associated with moderate drinking indicate a causal relationship between the moderate drinking and the improved health outcomes? There certainly are, and the Times points out why:

“All of this is based on observational studies,” which can turn up correlations between moderate drinking and health outcomes but do not demonstrate a cause-and-effect relationship, said Dr. Bruce N. Calonge, an epidemiologist at University of Colorado Denver, who chaired the National Academies panel. . . .

That research was observational, like all of the studies that were examined. There are few randomized controlled trials of moderate drinking, so it is difficult to know why moderate drinkers might have fewer heart attacks and strokes.

Their better cardiovascular health may not be a result of the alcohol they consume, but of a slew of other characteristics, experts say.

People who drink in moderation may be moderate in other ways, eating well and engaging in other healthy behaviors, and they tend to be more educated and have socioeconomic advantages — all factors associated with good overall health.

There are two big statistical problems here.

The first is that a small percentage increase over a very low baseline risk still means a very low risk. For example, suppose you follow 10,000 middle-aged (35-54) moderate drinkers and 10,000 middle-aged non-drinkers for five years. In the population as a whole, you would expect to see about 150 deaths among 10,000 randomly selected Americans of that age over a five-year period. You would expect about 25 of those deaths to be from cardiovascular disease. So let’s say that your study finds 22 deaths from CVD in the moderate drinking group (again this is among 10,000 middle-aged people over five years), and 27 deaths from CVD among the non-drinkers. Those five extra deaths among 10,000 people over five years represent a 20% increase in the relative risk of death from CVD among the non-drinkers relative to the moderate drinkers.

The second big statistical problem is that it’s to say the least unclear that those five extra deaths were caused by the non-drinking of the non-drinkers, as opposed to by one or more of the countless statistically uncontrolled differences between the people in the two groups.

But now we get to the bait and switch that’s so common in these sorts of articles. And that there’s no acknowledgement anywhere in the piece that these perfectly valid criticisms of drawing strong statistical conclusions from weak evidence, in the form of small percentage changes in relative risk in observational studies, also apply, and in exactly the same way, to observational studies finding an association between, for example, cancer and moderate drinking, or increased all-cause mortality, etc.

The anti-drinking zealots, in other words, are perfectly happy to point out the very serious limitations of observational studies when doing so supports their views, but instantly forget all about these very same methodological limitations when they quote exactly the same sorts of statistics to buttress their message that moderate drinking is supposedly so bad for your health that the US government should come out against it as a matter of public health policy.

And the Times’s story on this issue is a perfect illustration of this.