In Defense of Moderate Drinking (Again)

Source: New York Times

Aaron E. Carroll

MARCH 23, 2016

A few months ago, I wrote an article here on The Upshot discussing the potential health effects of alcohol. I pointed out that the scientific research said moderate drinking isn’t bad for you; indeed, it might even offer some benefits.

At the same time, I cautioned that this doesn’t mean I’d treat alcohol like medicine, or advise people to start drinking. The take-home message was that if you’re otherwise healthy, and enjoy no more than two drinks a day, you’re unlikely to suffer from it, and potentially might benefit.

However, a new study has been published that is leading some people to write stories proclaiming that “alcohol may not be good for you after all.” Is that the case?

Like any good Bayesian, I use new data and evidence to update my priors. In this case, the recent publication isn’t a new trial or experiment. It’s a new systematic review and meta-analysis that argues that it does a better job than prior work, by excluding a lot of research its authors declared was flawed.

The main problem, they argue, is that a lot of prior work lumps people who used to drink but quit and no longer do so with people who never drank. They further assert that many of these people may actually have quit because they were sick, and were told to give up alcohol. This would mean that the sickest patients, who had been drinking alcohol, were being counted as if they were abstainers, which would bias results in a way to make drinking look healthier than not.

They also only looked at studies that examined all death from all causes considered together. Of the 87 studies they found that met this criterion, only 13 strictly coded lifetime abstainers (and not quitters) as the reference group. When they looked at those studies and compared lifetime abstainers with everyday drinkers, there wasn’t a statistically significant difference. Only those who drank at least 65 grams of alcohol a day (about 4.5 drinks) had an increased risk of death.

The researchers refined their model further, excluding “lesser quality studies,” to consider only seven. The results were unchanged. They then excluded one more study that had results heavily favoring alcohol; the remaining six studies suggested that people who drank two to three drinks a day had a slightly elevated risk of death. But those who drank one to two, or three to 4.5 drinks, didn’t.

When it comes to observational research, that’s a pretty weak finding. But that conclusion is somehow making headlines.

If you look closely, it’s clear the authors excluded a lot of studies that seem to have already taken into account their concerns. For instance, one study I cited looked at all-cause mortality for those 55 to 65 years old. It found that, after controlling for the prior drinkers, as well as other confounders, those who drank a moderate amount had a lower risk of death than abstainers as well as heavy drinkers. It also cited five different earlier studies (1, 2, 3, 4, 5) that accounted for the “former drinker” problem, yet still found the protective effect of light-to-moderate drinking against all-cause mortality. It’s unclear to me why these studies didn’t factor in to the new analysis.

That doesn’t mean the researchers did anything wrong. You’ve always got to make judgment calls about what to include in a meta-analysis. But the researchers don’t clearly explain why they made the judgment calls they did in the text. That makes me concerned they might be cherry-picking.

It’s also important to recognize that this is only a study of all-cause mortality. Many other outcomes exist. A 2011 meta-analysis, for example, looked at mortality from a variety of cardiovascular causes.

This analysis also performed a sub-analysis of studies that classified former drinkers correctly. It found that – with or without this adjustment – active drinkers had both a lower incidence and mortality from a variety of cardiovascular disease. In the same vein, other studies have found associations with alcohol and better cognitive function and lower rates of diabetes. (The story for cancer is much more mixed.)

On top of that, randomized controlled trials suggest that moderate drinking can help with diabetes, blood pressure and cholesterol levels. There’s even a meta-analysis of 63 controlled trials showing alcohol’s positive effects on HDL (good) cholesterol. Authors of articles now declaring alcohol to be benefit-free will need to explain why they ignored those trials.

At best then, this new study shows that for those drinking two or fewer drinks of alcohol a day, there’s no association with higher or lower risks of death. At worst, it’s leaving out trials that show a benefit.

In others words, the study doesn’t make me think I should update my priors much at all. The evidence still says that a moderate amount of alcohol appears to be safe, and that it might even be healthy for many people. There’s nothing in this new analysis that would make me change my mind.