What does drinking do to your health? We can say two things with confidence:

Drinking is associated with lots of health problems.

Heavy drinking is bad for you.

Here’s a graph of some associations:

Someone who averages 10 drinks per day is 50 times more likely to get cirrhosis than someone who doesn’t drink at all (controlling for age, sex, and drinking history).

This looks bad, but there are two caveats. First, it doesn’t establish causality. It could be — if all you had was this figure — that cirrhosis causes hormonal changes that in turn create the urge to drink more.

But we do know that heavy drinking is bad. That’s partly because we know how alcohol causes problems. It causes cirrhosis by destroying liver cells. It causes cancer by getting converted to acetaldehyde and then damaging DNA. There are also randomized controlled trials (RCTs) that take heavy drinkers and get them to drink less. These inevitably show improved health (either health outcomes or biomarkers like blood pressure).

The second caveat is the little dip in relative risk for diabetes and heart disease around 1-2 drinks. Some people think alcohol is causing this dip. Lots of mechanisms have been proposed: Maybe it reduces inflammation. Or maybe it impairs the cells that build up plaques in arteries. Or maybe it creates a hormonal imbalance that changes blood pressure regulation. Or maybe it increases HDL cholesterol or insulin sensitivity or adiponectin levels.

Or, maybe alcohol doesn’t help diabetes and heart disease at all. Mathews et al. (2015)
tried to model how alcohol affects the heart, ending up with this terrifyingly tangled figure.

Alcohol does a lot of different things and interacts with a lot of other factors. It’s great to try to unravel all this, but I don’t trust anyone who says they understand everything with confidence.

If alcohol doesn’t improve heart health, then why the dip? Well, it could just be that the same people who drink moderately are also more likely to exercise and eat well.

So we don’t know if moderate drinking is bad for you. It almost certainly causes harms like cancer, but it might help heart disease enough to offset those harms. In the US, around 20% of adults drink 1-2 drinks per day. Even if the effects are modest, the collective impact is huge. Second perhaps to caffeine, alcohol is humanity’s favorite drug. We need to know what it does.

This is the story of a trial that came close to answering this question and then exploded. At first, this looks like a simple story of corruption, but when you look closely, it’s a very complicated story of corruption.

You might be thinking, “What we need to do is compare the health of people who drink different amounts, while controlling for income, diet, education, exercise, et cetera.” The problem is that “controlling” for things is a dangerous business. It requires tons of different assumptions, like what you control for, how you code stuff, and how you model everything. For example, if you “control for exercise,” do you measure the number of hours people exercise each week? Should you distinguish different kinds of exercise? Reasonable people can disagree about these choices. For alcohol, reasonable people do disagree. Some, like Ronksley et al. (2011)
find a strong association between moderate alcohol consumption and improved cardiovascular health, and argue that the association is likely causal. But others, like Goel et al. (2018)[^S. Goel, A. Sharma, & A. Garg, “Effect of Alcohol Consumption on Cardiovascular Health,” Current Cardiology Reports 20, no. 19 (2018).] are unconvinced and suggest there could still be confounding variables, while Wood et al. (2018)
do a meta-analysis of observational studies that suggests even small amounts of alcohol hurt cardiovascular health.

There’s also some recent research using Mendelian randomization, which suggests alcohol could be bad for cardiovascular health. The idea is that a variant of the ADH1B gene makes it hard to metabolize alcohol. People who have it drink less. If you assume that the gene is random in the population and that it’s causing reduced drinking, then you can treat it like a random assignment to drink less. Holmes et al. (2014)
did this and found that carriers of ADH1B had better cardiovascular health by every measure. This suggests alcohol makes cardiovascular disease worse, not better.

So what do we do? We take the long, slow, hard path:

1. Get a large group of people.

2. Tell some of them to drink moderately, tell the others not to drink at all.

3. Wait years, monitoring people to make sure they are actually drinking (or not) like they’re supposed to.

4. Follow up and see which group is healthier.

Lots of things make this difficult. Because the expected effects aren’t huge, you need a large group of people. Because culture and genetics vary, you need people from around the world. Because diseases take a long time to show up, you need to wait years. And imagine the challenge of telling people how much to drink and then making sure they follow your instructions.

An international effort monitoring thousands of people around the world for years — does that sound expensive?

Back in 2013, the NIH’s National Institute on Alcohol Abuse and Alcoholism (NIAAA) got interested in funding this. They figured it would cost on the order of $100 million for the full trial. This doesn’t seem crazy given the NIAAA’s $500 million annual budget, but the NIAAA has lots of other priorities and didn’t feel they had the money.

You know who has a lot of money, though? The alcohol industry. Worldwide, $85 million of booze is sold every 30 minutes. In principle, the industry could directly fund a study, but who would trust it?

In 2016, it looked like the NIAAA had found an elegant solution:

• Five alcohol companies would donate money for a trial.

• The NIH would ask researchers to send proposals for how they’d run a trial.

• The NIH would choose the scientifically best proposal, just like they do with any government-funded grant. The donors would have no influence on the process.

• To make the results trustworthy, there would be a “firewall”, with no communication between the industry and the research team.

Sounds promising. But if we go forward a couple of years, everything suddenly blows up.

June 15, 2018

What happened? You might imagine banal corruption, with cocaine and overseas bank accounts, but it’s nothing like that.

The real story is a much more interesting cocktail of science, academia, bureaucratic maneuvering, ambition, politics, capitalism, the “deep state,” secret emails, and slippery ethical slopes. It’s a huge stroke of luck that we know about any of this. You have to ask how often similar things happen and don’t blow up.

If you’re brave, you can read the 165-page report the NIH prepared before canceling the program. But I warn you: it’s mostly out-of-order redacted emails written by people who wanted to conceal what was happening. There’s an executive summary, but it’s written in a frustratingly bureaucratic style. There are also newspaper stories, but they don’t try to give the full timeline.

After way too much time reconstructing things, here’s the full story as best as I can tell.

2001-2013. Kenneth Mukamal, a physician at Beth Israel Deaconess Medical Center and faculty member at Harvard Medical School, published many papers that argue that moderate alcohol consumption has health benefits, usually for heart disease or diabetes. During the same period, John Krystal, a psychiatrist and professor at Yale, also published many papers on alcohol, mostly focusing on addiction and mental health. (Many other researchers were involved in this study, but these two were most prominent.)

Here’s a characteristic sample of Mukamal’s 189 papers on alcohol:

In summary, all of this evidence implicates alcohol consumption rather than lifestyle factors … as the primary factor in the lower rates of cardiovascular disease found among moderate drinkers. (2001)

In this large cohort study of older adults, there was a lower risk of congestive heart failure associated with moderate drinking compared with abstention. (2006)

There is convincing evidence that light-moderate, non-binge alcohol intake reduces the risk of coronary heart disease. (2009)

In 9 nationally representative samples of U.S. adults, light and moderate alcohol consumption were inversely associated with cardiovascular disease mortality, even when compared with lifetime abstainers. (2010)

Long-term moderate alcohol consumption is inversely associated with all-cause and cardiovascular mortality among men who survived a first myocardial infarction. (2012)

You may notice that all of them find that moderate drinking has health benefits. 

Early 2013. Some NIAAA staff were convinced that moderate drinking is good for you, and an RCT could prove it conclusively enough that doctors might recommend it to patients like they do with aspirin now. They had the idea of getting the alcohol industry to fund the study, but faced two problems. First, the alcohol industry wants lots of details before forking over any cash. Second, the NIAAA isn’t allowed to solicit from industry. They tried to get around these problems by having outside researchers (including Mukamal and Krystal) meet with industry to give details on how such a trial might work. This created a dynamic where everyone (the NIAAA, the alcohol industry, Mukamal) wanted to coordinate with each other, but maintain a pretense of being isolated. There was lots of scheming about how information should flow to maintain this pretense.

The NIAAA wanted someone else to present the idea of the study to overcome their prohibition of solicitation, even though they’d obviously set this whole thing in motion. The alcohol industry was excited about what they heard directly from the researcher, but wanted the plan to come “from NIAAA.”

July 12, 2013. The NIAAA published NOT-AA-13-004. This was a “planning grant,” which basically means that the NIH would give you some money to do some work that would allow you to successfully submit a much larger grant soon. By NIH rules, this was a public opportunity, meaning any researcher could submit and win the grant if they had the best science. Yet they obviously wanted “their” PI to win: