"The safest level of drinking is none." That was the amazing conclusion of a major newspaper that appeared last week in the Lancet – which warns dozens of news stories that are facing the dangers of even the lowest levels of alcohol consumption.
Researchers wanted to estimate alcohol consumption and exposure to alcohol-related illnesses in 195 countries. So they looked at more than 700 studies from around the world that involved millions of people and explained how safe drinking looks like.
Their reasons for all these problems were clear: Alcohol is a massive health and social problem worldwide. Excessive consumption of alcohol may, over time, increase the risk of liver disease, high blood pressure, injury, memory and mental health problems. It has been a worthwhile effort to understand the global impact of drinking on health ̵
"But while the paper is so beautiful and so useful [at estimating alcohol’s disease burden]," said Stanford Metaforscher John Ioannidis, "it destroys everything at the last moment." Instead of focusing on the message about the dangers of over-drinking "It's about claiming that no drinking is safe."
The data in the paper not only supported a zero drink recommendation, but the authors were also guilty of what too many nutritionists are doing: they used definitive, causal language to talk about correlative studies only. This is something that Ioannidis, a long-standing critic of nutrition science, has recently described as a major source of confusion to the public. In a new paper he argues that the field of nutritional epidemiology needs radical reforms.
The Alcohol Study and related discussion encapsulates everything that is wrong with nutritional epidemiology and the way we talk about it. To avoid being fooled into the future, here's what you need to know.
The Way We Speak About Nutrition Science Is Wrong
Most of what we know about the effects of diet on chronic diseases comes from observational studies. With them, researchers are tracking what a large number of people eat over time and then look at their disease rates, trying to tease out relationships in the data and generate hypotheses for future research. Do people who drink more red wine have fewer heart disease? Does yogurt and nuts have a longer life?
These studies are not controlled as randomized trials in which study participants are randomized into groups and assigned to different interventions. That is, the authors of observational studies can not tell us whether one thing has caused another thing – only that two things are associated .
Observational studies are also riddled with confounders, the unmeasured variables that can actually lead to specific outcomes.
For example: Let's say you want to compare people who drink spirits and beer to wine drinkers. These two groups of people might have other differences at the beginning apart from their choice of wet drink. As we saw with another recent alcohol study, beer and spirit drinkers were rather lower income, men and smokers and had jobs that included manual labor compared to the wine drinkers. They had a higher risk of death and cardiovascular disease than wine drinkers, but were it these lifestyle factors – or just the selection of beer and spirits – that raised their risk of illness? Researchers are trying to control these confounders, but they can not capture all of them.
And even in the most controlled studies on eating and drinking, it is incredibly difficult to find out whether a single nutrient, food, or drink has really led to a particular health outcome.
As Ioannidis points out in his new article, "Individuals consume thousands of chemicals in millions of possible daily combinations." We also prepare our food in thousands of different ways, and if you add something to your diet, you take a different route. To filter out the impact of these variables on health outcomes is "challenging, if not impossible," Ioannidis added.
For this reason, observational studies in nutrition should be hypothesized rather than a source of definitive information about how a single food or nutrient increases or decreases the risk of disease by a certain percentage.
You would not know that if you read the conclusions of nutritional studies and, in particular, much of the reports on nutritional studies. If someone meta-analytically analyzed the evidence from cohort studies (a type of observational research) on different foods, Ioannidis ironically writes:  12 hazelnuts daily (1 ounce) would prolong life by 12 years (ie 1 year per hazelnut) Drinking 3 cups of coffee daily would achieve a similar gain of 12 additional years, and the daily meal of a single tangerine (80 g) would add 5 years of age. Conversely, consuming 1 egg daily would reduce life expectancy by 6 years, and daily consumption of 2 slices of bacon (30 g) would shorten life by a decade, an effect worse than smoking.
All these claims can not possibly be true, yet we often use this kind of causal language to talk about nutritional outcomes.
"Readers and guideline developers may ignore hasty statements about causal inference and advocacy of public policies made by previous articles on nutritional epidemiology," Ioannidis suggested in his new paper. And we – journalists, researchers, policymakers – should avoid making such statements based on nutritional epidemiology as a whole.
What the new alcohol paper really showed
But unlike the spurious conclusions that Ioannidis quotes – about eating a little bacon or an egg will shave years out of your life – heavy drinking has been proven to have terrible health consequences.
"If eggs are bad, even if you eat eight eggs a day, that's not a big deal, but with eight drinks a day, the risk of illness and death is enormous," Ioannidis summarized. And since heavy drinking is socially unacceptable, there is a high level of under-detection and a real chance that the actual burden of alcohol is underestimated.
While no one disputes the harmful effects of heavy drinking, there is a lively debate about what makes healthy, moderate drinking (and concerns about the alcohol industry, which influences research on the benefits of light drinking). But the new Lancet newspaper went much further and made the bold claim that people should not drink, because even a single drink a day is problematic.
To understand why this claim is absurd, take a look at this figure from the study, which shows how the risk of alcohol-related health problems increases by the number of drinks consumed daily:
The authors focused on the risk increase between zero and one drink a day and suggested, "Consuming zero standard drinks daily minimizes the overall risk of all health damage." But you'll see the risk between zero and one, in the bottom left corner of the chart, is virtually indistinguishable. In fact, the risk curve begins to increase after a drink or even 1.5 .
Therefore, it is wrong to claim that people who do not drink on the basis of this data are misleading, said Cecile Janssens, a professor of epidemiology at Emory University, to Vox. "This paper shows that heavy drinking is unhealthy," she said, "but it does not show that zero drinks are the safest."
The most important finding in the paper's press release, and the most prominent on the news The study found that adults who consumed a single alcoholic beverage per day reduced their risk of 23 alcohol-related health problems (from cancer to cardiovascular disease). Illnesses and personal injury) by 0.5 percent compared to non-drinkers.
In the New York Times, Aaron Carroll got this risk under control. A relative increase in risk of 0.5 percent between no drinking and one drinking a day means that four more people in 100,000 a year will have an alcohol-related problem. Here's Carroll:
For every set of 100,000 people who have one drink a day per year, 918 can expect one of 23 alcohol-related problems every year. Of those who do not drink, 914 can expect a problem. … With two drinks a day, the number of subjects rose to 977. Even with five drinks a day, which most agree, the vast majority of people are unaffected.
In other words, statistician David Spiegelhalter estimated that 25,000 people would need to drink 400,000 bottles of gin to experience an additional health problem compared to non-drinkers, "indicating a rather low level of harm to these occasional drinkers."
Again, the difference in health risk between those who do not drink and those who drink daily is tiny. And given the weak observational research on which it is based, it may not make sense.
"To tell people nothing to drink is a stretch," said Ioannidis. "We really need large randomized trial proofs to make such a recommendation, and we have none."
So, now we do not know exactly what threshold alcohol use is risking, but based on this study, it certainly looks like more than zero drinks. The question may be why the authors of the study and the Lancet press office have stretched their findings so far. If I am generous, I would say that they were driven by a desire to draw attention to the great health burden of alcohol. But I believe they also resonated with the public's craving for wild nutritional information – a desire we need to stop feeding.