So here goes. If I were writing the dietary guidelines, I would give them a radical overhaul. I’d go so far as to radically overhaul the way we evaluate diet. Here’s why and how.
The reason we know so little about what to eat despite decades of research is that our tools are woefully inadequate. Lately, as scientists try, and fail, to reproduce results, all of science is taking a hard look at funding biases, statistical shenanigans and groupthink. All that criticism, and then some, applies to nutrition.
Prominent in the charge to change the way we do science is John Ioannidis, professor of health research and policy at Stanford University. In 2005, he published “Why Most Research Findings Are False” in the journal PLOS Medicine, and he has been making science headlines (although not always friends) ever since. He came down hard on nutrition in a pull-no-punches 2013 British Medical Journal editorial titled, “Implausible results in human nutrition research,” in which he noted, “Almost every single nutrient imaginable has peer reviewed publications associating it with almost any outcome.”
Ioannidis told me that sussing out the connection between diet and health — nutritional epidemiology — is enormously challenging, and “the tools that we’re throwing at the problem are not commensurate with the complexity and difficulty of the problem.” The biggest of those tools is observational research, in which we collect data on what people eat, and track what happens to them.
The trouble begins with that “collect data” part. There are a few ways to do this, none of them particularly good. You can use a 24-hour recall, which gives respondents a fighting chance of remembering what they actually ate but doesn’t give you a representative sample of overall diet. Food diaries over a long period do that better but people tend to eat differently when they’re tracking their diet for researchers. Most large population studies use food frequency questionnaires (FFQs, in industry lingo), where they ask people to count up the servings they’ve eaten of a wide range of foods, often over the course of a year.
There’s no better way to understand the shortcomings of an FFQ than to fill one out. Maybe you know how often you ate pie last year, but do you know how often you ate “foods with oils added or with oils used in cooking (do not include baked goods or salads)”? A host of studies of self-reported data have found that up to two-thirds of respondents report eating a diet so inconsistent with their caloric needs as to be implausible.
Give tens of thousands of people that FFQ, and you end up with a ginormous repository of possible correlations. You can zero in on a vitamin, macronutrient or food, and go to town. But not only are you starting with flawed data, you’ve got a zillion possible confounding variables — dietary, demographic, socioeconomic. I’ve heard statisticians call it “noise mining,” and Ioannidis is equally skeptical. “With this type of data, you can get any result you want,” he said. “You can align it to your beliefs.”
Ah, beliefs. Just about every week there’s a new study of a food funded by the people who profit by it. (New York University’s Marion Nestle has been tracking this for years; her 2018 book “Unsavory Truth” details her findings.) But funding bias isn’t the only kind. “Fanatical opinions abound in nutrition,” Ioannidis wrote in 2013, and those have bias power too.
So what do we do about this? “Definitive solutions won’t come from another million observational papers or small randomized trials,” reads the subtitle of Ioannidis’s paper. His is a burn-down-the-house ethos.
Frank Hu lives in the house and is understandably less enthusiastic about the incendiary approach. He chairs the department of nutrition at Harvard’s T.H. Chan School of Public Health, arguably ground zero of nutritional epidemiology. While he acknowledges shortcomings of the research in his field, and is respectful of Ioannidis’s criticism, he says nutrition researchers have brought much to our understanding of a healthful diet and can address the problems.
He pointed out that data collection is improving, with new tools to better assess diet and reality checks with measurable biomarkers (testing the urine of respondents in a salt study for sodium, for example). And he doesn’t believe biases undermine the credibility of the field. Often, they cancel each other out, he says, and the most authoritative recommendations, such as the dietary guidelines (he was on the 2015 committee), are the consensus of large groups looking at the preponderance of evidence. Still, he acknowledges that there’s work to be done. “If we don’t have challenges, our life will be very boring.”
When it comes to actual dietary recommendations, the disagreement is stark. “Ioannidis and others say we have no clue, the science is so bad that we don’t know anything,” Hu told me. “I think that’s completely bogus. We know a lot about the basic elements of a healthy diet.” He lists plant-based foods — fruit, veg, whole grains, legumes — but acknowledges that we don’t understand enough to prescribe specific combinations or numbers of servings. The ongoing controversy, he says, has generated “a lot of heat but not much light,” and he’s afraid Ioannidis’s dismissal of the entire field undermines nutritional advice.
But if nutritional advice is unsupported, a little undermining is in order. Ioannidis is in favor of fruits, vegetables and whole grains, but “the evidence behind them is pretty soft,” he wrote in an email. Older observational studies showed big reductions in cancer risk but newer studies show small benefits, if any. “When the benefit in published studies in the literature shrinks 10-fold or 100-fold over time,” he continued, “you have every reason to worry about whether this type of research effort can give you any reliable answers.” Heart disease risk reduction has remained sizable, Ioannidis noted, but it’s still observational data, and confounding and data reporting issues mean we can’t definitively link diet to health, a point Hu makes in his own research.
Big differences in what people eat track with other differences. Heavy plant-eaters are different from, say, heavy meat-eaters in all kinds of ways (income, education, physical activity, BMI). Red meat consumption correlates with increased risk of dying in an accident as much as dying from heart disease. The amount of faith we put in observational studies is a judgment call.
In the two decades I’ve been writing about nutrition, my confidence in what we know about food and health has eroded, and I find myself in Ioannidis’s camp. What have we learned, unequivocally enough to build a consensus in the nutrition community, about how diet affects health? Well, trans-fats are bad. Anything else, and you get pushback from one camp or another.
And then there’s eggs, poster food for we-don’t-know-jack-about-diet. We used to think they were bad, because their cholesterol content contributed to heart disease. Then they were exonerated. Eggs are okay! And just last week, a new study came out saying not so fast, they might be bad after all. Let the eye-rolling begin.
Over and over, large population studies get sliced and diced, and it’s all but impossible to figure out what’s signal and what’s noise. Researchers try to do that with controlled trials to test the connections, but those have issues too. They’re expensive, so they’re usually small and short-term. People have trouble sticking to the diet being studied. And scientists are generally looking for what they call “surrogate endpoints,” like increased cholesterol rather than death from heart disease, since it’s impractical to keep a trial going until people die. While I hold out hope that we’ll get better at all this, it’s going to take a while.
Meantime, what do we do? Hu and Ioannidis actually have similar suggestions. For starters, they both think we should be looking at dietary patterns rather than single foods or nutrients. They also both want to look across the data sets. Ioannidis emphasizes transparency. He wants to open data to the world and analyze all the data sets in the same way to see if “any signals survive.” Hu is more cautious (partly to safeguard confidentiality) but does believe wider access to data and checking results against multiple data sets will help identify genuine effects.
I don’t think anyone would be against this — I’m certainly not — but remember those FFQs? You’re still working with flawed data, and I am not optimistic that we’ll get much actionable advice out of the effort. Neither is Ioannidis. When I asked him whether that approach would be more likely to solve the dietary advice problem or tell us how little we know, he said “probably the latter.”
The important question — what are we supposed to eat already?! — is still on the table, and I have a suggestion. Let’s give up on evidence-based eating. It’s given us nothing but trouble and strife. Our tools can’t find any but the most obvious links between food and health, and we’ve found those already. Instead, let’s acknowledge the uncertainty and eat to hedge against what we don’t know. We’ve got two excellent hedges: variety and foods with nutrients intact (which describes such diets as the Mediterranean, touted by researchers). If you severely limit your foods (vegan, keto), you might miss out on something. Ditto if you eat foods with little nutritional value (sugar, refined grains). Oh, and pay attention to the two things we can say with certainty: Keep your weight down, and exercise.
When I first started writing about nutrition, I used to say I could tell you everything important about diet in 60 seconds. Over the years, my spiel got shorter and shorter as truisms fell by the wayside, and my confidence waned in a field where we know less, rather than more, over time. I’m down to five seconds now: Eat a wide variety of foods with their nutrients intact, keep your weight down and get some exercise.
Oh, and playing well with others is highly overrated.
More from Food: