The Real Reason Ultraprocessed Foods Are Ruining Our Health

This transcript has been edited for clarity. 

Welcome to Impact Factor, your weekly dose of commentary on a new medical study. I’m Dr F. Perry Wilson from the Yale School of Medicine.

If you have any interest at all in nutrition and have been conscious over the past 5 years or so, you are no doubt aware of a torrent of literature blaming a lot of health woes on ultraprocessed foods. 

If I were to show you some foods, for the most part you’d have a pretty good sense of whether they were ultraprocessed or not. You know what? Let's try it.

The classic Oscar Mayer wiener? Ultraprocessed. It contains multiple additives like sodium lactate, dextrose, and glucose, and “mechanically separated meat.”

How about a handful of almonds? 

Not ultraprocessed. In fact, on the standard NOVA processing scale— which goes from a score of 1 for unprocessed or minimally processed foods to 4 (ultraprocessed) — a nice handful of almonds is a 1. 

Doritos? Ultraprocessed. 

Tostitos? Only “processed.” Containing just corn, canola oil, and salt, Tostitos scores a 3 on the NOVA scale. A good example of a 2 on the NOVA scale is California Olive Ranch olive oil. You need to process olives to turn them into oil, but there’s not much more going on than that. 

This can be called the Potter Stewart method of classifying foods. You may know this story. Stewart, a Supreme Court Justice from 1958 to 1981, famously weighed in on a censorship case. The question was whether a particularly racy French movie could be banned because it was hardcore pornography. He admitted that it was hard to come up with a legal definition of hardcore pornography but “I know it when I see it.” 

And so it is with ultraprocessed foods. In general, we have a sense of what they are, but you might be hard-pressed to come up with a universal definition. And you may be surprised. For example, most almond milks are technically ultraprocessed foods, whereas most regular milk isn’t. In other words, ultraprocessed doesn’t always mean less healthy. But for the most part, ultraprocessed foods are worse for you than their less-processed alternatives. Still, up until now, deciding what foods are ultraprocessed was, honestly, somewhat subjective. 

But there’s a new tool available that takes the guesswork out of this process. It will be crucial for researchers, but also for consumers trying to eat less-processed foods. That’s because in general, ultraprocessed foods are really bad for you, but maybe not entirely for the reason you think.

This is how the standard NOVA system classifies an ultraprocessed food:

Lots of food products don’t fit neatly into this category, though. To fix this, enter Giulia Menichetti and colleagues, writing in Nature Food, developed a machine-learning model that can ingest any ingredient list — like the kind that appears on virtually every packaged item you buy in the grocery store — and spit out a food processing (FPro) score that tells you exactly how processed this specific food is.

They developed the algorithm by training it on a database of foods that had been painstakingly categorized into the four NOVA categories by humans. But the nice thing about machine-learning models is that you can apply them to data they haven’t seen before. And that’s exactly what they did. 

Let me give you an example.

The score ranges from 0 (raw foods, basically) to 1, the most processed thing possible. Yes, I did dig into their actual database to see what specific food products bookended the research. At a score of 0 is organic ground beef, and all the way at 0.999056, it’s a tie between Wonder White Hamburger Buns and Ball Park Everything Hamburger Buns. So, I guess go bunless on your next burger.

Less extreme examples are two cheesecakes.

On the left we have Edwards Desserts Original Whipped Cheesecake and on the right Pearl River Mini No Sugar Added Cheesecake. Potter Stewart doesn’t help me as much here, although the long ingredient list on the left is a bit of a red flag. They both strike me as probably very processed. But the algorithm gives us a number, an actual score. The Edwards dessert, with 43 ingredients including 26 additives, hits 0.953 and the Pearl River (14 ingredients, five additives) scores 0.720. The authors suggest that providing scores like this will help consumers make better choices. There’s a small problem with that logic, though; stay tuned.

We’re not here to call out individual food items. The whole point of this new model is that it can work on every food item you shop for. And it can be automated. The authors used that automation to score every single food item sold at Whole Foods, Target, and Walmart. 

Let’s look at the distribution of scores at these three grocery chains. Overall, you see that there tend to be more products at the higher end of the processing scale, though Whole Foods seems to have a bit of a flatter curve. Of interest, more of Target’s products are at the very high end of the ultraprocessed scale than Walmart’s, which would not have been my hypothesis a priori.

These boxes show you the range of processing scores within a given category. There’s a lot of variability out there. Of course, it’s hard to find cookies and biscuits that aren’t ultraprocessed, but if you look you can find relatively less-processed spreads and coffee drinks and yogurts. 

Does all this processing come at a cost? Not to your bank account. The researchers found that by and large, the more processed a food was, the cheaper it was per calorie. That Edwards whipped cheesecake? $1.87 per serving, and with 260 calories, it’s 1.4 calories per cent. The Pearl River cheesecake? $4 per serving, 260 calories, and 0.65 calories per cent.

If you’re trying to stretch your food dollar, the more-processed stuff is better. You’ll pay in other ways, of course. 

I couldn’t help it. I searched their database for the food with the lowest cost per calorie. The winner? Betty Crocker Super-Moist German Chocolate Cake Mix: 15 cents a serving, 160 calories, or about 11 calories per cent. Impressive.

This “more-processed foods are cheaper” paradox was seen across nearly all food categories. I suspect this occurs because, while processing takes money (machines, labor, raw materials, etc.), it is done at an industrial scale, driving individual costs down.

There is no single ingredient that makes a food ultraprocessed. It’s the totality of what goes into making a food product. But because each of the 50,000-plus foods in their database now had a score, the researchers could determine which single ingredients appear most in foods with high total scores. I have dubbed these “red flag” ingredients; they might not even be all that processed themselves, but they are markers that the food has been very processed. If you’re reading a food label, these are probably the ones you want to watch out for.

This chart tells the story focusing on added oils.

Palm and corn oil are good markers that a food is highly processed; flaxseed and peanut oil, not so much. Once again, I looked in their raw data to see what ingredients were at the tippy top of the red-flag list, restricting my search to ingredients that were seen across at least 10 products. That left oat blend, crust grain oat, "palm kernel oil with TBHQ for freshness,” spice oil, and sorbic acid. 

At the start of this, I told you you’re right to worry about ultraprocessed foods, but you might be worrying for the wrong reason. When you look at these ingredient lists and see things like “sorbitan monostearate” or “palm kernel oil with TBHQ for freshness,” you might be worried that these additives are somehow toxic. And sure, if you google enough, you may find some study in rats that shows that this dye or that preservative causes cancer or something. But honestly, this is not how ultraprocessed foods are killing us.

It's simpler than that.

The reason ultraprocessed foods are so bad for us is that they are easier to eat. These additives include chemicals to keep the food fresh longer — so it stays soft if it’s supposed to be soft, crunchy if it’s supposed to be crunchy. They add salts, spices, and oils that are delicious. Grains that have been stripped of fiber so that they are easier to chew and have a better mouthfeel. These are all designed by very smart scientists to be highly delicious and easy to eat. 

A landmark 2019 study took 20 volunteers and had them stay at the National Institutes of Health for a month. For 2 weeks they got a normal diet; for 2 weeks, a diet high in ultraprocessed foods. They could eat as much or as little as they wanted, and the foods they were presented with were matched by macronutrients, energy density, calories, etc. The average volunteer ate 500 extra calories per day while randomized to the ultraprocessed diet. These foods kill us, quite simply, because we can’t help but eat more of them. 

There is even a term they use for this in the biz: cravability. That’s what they are going for; it’s basically saying, “We want this to be as close to addictive as possible.”

And it works. Look, I love blueberries, but I don’t really have a problem stopping eating them after I’ve had some. Have you ever tried to stop eating Doritos, though? It’s tough. 

That’s where I part with the authors here, who suggest that we can use FPro scores to choose the better options for us within a food category — the less-processed cheesecake over the more-processed cheesecake. The problem with that idea is that the healthier one, the one with the lower score, may not taste as good. 

And we need to be okay with that. This is actually how we break the cycle of ultraprocessed foods. We have to learn to enjoy food that doesn’t smack us straight in the tastebuds the moment it goes in our mouths, that doesn’t give a dopamine rush just from the smell alone. We have to learn to love real food again. 

I think this is going to start with our kids, which is why I hate how much ultraprocessed food is directed toward children. They learn what tastes “good” at a young age, and those habits can be incredibly hard to break. But maybe a bit of feedback from those FPro scores will nudge us toward food that tastes more like food and less like bottled rainbows. And maybe, just maybe, we can learn to be happy about that.

F. Perry Wilson, MD, MSCE, is an associate professor of medicine and public health and director of Yale’s Clinical and Translational Research Accelerator. His science communication work can be found in the Huffington Post, on NPR, and here on Medscape. He posts at @fperrywilsonand his book, How Medicine Works and When It Doesn’t, is available now.