Should You Fear Ultra-Processed Foods?
The complex problem of UPFs and why the headlines are very misleading
For decades, we’ve been trying to figure out how to separate food into easy “healthy” and “unhealthy” categories. There was the “calorie-dense and nutrient poor” descriptor, there’s the ever-present discussion about full vs low fat, and the ongoing war about carbs, seed oils, gluten, and everything else under the sun.
It’s enough to make you ignore dietary advice entirely, really.
The most recent fad is processing. If you believe the recent headlines, processed foods dramatically increase your risk of heart disease, diabetes, cancer, and even death. According to the stories, a new study has proven that Ultra-Processed Foods (UPFs) are essentially the devil disguised as a chocolate bar or corn chips.
But if you look at the actual evidence here, it’s really not that strong. UPFs are probably, on the whole, worse for health than other foods, but the real issue is that our food system is far more complex than simple designations like “ultra-processed” allow.
Let’s look at the science.
Umbrellas Everywhere
The new paper that has been making headlines is what’s known as an umbrella systematic review and meta-analysis. This is a type of meta study where scientists combine estimates from other meta-studies which themselves aggregate all of the evidence around on a topic in question. Basically, a systematic review tries to capture every study published on a question, and an umbrella review tries to capture every systematic review in turn.
In this umbrella review, the authors managed to find a total of 14 other reviews to pull together. Those reviews included a total of 45 separate statistical analyses that the umbrella review then pooled together, looking at how UPFs impacted everything from ulcerative colitis to heart attacks. Those 45 analyses included the data from an impressive 10 million individual patients.
Overall, after running this lengthy series of analyses on other people’s analyses in a wonderful chain of meta-science, the authors concluded:
“Greater exposure to ultra-processed food was associated with a higher risk of adverse health outcomes, especially cardiometabolic, common mental disorder, and mortality outcomes.”
And from here we get the scary headlines. It sounds as if eating UPFs is slowly killing us all. Even the author’s conclusions imply this, arguing that this provides a rationale to “target and reduce dietary exposure to ultra-processed foods for improved human health.”, which implies that they consider this to be a slam dunk of UPFs causing harm to our health.
Problem is, the evidence here is a whole lot more complicated than that.
The Data
The problem with umbrella reviews is something I’ve written about several times before - they are essentially a form of citation laundering, albeit usually unintentionally. Because the authors are filtering other meta-science studies, you have to get through several layers before you get to the actual research that has been done on humans which is driving all of this concern.
For example, here’s one of the analyses that the umbrella review cited. It’s a systematic review from June 2023 looking at UPFs and risk of cancer. They aggregated together just 6 studies, and came to the conclusion that there is a tiny associated increase in risk of cancer for people who eat more UPFs than those who don’t.
But you have to go a step further to look at the actual studies that the review found. They looked at this paper on the large UK Biobank cohort, this paper on the Nutrinet Santé cohort from France, this much smaller Spanish study, a tiny Iranian study, a very small Moroccan paper, and another large paper that aggregates the results from three US cohorts.
Of these, the UK, French, and US papers are quite strong, and the others are far weaker, but they all go into the same statistical model. Even the better papers here are only looking at associations, rather than causation - they are trying to see if people who eat more UPFs are worse off, not whether UPFs cause ill health per se.
There are two big problems here. The first is that this umbrella review by definition is including a whole heap of low-quality research, and we don’t have much of an idea of how those bad studies are impacting the final numbers. In the study above, a large portion (~30% or so) of the main result was drawn from the extremely small, poorly-controlled papers. This problem has a big impact on the final numbers, the 50% increased risk of death from heart disease that’s quoted in many of the news stories.
The other issue is that there just aren’t that many studies on UPFs that can be included in this sort of review, and even the different studies look at the same patient populations just in different years. The umbrella review included estimates from both Isaksen 2023 and Lian 2023, both systematic reviews of the impact of UPFs on cancer mortality. But these studies have a huge overlap, including 5 of the same papers in their analyses. In fact, the primary statistical model in both cases is almost entirely drawn from the exact same two papers - Wang 2018 and Fiolet 2018, both published in the BMJ.
This means that the umbrella review is double, triple, or even quadruple counting the data. I found Fiolet 2018 included in three of the reviews that are part of the umbrella review. Perhaps more importantly, there is at least one paper in every single review I could find on either the NutriNet-Santé cohort, the UK Biobank cohort, the US NHANES, or the US Nurses/Physicians health studies. Rather than 14 entirely separate analyses that the umbrella review has collated into one, we’ve got a handful of samples that have been analyzed many dozens of times. If even one of these samples has a bias that we are unaware of, it could completely undermine the entire analysis by the authors.
While the idea of an umbrella review sounds robust, ultimately you’re relying on the included research, and often it just isn’t that strong. A key point here is that none of the included papers actually show that UPFs cause health issues. They all just look at correlations, with varying degrees of success. That’s why, if you go back and read the author’s quote from their conclusions above, you’ll see the all-important word “associated”. UPFs are “associated” with worse health, but we still don’t really know if they cause it or not.
Should I Eat The Cheerios?
This brings me to the real complexity of this whole situation - UPFs are something of a meaningless designation in the first place. Most of these studies used the NOVA classification system, which designates food into different levels of processing, but it’s not a strict definition. There’s a great deal of debate about which foods are minimally processed, processed, and ultra-processed. Even within individual teams, you get a fair bit of disagreement when people have to rate products into these categories.
You can conceptualize this by thinking about individual foods. Homemade granola with nuts and dried fruit is unprocessed. If you add honey, it becomes minimally-processed (you might say this is arbitrary, but that’s true of the entire rating system anyway). However, if you buy that same bagged granola on a supermarket shelf, and they’ve added salt to the mix and used a seed oil rather than butter to make it, suddenly you’re eating a processed food. If the fruit has been dried using citric acid, the food would probably be considered ultra-processed.
Now think back to all the research above. How did they even know how many UPFs people ate in all those studies? We generally use either 24-hour recall questionnaires, which ask people what they ate in the last 24 hours, or massive food frequency questionnaires which ask people about everything they remember from the previous year. But both of these approaches are limited and fraught with biases, making it hard to infer how much people eat of specific foods, never mind entire groups of poorly-defined products.
Not only do we not know whether UPFs actually cause disease, we also don’t even really know how much UPF the people in all these studies were eating. It’s possible that the findings of the umbrella review are just spurious correlations that tell us nothing about human health.
So, it’s hard to draw a great deal of meaning from the study that made all these headlines. But even more than that, UPFs are a conceptually problematic category as well. Processing isn’t a particularly useful bar against which to measure the healthiness of foods, and the more we lean on it the less useful it becomes. MacDonald’s burgers would fall into the “minimally processed” category if they slightly reformulated their buns and sauces. Low calorie diet shakes, which are used as a treatment to reverse diabetes, are ultra-processed, while handmade potato chips are minimally or entirely unprocessed. There’s really no health difference between processed and unprocessed cheese, but one is a UPF and the other minimally processed or processed depending on how it’s made.
All of this comes back to the question of whether UPFs are useful as a category, and fundamentally I’m unconvinced. It’s true that many unhealthy foods are UPFs, but so are healthy foods, and demonizing processing is unhelpful. Without food processing, quite a lot of people would be starving, or dead!
Moreover, adding the designation UPF to unhealthy foods doesn’t really change anyone’s understanding of the problems. You don’t need to know that ice cream is ultra-processed to know that it’s a treat food, or that you should try and avoid highly-sweetened cereals if possible. No one eats frozen lasagna because they’re convinced of the health benefits - they eat it because it’s tasty, convenient, and cheap.
In general, I’d say that eating processed food remains just as fine as it was before this research came out. Yes, you should avoid calorie-dense foods with few nutrients in them, but we’ve known that since at least the 60s and little has changed since. My advice is to try and eat healthily, but don’t worry too much about how much processing has gone into your food.
thank you so much for this.
What you said. I’ve always been a little bit annoyed at people who talk about processing as if it’s an evil in itself. If you pick cucumbers out of your garden and take them into your kitchen and make pickles out of them and then can those in your home canner- they are processed.
It isn’t the processing that’s bad for you. It’s what’s in it.