June 24, 2024

Last summer, I got a tip about a curious scientific finding. “I’m sorry, it cracks me up every time I think about this,” my tipster said.

Back in 2018, a Harvard doctoral student named Andres Ardisson Korat was presenting his research on the relationship between dairy foods and chronic disease to his thesis committee. One of his studies had led him to an unusual conclusion: Among diabetics, eating half a cup of ice cream a day was associated with a lower risk of heart problems. Needless to say, the idea that a dessert loaded with saturated fat and sugar might actually be good for you raised some eyebrows at the nation’s most influential department of nutrition.

Magazine Cover image

Explore the May 2023 Issue

Check out more from this issue and find your next story to read.

View More

Earlier, the department chair, Frank Hu, had instructed Ardisson Korat to do some further digging: Could his research have been led astray by an artifact of chance, or a hidden source of bias, or a computational error? As Ardisson Korat spelled out on the day of his defense, his debunking efforts had been largely futile. The ice-cream signal was robust.

It was robust, and kind of hilarious. “I do sort of remember the vibe being like, Hahaha, this ice-cream thing won’t go away; that’s pretty funny,” recalled my tipster, who’d attended the presentation. This was obviously not what a budding nutrition expert or his super-credentialed committee members were hoping to discover. “He and his committee had done, like, every type of analysis—they had thrown every possible test at this finding to try to make it go away. And there was nothing they could do to make it go away.”

Spurious effects pop up all the time in science, especially in fields like nutritional epidemiology, where the health concerns and dietary habits of hundreds of thousands of people are tracked over years and years. Still, the abject silliness of “healthy ice cream” intrigued me. As a public-health historian, I’ve studied how teams of researchers process data, mingle them with theory, and then package the results as “what the science says.” I wanted to know what happens when consensus makers are confronted with a finding that seems to contradict everything they’ve ever said before. (Harvard’s Nutrition Source website calls ice cream an “indulgent” dairy food that is considered an “every-so-often” treat.)

This article was featured in One Story to Read Today, a newsletter in which our editors recommend a single must-read from The Atlantic, Monday through Friday. Sign up for it here.     

“There are few plausible biological explanations for these results,” Ardisson Korat wrote in the brief discussion of his “unexpected” finding in his thesis. Something else grabbed my attention, though: The dissertation explained that he’d hardly been the first to observe the shimmer of a health halo around ice cream. Several prior studies, he suggested, had come across a similar effect. Eager to learn more, I reached out to Ardisson Korat for an interview—I emailed him four times—but never heard back. When I contacted Tufts University, where he now works as a scientist, a press aide told me he was “not available for this.” Inevitably, my curiosity took on a different shade: Why wouldn’t a young scientist want to talk with me about his research? Just how much deeper could this bizarre ice-cream thing go?

“I still to this day don’t have an answer for it,” Mark A. Pereira, an epidemiologist at the University of Minnesota, told me, speaking of the association he’d stumbled upon more than 20 years earlier. “We analyzed the hell out of the data.”

Just that morning, I’d been reading one of Pereira’s early papers, on the health effects of eating dairy, because it seemed to have inspired other research that was cited in Ardisson Korat’s dissertation. But when I scrolled to the bottom of Pereira’s article, down past the headline-making conclusions, I saw in Table 5 a set of numbers that made me gasp.

Back then, Pereira was a young assistant professor at Harvard Medical School. Hoping to address the newly labeled epidemics of obesity and diabetes, he initially focused his research on physical activity, but soon turned to the unsettled science of healthy eating. The status of dairy, in particular, was bogged down in simplistic and competing assumptions. “We just thought, Oh, you know, calcium and bones: It’s good for kids. But, oh, the saturated fat! Don’t eat too much dairy! 

Pereira and his co-authors tested these old ideas using data from a study, begun in 1985, that tracked the emergence of heart-disease risk factors in more than 5,000 young adults. After seeing the results, “we knew it was going to be very high-profile and controversial,” Pereira recalled. Pretty much across the board—low-fat, high-fat, milk, cheese—dairy foods appeared to help prevent overweight people from developing insulin-resistance syndrome, a precursor to diabetes. “I’ll tell you, this study surprised the heck out of me,” said one CNN correspondent, as Pereira’s study spiraled through the press.

But the international media coverage didn’t mention what I’d seen in Table 5. According to the numbers, tucking into a “dairy-based dessert”—a category that included foods such as pudding but consisted, according to Pereira, mainly of ice cream—was associated for overweight people with dramatically reduced odds of developing insulin-resistance syndrome. It was by far the biggest effect seen in the study, 2.5 times the size of what they’d found for milk. “It was pretty astounding,” Pereira told me. “We thought a lot about it, because we thought, Could this actually be the case? 

There were reasons to be wary: The data set wasn’t huge, in epidemiological terms, and participants hadn’t reported eating that many dairy-based desserts, so the margin of error was wide. And given that the study’s overall message was sure to attract criticism—Pereira recalled getting “skewered” by antidairy activists—he had little desire to make a fuss about ice cream.

Pretty soon, Pereira’s peers found themselves in the same predicament. Building on the 2002 study and the growing interest in dairy, researchers at the Harvard School of Public Health decided to break out some of their most powerful tools. Since the 1980s, Harvard’s scientists have been collecting “food-frequency questionnaires” and medical data from many thousands of nurses, dentists, and other health-care workers. These world-famous studies have fueled a stream of influential findings, including some of the data that sparked the removal of trans fats from the food supply.

The results of Harvard’s first observational study of dairy and type 2 diabetes came out in 2005. Based on data collected from just one of their three cohorts, following men between 1986 and 1998, the authors reported that higher dairy intake, and higher low-fat-dairy intake in particular, was associated with a lower risk of diabetes. “The risk reduction was almost exclusively associated with low-fat or non-fat dairy foods,” a Harvard news bulletin explained. An article on Fox News’s website underscored the low-fat message: “There was no decrease in men who drank whole milk,” the story said.

Perhaps not whole milk, but what about butter pecan? Near the end of the Harvard paper, where the authors had arrayed the diabetes risks associated with various dairy foods, was a finding that was barely mentioned in the “almost exclusively” low-fat narrative given to reporters. Yes, according to that table, men who consumed two or more servings of skim or low-fat milk a day had a 22 percent lower risk of diabetes. But so did men who ate two or more servings of ice cream every week. Once again, the data suggested that ice cream might be the strongest diabetes prophylactic in the dairy aisle. Yet no one seemed to want to talk about it.

In the years that followed, research summaries generally agreed that high dairy intake overall was associated with a slightly reduced risk of diabetes, but called for more investigation of which specific dairy foods might have the greatest benefits. In 2014, Harvard’s nutrition team brought another dozen years of diet-tracking data to bear on this question. In this new study, total dairy consumption now seemed to have no effect, but the ice-cream signal was impossible to miss. Visible across hundreds of thousands of subjects, it all but screamed for more attention.

Following a pattern of incredulousness that was by then more than a decade old, Frank Hu, the study’s senior author and the future chair of Harvard’s nutrition department, asked the graduate student who’d led the project, Mu Chen, to double-check the data. “We were very skeptical,” Hu told me. Chen, who is no longer in academia, did not respond to interview requests, but Hu recalled that no errors in the data could be found.

The Harvard researchers didn’t like the ice-cream finding: It seemed wrong. But the same paper had given them another result that they liked much better. The team was going all in on yogurt. With a growing reputation as a boon for microbiomes, yogurt was the anti-ice-cream—the healthy person’s dairy treat.

“Higher intake of yogurt is associated with a reduced risk” of type 2 diabetes, “whereas other dairy foods and consumption of total dairy are not,” the 2014 paper said. “The conclusions weren’t exactly accurately written,” acknowledged Dariush Mozaffarian, the dean of policy at Tufts’s nutrition school and a co-author of the paper, when he revisited the data with me in an interview. “Saying no foods were associated—ice cream was associated.”

But yogurt made so much more sense. In a way, it was confirmation of something that everyone already knew. From the start of yogurt’s entrée into the American diet, it had been perceived as an exotic food from a faraway land, quivering with vague health-giving properties. Even after being spiked with sugar in the ’70s and ’80s to better suit the U.S. market, yogurt still retained its image as an elixir.

Furthermore, a growing body of literature suggested that yogurt’s health benefits might be real. Harvard had found, a few years earlier, that eating yogurt was associated with reduced weight gain; researchers at the university were interested in its possible effects on gut bacteria as well. Other studies—including those that first revealed the ice-cream signal—had also sketched the slender outlines of a yogurt effect. When Chen and Hu pooled together findings from this research, added in their latest data, and performed a meta-analysis, they concluded that yogurt was indeed associated with a reduced risk of diabetes—a potential benefit, they wrote, that warranted further study.

Regarding ice cream’s potential benefits, they had much less to say. I asked other experts to compare the 2014 yogurt and ice-cream findings. Kevin Klatt, a nutrition scientist at UC Berkeley, said the ice-cream effect was “more consistent” than yogurt’s across the studied cohorts. Deirdre Tobias, an epidemiologist at Harvard, the academic editor of The American Journal of Clinical Nutrition, and a member of the advisory committee for the 2025 update to the U.S. dietary guidelines, agreed with that assessment. Even Dagfinn Aune, an epidemiologist at Imperial College London and a peer reviewer of the Chen and Hu paper, said that the ice-cream effect was “similar” in magnitude to, or “slightly stronger” than, the one for yogurt.

So how did the Harvard team explain away the ice-cream finding? The theory went like this: Maybe some of the people in the study had developed health problems, such as high blood pressure or elevated cholesterol, and began avoiding ice cream on doctors’ orders (or of their own volition). Meanwhile, people who didn’t have those health problems would have had less reason to give up their cookies and cream. In that scenario, it wouldn’t be that ice cream prevented diabetes, but that being at risk of developing diabetes caused people to not eat ice cream. Epidemiologists call that “reverse causation.”

To test this idea, Hu and his co-authors set aside dietary data collected after people received these sorts of diagnoses, and then redid their calculations. The ice-cream effect shrank by half, though it was still statistically significant, and still bigger than the low-fat-dairy effect that Harvard had publicized in 2005. In any event, if people who received adverse diagnoses cut back on their ice cream, you might expect that they’d also cut back on, say, cake and doughnuts. So shouldn’t there be mysterious protective “effects” for cake and doughnuts too? “There should be,” Mozaffarian said. “That’s why the finding for ice cream is intriguing.”

The new analysis was hardly a slam dunk. On paper, the yogurt and ice-cream effects still looked pretty similar. “Within the realm of statistical uncertainty, they’re identical,” Mozaffarian told me. But in the 2014 paper, he and the other authors had argued that “reverse causation may explain the findings” for ice cream. And as academia’s public-relations machinery came to life, nuance went out the window.

“Does a yogurt a day keep diabetes away?” asked the press release that went out on publication day. “Other dairy foods and consumption of total dairy did not show this association,” said Hu, the senior author, in an ice-cream-free appraisal included in the release and echoed in Harvard’s own press bulletin. “Yogurt has approached wonder-food status in recent years,” a Forbes article on the paper noted. “In the new study, other forms of dairy like milk and cheese, did not offer the same kind of protection as yogurt.”

Hu says today that the Harvard researchers felt confident in their conclusions about yogurt largely on account of their meta-analysis, and the fact that prior clinical studies and basic science research supported the idea that probiotics improve metabolic outcomes. “For ice cream, of course, there is no prior literature,” he said. Given that the ice-cream effect was diminished when they tested their reverse-causation theory, he called it “much more plausible” that yogurt would help prevent diabetes than ice cream.

A photograph of a freezer filled with pints of ice cream.
Kenji Toma / Trunk Archive

After his paper was published, it didn’t take long for the Harvard group’s good news about yogurt to take hold as a dominant scientific narrative. Two years later, when a team of researchers based in the Netherlands and at Harvard analyzed all the evidence it could find on dairy and diabetes, the yogurt effect popped out. A featured graph from the team’s 2016 paper in The American Journal of Clinical Nutrition summarizes data from about a dozen studies: As someone’s yogurt intake mounts to roughly one-third of a cup a day, their risk of getting diabetes shrinks by 14 percent.

The authors also found the ice-cream effect: Consuming as little as a half a cup per week was associated with a 19 percent reduced diabetes risk. But that finding’s epitaph was already written. The researchers concluded that consuming “dairy foods, particularly yogurt,” might help curb the diabetes epidemic, and noted that the benefits of ice cream had elsewhere been written off as a product of reverse causation. The evidence in yogurt’s favor was much better established, Sabita Soedamah-Muthu, an epidemiologist at Tilburg University and the paper’s senior author, told me. The ice-cream effect had fewer studies in its corner. “We didn’t believe in it,” she said.

There’s a thing that happens when you start writing a story about how maybe, possibly, believe it or not, ice cream might be sort of good for you, and how some of the world’s top nutritionists gathered evidence supporting that hypothesis but found reasons to look past it. You begin to ask yourself: Am I high on my own ice-cream supply? I asked the experts for a gut check. Pereira, the first to hit upon the ice-cream effect, told me that it just wasn’t the kind of result that goes down well in the “closed-minded” world of elite nutrition. “They don’t want to see it. They might ponder it for a second and kind of chuckle and not believe it,” he said. “I think that’s related to how much the field of nutritional epidemiology in the modern era is steeped in dogma.” Tobias, the journal editor and member of the 2025 U.S. dietary-guidelines advisory committee, called it “totally fair criticism” to ask why yogurt was played up while ice cream was played down. She expressed support for the Harvard team’s handling of the data, while acknowledging the tensions involved: “You don’t want to overstate stuff that you know probably has a high likelihood of bias, but you also don’t want to do the opposite and seem to be burying it, either.”

Hu, the Harvard nutritionist, said that deciding what a study means requires looking beyond the numbers to what is already known about dietary science: “You need to interpret the data in the context of the rest of the literature.” Mozaffarian, Hu’s co-author, echoed this view. Still, he noted, “you’re raising a really, really important point, which is that when, as scientists, we find things that don’t fit our hypotheses, we shouldn’t just dismiss them. We should step back and say, ‘You know, could this actually be true?’ ”

Could the idea that ice cream is metabolically protective be true? It would be pretty bonkers. Still, there are at least a few points in its favor. For one, ice cream’s glycemic index, a measure of how rapidly a food boosts blood sugar, is lower than that of brown rice. “There’s this perception that ice cream is unhealthy, but it’s got fat, it’s got protein, it’s got vitamins. It’s better for you than bread,” Mozaffarian said. “Given how horrible the American diet is, it’s very possible that if somebody eats ice cream and eats less starch … it could actually protect against diabetes.” The “Got Milk?” crowd also loves to talk about the “milk-fat-globule membrane,” a triple-layered biological envelope that encases the fat in mammalian milk. Some evidence suggests that dairy products in which the membrane is intact, such as ice cream, are more metabolically neutral than foods like butter, where it’s lost during the churn. (That said, regular cream has an intact membrane, and it hasn’t been consistently associated with a reduced diabetes risk.)

Then there is what might charitably be termed the “real-world evidence.” In 2017, the YouTuber Anthony Howard-Crow launched what Men’s Health called “a diet that would make the American Dietetic Association shit bricks”: 2,000 calories a day of ice cream plus 500 calories of protein supplements plus booze. After 100 days on the ice-cream diet, he’d lost 32 pounds and had better blood work than before he’d started pounding Irish-whiskey milkshakes. Still, the method is unlikely to take the slimming world by storm: Howard-Crow called his ice-cream bender “the most miserable dieting adventure I have ever embarked upon.”

But overall, I found more receptiveness to the ice-cream signal than I was expecting. “It’s been more or less replicated,” Pereira noted. “Whether it’s causal or not still remains an open question.” Mozaffarian agreed: “I think probably the ice cream is still reverse causation,” he said. “But I’m not sure, and I’m kind of annoyed by that.” If this had been a patented drug, he continued, “you can bet that the company would have done a $30 million randomized controlled trial to see if ice cream prevents diabetes.”

To be clear, none of the experts interviewed for this article is inclined to believe that the ice-cream effect is real, although sometimes for reasons that differ from Hu’s. Pereira, for example, pointed out that people aren’t always truthful when they’re quizzed on what they eat. His 2002 study found that overweight and obese people reported eating fewer dairy-based desserts than other people. “I don’t believe that the heavier people consume less desserts,” he said. “I believe they underreport more.” If that’s true, then admitting to eating ice cream might correlate with metabolic health—and the ice-cream effect would be, in its way, a marker of fat stigma in America.

The problem with this line of thinking is that once you start contemplating all the ways that cultural biases can seep into the science, it doesn’t stop at dairy-based desserts. If the ice-cream effect can be set aside, how should we think about other signals produced by the same research tools? “I don’t know what I believe about yogurt,” Tobias told me. It’s widely known that yogurt eaters on average are healthier, leaner, wealthier, better educated, more physically active, more likely to read labels, more likely to be female, and less likely to smoke or drink or eat Big Macs than never-yogurters. “You can’t confidently adjust away all of that kind of stuff,” said Klatt, the UC Berkeley nutritionist.

In 2004, the English epidemiologist Michael Marmot wrote, “Scientific findings do not fall on blank minds that get made up as a result. Science engages with busy minds that have strong views about how things are and ought to be.” Marmot was writing about how politicians deal with scientific evidence—always concluding that the latest data supported their existing views—but he acknowledged that scientists weren’t so different.

The ice-cream saga shows how this plays out in practice. Many stories can be told about any given scientific inquiry, and choosing one is a messy, value-laden process. A scientist may worry over how their story fits with common sense, and whether they have sufficient evidence to back it up. They may also worry that it poses a threat to public health, or to their credibility. If there’s a lesson to be drawn from the parable of the diet world’s most inconvenient truth, it’s that scientific knowledge is itself a packaged good. The data, whatever they show, are just ingredients.

This article appears in the May 2023 print edition with the headline “The Ice-Cream Conspiracy.”


Leave a Reply

Your email address will not be published. Required fields are marked *