How One Keto Trial Set Off a New War in the Nutrition World

Whenever the keto diet is discussed, there’s going to be beef. But this time it’s different. A new research paper on the diet’s effects has whipped the nutrition field into a frenzy. Researchers behind the study say it supports keto’s health credentials; opponents claim the research shows the exact opposite. Cue public spats on social media, questions about the study’s rigor, and calls for it to be withdrawn. “It’s a collective mess,” says Kevin Klatt, an assistant research scientist and instructor in the Department of Nutrition Sciences and Toxicology at UC Berkeley.

Published on April 7 in JACC: Advances, the paper examines the relationship between cholesterol and the ketogenic diet—the practice of consuming low-carb, high-fat foods to try to push the body into “ketosis,” where cells burn fat instead of carbs for energy. Keto diets have become a popular strategy used by millions for losing weight, though detractors have questioned how healthy it is to consistently consume high amounts of fat.

The study, by getting to the heart of this question, has received a significant amount of attention. Altmetric, which measures the attention a publication receives in the press and on social media, puts it in the top 5 percent of the papers it tracks—more than 24 million pieces of research. The majority of attention has come from X.

According to some keto advocates, the paper’s findings are a step towards refuting the widely accepted theory that LDL cholesterol, also referred to as “low-density lipoprotein” or “bad” cholesterol, has a causal relationship with heart disease and other cardiovascular conditions. Overturning this hypothesis would undermine long-standing medical advice that people should keep their LDL cholesterol low, and potentially rewrite the rules on consuming fatty foods.

The trial recruited 100 subjects who were otherwise healthy but had been following a ketogenic diet for at least 2 years, and who as a result had unusually high cholesterol levels in their blood. Patients who exhibit these characteristics, but who show other signs of being in good cardiometabolic health—having lean physiques, low body fat, low blood pressure, and good insulin sensitivity—are sometimes called Lean Mass Hyper-Responders (LMHRs). The study set out to show if its cohort of LMHRs were developing fatty deposits known as plaque in their arteries, a known risk for people with high levels of LDL cholesterol in their blood. Participants were followed for one year, during which they continued with their keto diets, and their plaque levels were observed at the beginning and end of the study.

One of the study’s authors is Dave Feldman, a software engineer and entrepreneur without a medical license or training, who has devoted himself to all things keto and cholesterol. In an email to WIRED, Feldman claimed that it was he who coined the term Lean Mass Hyper-Responders, back in 2017. In the past he has organized his own experiments—without the guidance of an institutional board review, which in formal experiments are used to ensure ethical behavior and participant welfare—to try to get the attention of scientists and have them study LMHRs. Feldman’s charity, the Citizen Science Foundation, crowdfunded the recent study, which was run through the Californian research organization the Lundquist Institute, with an institutional review board.

In a video released on X the day the paper came out, Feldman claimed the study found no association between LDL cholesterol and plaque in the patients, and no association between apolipoprotein B (ApoB) and plaque. (ApoB helps carry fat molecules around the body, and higher levels of it are associated with cardiovascular disease.) These alleged findings run counter to large amounts of already-existing evidence suggesting both LDL and ApoB have a causal relationship with the development of plaque in the arteries. In Feldman’s view, the study shows that despite their high levels of LDL cholesterol, the patients’ keto diets weren’t raising their risk of plaque.

However, many doctors and researchers reached the opposite conclusion when looking at the work. On May 7, JACC: Advances published a pre-proof version of a Letter to the Editor, written by two researchers specializing in nutrition, Miguel López-Moreno and José Francisco López-Gil. They highlighted concerns with the study, including what they alleged to be “selective reporting” of data, the study’s lack of a comparator group, the validity of the statistical modeling used, and the weakness of using a one-year timeframe.

The study was also heavily criticized for seeming to mask its original focus. Originally, it was supposed to look at the percentage change in non-calcified plaque volume (NCPV)—soft plaque that had not yet hardened inside participants’ blood vessels—in the participants over the course of the study. A graph of NCPV change appeared in the paper, but measurements were not provided or mentioned. Instead, the paper ended up offering an exploratory analysis—that ApoB does not beget plaque—“that was implausible to do based on the data they had,” says Spencer Nadolsky, a Michigan-based physician specializing in obesity medicine and lipidology.

This means the paper “shouldn’t have made it through peer review in the first place,” Nadolsky believes. If researchers leave out the intended goal of a study, critics allege that they could then cobble together any data once the experiment has been done, without clarifying what they were initially looking for, and try to pass this off as evidence of something. Because the study wasn’t designed to investigate the alternative hypothesis of the explanatory analysis, there may be flaws in the data being used to support it—biases in how it was obtained, or not enough of it to reach a robust conclusion.

“This is the first thing you’re not supposed to do,” says Nadolsky of the decision to shift focus. “That’s why we’re hammering them.”

“Interpretative disagreements are far from rare in nutritional science,” wrote Adrian Soto-Mota, lead author of the study, in response to a request for comment from WIRED. He notes that all of the limitations of the study’s design were acknowledged in the paper, and says that when they used the alternative statistical model suggested by López-Moreno and López-Gil, this still corroborated the paper’s conclusions.

Soto-Mata also says the focus wasn’t switched. The changes in participants’ NCPV were shown in the graph, he says, and notes that these changes were used in “almost all the analyses in our paper.” Moreover, he says it’s wrong to describe the analysis that ApoB does not beget plaque as implausible based on the data gathered. “Our analysis was carried out by two experts in data analysis, and it was independently reviewed by an expert in statistics during the peer review process,” he says.

Nadolsky, though, has called for the paper to be retracted, and co-wrote a response to the research, which has been released as a preprint, which takes issue with the paper’s findings, interpretation, and statistical analysis, among other concerns. The response says the study’s conclusions—a “clear example of scientific spin”—are not supported by the data, and have the potential to misinform both doctors and patients about the risks of following a high-fat diet.

“Nothing was spun, and our conclusions remained unchanged after multiple sensitivity analyses and an independent expert data analysis review,” says Soto-Mota.

What sets Nadolsky’s criticism apart from others’ is that he had a part in designing the study. Feldman and Nadolsky had gone back-and-forth for years online about the risks of high cholesterol, with Feldman suggesting that the traditional consensus around its risks might be wrong, particularly for the LMHR population.

Instead, Feldman proposed a new alternative theory—the lipid energy model—which he and some of his coauthors on the current paper outlined in a study published in Metabolites in 2022. In this unproven theory, high LDL is thought to be unconcerning in LMHRs because their bodies have become more efficient in transporting cholesterol while running primarily on fat.

Nadolsky, though a believer in the consensus view on cholesterol, was still interested in getting some data on the effects of LDL cholesterol in LMHRs, and a study investigating Feldman’s theory was a way to reach across the aisle to get it.

But in putting together a study to test Feldman’s hypothesis they faced difficulty, Nadolsky explains. It would be shot down by an institutional review board, as it would require people with extremely high LDL cholesterol levels to go untreated, when this is known to be potentially dangerous. However, a workaround would be to observe plaque progression in people experiencing diet-induced hypercholesterolemia (high LDL cholesterol due to their keto diet) who were refusing lipid-lowering medications.

The recruitment and promotion of the study was done on X via the #LMHRstudy hashtag, in addition to Feldman’s LMHR Facebook group, which also called for fundraising contributions—and it was during this process that Nadolsky began to grow concerned. During recruitment, Feldman also presented some of the preliminary data at a low-carb conference, using it “to try to present that the [LMHR] phenotype was benign, because most of the individuals seemed to not have plaque at baseline,” Nadolsky says. He says that Feldman was doing this to recruit more subjects and donations for the research; but in essence, this was presenting supposed findings of the research before it had been properly conducted.

At this point, Nadolsky conferred with multiple scientists and researchers outside of the study, and was advised to wash his hands of the project. “It was clear there was going to be a spin, no matter what the data showed,” Nadolsky claims. Nadolsky filed a complaint with the institutional review board overseeing the study, for ethical concerns. The board, says Soto-Mata, “allowed the study to proceed after concluding that no ethical transgressions had been made.” The Lundquist Institute did not respond to a request for comment from WIRED.

While the study was still in the recruitment phase, Nadolsky left the team.

Klatt, of UC Berkeley, is extremely well-versed in nutrition research and the current online debates around cholesterol. He’s written about this study and its fallout on his personal Substack, and calls Nadolsky a friend. Klatt discussed the study with Nadolsky while it was ongoing, and many aspects concerned him.

Klatt brought up issues of undisclosed biases to the Lundquist Institute, the host of the trial, along with Dave Feldman’s “strongly vested interest” in the results of the study that was not properly disclosed, claiming he was “a conflicted party with no training in the biomedical sciences.” His email to the Institute about these issues went unanswered. “I think this study has gotten to the point of being extremely unethical,” Klatt says.

“All authors adhered to the conflicts-of-interest disclosure guidelines the journal required,” Soto-Mata says. “Our study was independently reviewed, approved, and monitored by an expert Research Ethics committee, all its recommendations were followed, and all its standards were met.”

While some researchers and physicians are tearing the study apart, or using it to show that keto can have adverse effects, Klatt doesn’t draw any strong conclusions. “People are talking past each other,” he says. Generally speaking, there are two clear camps, with one thinking the traditional lipid hypothesis holds up, and another thinking the new lipid energy model might work. Klatt puts himself in a third camp, asking: “Why are we trying to interpret this study at all?”

“I’m an editor at the American Journal of Clinical Nutrition,” Klatt says, “and I would like to believe that we would have rejected this outright without even sending it out for peer review, because it has so many obvious issues.” He is worried about people using this flawed study as proof the consensus on the risks of LDL cholesterol has been “debunked,” which it has not.

One of the study’s coauthors, Matthew Budoff, a professor of medicine at UCLA as well as an investigator at the Lundquist Institute, acknowledged in an email to WIRED that there had been “incredible scrutiny of the data on social media, which is more than expected based on my prior publications.” He noted the research team is seeking to have the paper incorporate corrections, but that this is ultimately at the discretion of the journal. A response to the Letter to the Editor from the coauthors clarifies some of the issues, he wrote.

That reply to the Letter to the Editor has now been published—and reveals that the study’s data may support the conventional position on the risk of cholesterol after all. The study’s authors share that the “pooled median change” in NCPV in the participants—the rise in the type of plaque the study was set up to investigate, but which originally wasn’t explicitly quantified in the paper—was an alarming 42.8 percent. The reply goes on to state that the study’s findings were “compatible with a causal role of ApoB in atherosclerosis”—the build-up of fat in the arteries—which they’ve “acknowledged and supported in previous publications.” The letter says that not mentioning this percentage increase in NCPV “was a sincere oversight, not intentional selective reporting.”

But this concession comes after the horse has bolted. Feldman’s hypothesis is already appearing in laypeople’s research—with the keto diet having been among the most Googled diets in recent years, and keto products a growing multibillion-dollar industry. Answering the query “What is special about Lean Mass Hyper-Responders,” ChatGPT provides the lipid energy model, Feldman’s argument against the consensus on cholesterol, among the initial explanations for why there is so much controversy and interest. There is also a Cholesterol Code documentary in the works—which covers Feldman’s personal experience and his research, including this study—which Feldman predicts will be available on a major streaming service sometime this year.