The Latest Attack on Low-Carb Diets: Science or Politics?

Nina E. Teicholz, MPhil; Fabiano M. Serfaty, MD


September 14, 2018

The low-carbohydrate diet has been shown in clinical trials to reverse diabetes, lead to weight loss, and improve most heart disease risk factors, all of which logically should lead to longer life.[1,2] Yet, a recent paper in Lancet Public Health[3] made headlines around the world when it declared that a low-carb diet will shorten life. Why the disconnect? The answer centers on the fundamentally weak data in the Lancet article, promoted by authors with a possible interest in preserving the status quo.

This caveat dwarfs everything else, such that ignoring it is like promoting a skyscraper's height while overlooking that it's built on a foundation of sand.

The paper examined data from the Atherosclerosis Risk in Communities (ARIC) study,[4] a project of the National Institutes of Health (NIH) that, since 1987, has followed some 15,500 middle-aged men and women in four US communities. Two of the Lancet authors were members of the Minneapolis center of this study; the other authors, all from Harvard, are not listed as ARIC study participants.[5]

Low-Carb Eating Is Not Actually Studied in This Paper

It's remarkable that the authors made such definitive claims about "low-carb diets," as the paper defined this diet as up to 37% of calories from carbohydrates—not "low-carb" by the latest standards of practice. Evidence from the past 5 years shows better health when carbohydrates are kept below 30% of calories,[6] with the greatest benefits,[6,7] including reversal of type 2 diabetes,[8,9] seen with a very low-carb or ketogenic diet, where carbohydrates normally total between 5% and 20%.[10,11]

The Limits of Observational Science

A fundamental limitation of ARIC is that, like all observational studies, it can only show association, not causation. The epidemiologists who do these studies will often mention this "association only" caveat but tend to breeze by it. Yet truly, this caveat dwarfs everything else, such that ignoring it is like promoting a skyscraper's height while overlooking that it's built on a foundation of sand.

An "association-only" study can never fully control for outside factors that researchers may or may not have measured. For instance, the low-carb group in ARIC also had higher rates of diabetes, higher average body mass index, smoked more, and exercised less. Thus, the poorer outcomes seen in ARIC could well have been due to any one of these factors or other unhealthy behaviors that the researchers didn't or couldn't measure.

No doubt epidemiology can, under certain conditions,[12] be used to suggest cause-and-effect relationships. Most important, the strength of association must be strong—eg, the 15-to-30 times greater risk of contracting lung cancer in heavy smokers compared with never-smokers. Yet, the association found in the Lancet study was a mere fraction of that—under two—which is considered by most epidemiologists to be too small for serious consideration.[13]

Weak associations are overwhelmed by other issues known to affect health, called confounders. For instance, whether a person drinks, smokes, exercises, or even goes to church has an effect on health—no one really knows how much—and researchers must try to adjust for all of them. Perhaps an unknown factor, such as an environmental toxin, has affected someone's health. Epidemiologists cannot adjust for that, as they will not have measured it.

A pointed example of this problem in ARIC is that the paper makes no mention of adjusting for alcohol, a potent confounder for longevity.

For this reason, Stanford professor and evidence-based medicine expert John Ioannidis wrote in an opinion piece recently published in the Journal of the American Medical Association[14] that, given all of the problems with nutrition epidemiology, "Reform has long been due." The claims of this science, when tested in clinical trials, have been shown in two separate analyses to be correct 0%-20% of the time.[15,16] This means that 80%-100% of the time, they're wrong.

Contradicted by the Gold Standard of Evidence

The Lancet paper asserts that this weak evidence should somehow trump the far more rigorous data from randomized controlled clinical trials. These are considered the gold standard of science, simply because, whatever their flaws, they can demonstrate cause and effect. This more sound science is ignored by the Lancet authors in two ways.

First, they give short shrift to the low-carb clinical trials, now numbering more than 70 with at least 7000 people. The authors acknowledged this literature in half a sentence, stating, "Although many randomized controlled trials of low-carbohydrate diets suggest beneficial short-term weight loss and improvements in cardiometabolic risk..." There is no recognition that the low-carb evidence includes three 2-year trials—considered long enough to flush out any negative side effects (yet finding none).

The Lancet authors, in recommending a "moderate" diet of 50%-60% carbohydrates, also ignore another body of gold-standard evidence—on exactly this diet. The "moderate" carb, low-fat diet has, after all, been enshrined as our existing official dietary guidelines for Americans for decades. Ever since the late 1970s, when the Senate launched the Dietary Goals for the United States—which later became the basis of the food pyramid—the government's number-one goal has been to "increase carbohydrate consumption to account for 55%-60% of the energy (caloric) intake."

And of course, this diet, because it is government policy, has been tested—in rigorous, clinical trials funded by the NIH. Indeed, the NIH has spent at least a billion dollars on these trials, on more than 50,000 people altogether.[17] The results were that a diet low in fat with "moderate" carbohydrates does not fight any kind of disease—not heart disease, obesity, type 2 diabetes, or any type of cancer—and does not reduce mortality.

Why would the Lancet authors go back to the stage of generating a hypothesis about a diet that has already been tested and found lacking?

Weak Dietary Data

The ARIC data, upon inspection, are exceptionally weak. Participants were queried on their diet only twice (1987-1989 and 1993-1995), after which they were assumed to have continued eating exactly the same way for the next 15-plus years. The Mediterranean diet craze hit; the junk food industry exploded. During these 15 years, American eating habits changed profoundly, yet ARIC captures none of this.

Also, the ARIC dietary questionnaire contains only 66 items, compared with the 100-200 normally used in the field.[18] ARIC's questionnaire does not even appear to have been independently validated or ever published for outside evaluation, or at least it's not cited in the Lancet nor in ARIC's basic paper on diet.[19] Instead, a reader is referred to a similar questionnaire,[20] with only 61 food items, by the Harvard School of Public Health. With so few questions, many foods are missed, including such high-carbohydrate items as popcorn, pizza, and granola bars. In fact, Harvard reported[21] that "total carbohydrates" in their questionnaire could not be verified when properly adjusted for calories.

Evidence that these dietary data are flawed can be seen by the fact that the average energy intake in ARIC was only about 1500 kcals per day, which is markedly lower than what would be expected for this population (~2000 kcals would be more reasonable) and suggests that many food items were missed.

The questionnaire was also skewed towards fruits and vegetables, with 18 questions on those items compared with only nine on all kinds of fresh and processed meats. This is bound to create an unfair assessment of total meat intake and animal foods in general and more likely to bias the results in favor of a plant-based diet, as the Harvard paper did indeed find.

Reliable Evidence?

Another crucial point is that the reported death rates in the paper are not actual death rates but estimated ones, relying on numerous assumptions and incomplete data inherent in any statistical modeling exercise on a complex subject such as diet and health.

This problem is exacerbated by a serious issue discovered by Dr Zoe Harcombe,[22] a British researcher, who found that the dramatic U-shaped graph illustrating the paper's death-rate results was not based on data from the main paper but instead on different data in its appendix. Here, the authors did not divide the study participants into equal groups of carbohydrate consumption, which would have been the correct, objective approach. Instead, they created groups of unequal sizes, covering unequal ranges of carbohydrate consumption. The group eating the fewest carbohydrates (0%-30% of calories) contained only 315 people; the highest-consumption group (> 65%) comprised 715. By contrast, the groups in the middle range of consumption were populated by somewhere between 2242 and 6097 people each.

The authors gave no explanation for why they have skewed the distribution of groups in this way. By doing so, however, the death rates in the small groups are less reliable because of the limited numbers of people. "The confidence of the estimations at the extremes is more of a reach, given that few people consume extreme diets," observed Andrew Mente, an epidemiologist at McMaster University in Ontario, Canada.

Also questionable was the Lancet authors' decision to throw out part of the evidence. They eliminated any data on carbohydrate consumption from participants who developed heart disease, diabetes, or stroke before the second diet visit "to reduce potential confounding from changes in diet that could arise from the diagnosis of these diseases." The authors don't state how much data were dropped, but one has to ask: For a study examining the relationship between carbohydrate consumption and disease outcomes, aren't precisely these data most relevant? It would be essential to know, for example, what happened 15 years later to patients with heart disease who increased their carbs in response to the standard government advice. Thus, it seems that the most critical evidence in this study was deleted, replaced instead by the authors' own estimates on expected death rates.

Politics and Potential Conflicts of Interest

Given institutional biases in favor of the status quo, one could reasonably question whether this paper might be an effort on the part of the nutrition establishment to shore up the government's long-time dietary advice, which continues to advise getting a majority of calories from carbohydrates.

In fact, a number of the paper's authors have been significantly involved in the science behind those guidelines. One author, Eric Rimm of Harvard, served on the expert committee for the government's guidelines in 2010. Two other authors hail from the University of Minnesota, home to physiologist Ancel Keys, who authored the original hypothesis that a diet lower in fat and higher in carbohydrates would benefit health. In fact, Henry Blackburn, Keys' closest colleague and heir of his lab when Keys died, is one of the ARIC study leaders.[23] In recent years, he and others at the university have been concerned about defending Keys' legacy from a new generation of science that rejects the vilification of dietary fat and cholesterol.[24] One could imagine that this study might be part of that defense.

Other significant yet undeclared conflicts of interest are intellectual and financial. Harvard's Walter Willett, for example, works closely with industry-funded groups, such as Oldways[25] and the International Carbohydrate Quality Consortium,[26] which actively promote carbohydrate consumption. Willett has also been a long-time advocate of a high-grain vegetarian diet[27] and is a regular speaker on the vegan conference circuit,[28,29] as well as a senior advisor to more than one group that advocates for a vegetarian diet high in carbohydrates.[30,31]

Implications for Clinical Practice

In the end, the public loses by reading such confusing expert advice. Patients who are recovering from diabetes and losing weight sustainably on low-carb diets are waking up to headlines about how this diet will kill them. Perhaps patients will even abandon a regimen that is making them healthier. Given the apparent safety and efficacy of low-carb diets for a number of pressing nutrition-related diseases, this could actually cause patients harm.

Doctors have been prescribing the government's low-fat, high-carbohydrate diets for decades now, and Americans have largely complied.[32] Yet, we have not seen better health. Should we abandon promising new approaches backed by rigorous science in favor of weak, speculative data? The answer for an evidence-based clinical practice seems clear.

Nina Teicholz is a science journalist, author of The Big Fat Surprise, and executive director of The Nutrition Coalition, a group devoted to evidence-based nutrition policy.

Fabiano M. Serfaty, MD, is an endocrinologist, Medscape advisor, and medical director of Serfaty Clinic in Rio de Janeiro, Brazil.

The authors report no financial conflicts of interest. They both report eating diets relatively low in carbohydrates, and Dr Serfaty prescribes this diet in his practice.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.