A 2019 review of red and processed meat research used GRADE to rate evidence, raising questions about how drug-trial standards apply to diet.
The dispute over a 2019 recommendation to keep eating red and processed meat raises a broader question: can a system built largely around clinical trials of treatments fairly assess long-term nutrition research? A series of papers in the Annals of Internal Medicine concluded that adults who eat meat need not reduce their intake. Critics argued that the evidence grading behind that conclusion gave too little weight to the limits of nutrition studies and to findings from observational research.
The papers prompted strong criticism from nutrition researchers. The chair of Harvard’s nutrition department called the recommendation “a very irresponsible public health recommendation.” Walter Willett, a former chair of the department, described it as “the most egregious abuse of data I’ve ever seen.” Those reactions reflect a disagreement about methods as much as about meat: how should researchers judge evidence when the strongest kind of experiment is impractical?
Why trial standards can be difficult to apply
One criticism concerned the use of GRADE, short for Grading of Recommendations, Assessment, Development, and Evaluation. The papers relied heavily on this system to rate certainty in the evidence. Critics argued that applying it to nutrition studies was a mistake because it was designed with drug trials in mind. The authors’ approach and its implications were discussed in a video about the GRADE assessments.
In drug research, randomized, double-blind, placebo-controlled trials can help establish whether a treatment causes benefits or harms. GRADE typically rates observational studies as having low or very low certainty, in part because they can be affected by differences between the people being compared. That caution makes sense in many settings. The problem, critics say, is that applying the same expectations to dietary exposures can discount evidence in an area where long-term randomized trials are rarely feasible.
Researchers cannot realistically assign people to eat a particular diet every day for decades and then track outcomes such as heart attacks or cancer. Nor can participants be blinded to whether they are eating red meat. The chair of Harvard’s nutrition department made this point by comparing food research with a drug trial: “For dietary and lifestyle factors, it’s impossible to use the same standards for drug trials.” The same challenge applies to studying behaviors and exposures such as smoking, exercise, air pollution, or sugar intake over many years.
Yet the meat papers downgraded studies in part because they lacked blinding. As one set of criticisms asks, people in a nutrition study generally know what they are eating. Blinding can reduce bias in some experiments, but it cannot be used in the same way for ordinary foods. Harvard’s nutrition researchers also questioned whether this feature of GRADE was appropriate for evaluating the available diet evidence.
Limits of randomized dietary trials
Observational studies have limitations: they can identify associations but cannot, by themselves, establish that a food caused a health outcome. Researchers try to account for other factors that may explain those associations, and different evidence-grading approaches weigh the strengths and limits of each study design. Critics say that the inability to conduct ideal randomized trials should lead to careful interpretation of all available evidence, rather than the assumption that there is no useful evidence.
Long-term trials also face a practical problem: people may not follow the assigned diet. Large dietary trials, including the Women’s Health Initiative and the Multiple Risk Factor Intervention Trial, illustrate the difficulty. Participants in different study groups can end up eating more similar diets than researchers intended. When that happens, a trial may show little difference in health outcomes even if the diets would have mattered under stronger adherence.
The source article compares this issue with a randomized trial in which middle-aged men were assigned either to receive advice to quit smoking or to receive no special instruction. During the study, 13.7% of the advice group died, compared with 12.9% of the comparison group. Those figures alone do not show that smoking is harmless. Participants were assigned to receive advice, not to stop smoking. At the last follow-up, the advice group smoked an average of eight cigarettes a day, compared with 12 in the other group. The limited separation in behavior makes it hard to draw conclusions about the effect of quitting. The trial’s design and the difficulty of changing behavior help explain why an experiment can fail to detect an effect.
Similarly, the source notes that randomized trials have not always shown an effect of smoking on mortality. That result does not erase the wider body of evidence about smoking. It shows why the results of a trial need to be interpreted in light of what participants actually did and what the study could measure.
Choosing a suitable way to assess evidence
GRADE is not the only available system. NutriGrade, for example, was developed to assess research on nutrition and lifestyle factors. Reviews challenged the fit between GRADE and the meat analyses, while other researchers have raised concerns about how evidence standards can shape dietary guidance. The authors of the meat papers said their recommendations differed from other guidelines because they used GRADE. Whether the method was the main reason for that difference is part of the debate.
The article also raises questions about potential industry influence. It points to the lead author’s earlier work on sugar and soda, and links to reporting about past food-industry ties and a prior study funded at the request of soda and candy companies. These details are relevant to transparency, but they do not, by themselves, settle whether the analysis was valid. The methods and evidence need scrutiny on their own merits.
Critics warn that if evidence is judged mainly by whether it comes from randomized trials, the same weakness could be used to cast doubt on established findings about exposures that cannot readily be assigned in experiments. Examples raised in the debate include secondhand smoke, air pollution, inactivity, and trans fats. A discussion of evidence grading makes the broader point: a lack of feasible randomized trials does not mean researchers have no evidence, but it does mean they need methods suited to the question.
That matters for clinicians and readers because a low certainty rating is not the same as proof that two choices have identical health effects. One senior co-author of the meat papers, when asked whether a clinician could say a salad was healthier than a bowl of sugar, said people’s preferences would matter because the evidence quality was low. Critics see that as a consequence of applying a grading tool that may not fit the evidence available.
The series this discussion came from examined how industry can affect dietary guidance. Its earlier installments covered sugar recommendations and observational research: “The Corporate Playbook behind Undermining Dietary Guidelines,” “Big Sugar Twists the Science behind the Dietary Guidelines,” and “When Should We Rely on Observational Studies?” The practical takeaway from the meat debate is narrower than a verdict on any one food: evidence standards should account for how a study was designed, what participants actually did, and which research methods are realistic for the question.
