A Feast of Observations About Diet

Mercedes R. Carnethon, PhD; Sadiya S. Khan, MD, MS


Circulation. 2020;141(14):1138-1140. 

The scientific community has devoted considerable effort in identifying how food and nutrition influence health. In 2019, the key words foods, nutrition, and diet patterns appeared 101 661 times in PubMed—markedly more often than other health behavior–related terms, including physical activity (36 590 records) and sleep (15 018 records). The lay public is eagerly seeking guidance on what and how much to eat. In 2019, 2 peer journals reported that original research articles with the highest Altmetric scores (bibliometric indices that reflect dissemination outside of traditional scientific venues) came from observational studies of specific dietary components.[1,2] Sifting through the large volume of scientific citations and media headlines is challenging for clinicians and patients who are trying to make decisions to inform their health behaviors. In the abundant field of nutritional research, some studies stand out because of their novelty, methodologic rigor, or relevance to a large or particularly high needs proportion of the population.

One such nutritional study with numerous strengths is featured in this issue of Circulation.[3] In a longitudinal study that pooled 3 cohorts and included 210 700 men and women ages 25 to 75 years, Ma and colleagues[3] report that higher intake of tofu (a condensed soy product) and isoflavones was associated with lower risks of incident nonfatal and fatal coronary heart disease (CHD). Whereas only an estimated 5% of US adults endorse vegetarian or vegan diets, more than one third (38%) of adults in India (the world's second most populous country) identify as vegetarian.[4] Consequently, findings from this study are relevant to billions of adults worldwide.

Soy and isoflavone intake were determined from self-reported food intake collected using food frequency questionnaires at baseline and repeated every 2 to 4 years over 28 years. In a small subset (n=47), self-reported isoflavone intake was validated using measurements of urinary excretion of isoflavone metabolites with a resulting Spearman rank correlation coefficient ranging from 0.18 to 0.33. Other aspects of diet, including quality and quantity, medical history, and other lifestyle behaviors that could confound the proposed associations were included in the statistical models. The multitude of strengths of the present study increases the likelihood of broad dissemination and places the study squarely at risk of overinterpretation and misinterpretation.

The urgent question most likely to arise from these findings is whether adults should increase their intake of soy products. The haste to generate dietary recommendations for public consumption disrupts the thoughtful consideration of what observational studies of diet are, and most importantly, what they are not. Longitudinal observational studies are fairly advanced along the continuum of study designs that allow for a determination of causality.[5] However, there are critical limitations of observational studies of diet that warrant caution about making causal statements. These critical limitations include the potential for misreporting dietary intake given the difficulty of isolating what people eat from the individual dietary components of their food and the potential for sociodemographic characteristics to inform an individual's food choices and other related health behaviors.

Misreporting of diet is common for at least 2 reasons. First, people do not reliably encode long-term memories that are uncoupled from emotions, and, in most cases, a mundane behavior such as eating is unlikely to elicit emotion. Forgetfulness can be mitigated using certain techniques such as providing cues (food models), using food diary mobile apps for real-time logging of dietary intake, engaging trained professionals in data collection, or using trained motivational interviewing techniques.[6] However, these techniques can be time-consuming and resource-intensive, thus limiting their practical implementation in large studies. Second, researchers in the social and population sciences have found that when asked, people prefer to represent themselves in a favorable light. The result is a type of information bias, alternately termed "social desirability bias," "lying bias," or "impression management," depending on the field of study, that can yield falsely favorable self-reports.[7] Given stigma around weight management, dietary behaviors are uniquely prone to deliberate misreporting.

Both scenarios misclassify the dietary exposure and can obscure an association when there really is one (bias toward the null) or falsely identify an association when there is none (bias away from the null). Device-based assessment has enhanced the precision of exposure assessment for health-related behaviors such as physical activity and sleep. By comparison, device-based assessment of diet relies primarily on self-reported inputs.[8] Devices remain impractical for long-term and large-scale studies and are unlikely to replace food frequency questionnaires or dietary recall methodologies.

The other important factor is that healthy (as well as unhealthy) behaviors cluster together, which can yield residual confounding. Adults who adopt high-quality diets and adhere to portion control are often the same adults who maintain a physically active, nonsmoking lifestyle and have more education and other socioeconomic resources than other adults. As a result, they have better overall metabolic profiles and are predisposed to lower rates of CHD than other adults. In the present study, these observations were borne out as participants in the uppermost quartile of soy intake had fewer CHD risk factors. The authors factored these patterns into their statistical models, but residual confounding remains relevant. Borrowing from another prominent example, the "healthy user bias" is one hypothesis posed to explain the disparate findings between observational studies of hormone replacement therapy showing protections for cardiovascular health[9] versus major clinical trials that suggested harm.[10]

Together, these limitations justify cautions against making recommendations based on observational studies alone. At best, longitudinal observational studies reporting strong, consistent, and biologically plausible associations can satisfy most of the criteria set out by Sir Bradford Hill for inferring causality.[5] Still, they fall short for the aforementioned reasons. Although the authors of the present study acknowledge these limitations, the media driven by the public's interest commonly overlooks these cautions when they digest and disseminate study findings.

Two studies published in 2019 highlight another media bias towards covering nutrition studies that focus on popular or controversial foods. The study published on the association of eggs and dietary cholesterol with mortality earned an Altmetric score of 4023[2] in contrast with a study describing a plant-based diet that earned an Altmetric score of 36.[11] Valid concerns about media misinterpretation and the public's rush to read headlines as recommendations has prompted opinion leaders to advise against high-profile publication of observational studies of nutrition. However, blindly deciding not to publish or disseminate research findings from nutritional epidemiology studies interrupts the scientific process. Rather, consistent findings across studies should be the benchmark for justifying the expense of large-scale randomized controlled trials.

Jumping directly to randomized trials without a solid base of observational work is unjustified. Although randomized trials are the gold standard, they, too, are subject to limitations given the difficulty of measuring and reporting behaviors and adhering to interventions as complex as diet over a prolonged time period. Of note, monitoring adherence to the intended dietary intervention can be far more challenging than pill counts used in pharmaceutical studies investigating the effect of medications. For example, in the landmark DASH trial (Dietary Approaches to Stop Hypertension), participants were required to keep food diaries and eat in a metabolic kitchen under the observation of the investigators. The investigators additionally measured 24-hour urinary output to confirm urinary biomarkers that reflected adherence to the diet.[12] Whereas clearly an outstanding effort that yielded results that inform blood pressure management, such an expense would not have been justified without the observational studies that preceded it.

Even if the scientific community and the media treat observational nutrition study findings in a responsible manner, clinicians remain at the forefront of discussing behavior change with patients. The large majority of clinicians are constrained by time or ill-equipped to dive more deeply than a recommendation to pursue a heart-healthy diet as recommended in guideline documents.[13] Nutrition counseling, coupled with evidence-based medications, is an integral part of primordial, primary, and secondary prevention for CHD. In most cases, discussing methodological limitations of nutritional epidemiology studies with patients may be confusing and counterproductive. Delaying discussions for the release of practice guidelines and recommendations fails to meet the needs of patients in the short term and misses an opportunity for behavior change counseling, particularly when fewer than 1% of Americans achieve an ideal diet for cardiovascular health as assessed by the American Heart Association.[14]

A responsible approach is to capitalize on one of the limitations of observational studies of diet, which is the clustering of healthful behaviors. When counseling on the intake of soy products for CHD risk reduction, highlighting the potential benefits of soy intake alongside other dietary guidelines[15] can leverage the strengths of well-done observational research without overpromising the benefits of specific behavior changes. Despite our desire for an easily digestible message about whether soy intake can reduce the risks of CHD, the evidence needed to make or reverse current recommendations remains a work in progress. It is through well-done studies such as that featured here that we can come closer to answering that question.