Article, see p 1127

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 colleagues³ 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 health⁹ 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 4023² 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 guidelines¹⁵ 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.

None.

The opinions expressed in this article are not necessarily those of the editors or of the American Heart Association.

https://www.ahajournals.org/journal/circ

文章，请参阅第1127页

科学界在确定食物和营养如何影响健康方面付出了巨大的努力。2019年，食品，营养和饮食方式等关键词在PubMed中出现101661次，这一数字比其他与健康行为相关的术语更为频繁，包括身体活动（36590条记录）和睡眠（15018条记录）。普通民众急切地寻求关于吃什么和吃多少的指导。在2019年，有2家同行期刊报道说，具有最高Altmetric得分（反映传统科学场所之外传播的文献索引）的原始研究文章来自对特定饮食成分的观察研究。^1,2对于试图做出决定以告知其健康行为的临床医生和患者而言，筛选大量的科学引文和媒体头条具有挑战性。在营养研究的广泛领域中，一些研究因其新颖性，方法严谨性或与大量或特别高的人口需求比例有关而脱颖而出。

本期“循环”刊载了一项具有许多优点的营养研究。³在一项涉及3个队列的纵向研究中，包括210700名年龄在25至75岁之间的男性和女性，Ma及其同事³报告说，豆腐（一种浓缩大豆产品）和异黄酮的摄入量较高，从而降低非致命性和致命性风险冠心病（CHD）。估计只有5％的美国成年人支持素食或纯素食，而印度（世界第二人口大国）的成年人中有超过三分之一（38％）的人认同素食。⁴因此，这项研究的发现与全世界数十亿成年人有关。

从基线时使用食物频率问卷收集的自我报告的食物摄入量确定大豆和异黄酮的摄入量，并在28年中每2至4年重复一次。在一个很小的子集中（n = 47），通过测量异黄酮代谢产物的尿排泄量来验证自我报告的异黄酮摄入量，由此得出的Spearman等级相关系数为0.18至0.33。饮食的其他方面，包括质量和数量，病史和其他可能混淆拟议的关联的生活方式行为，都包括在统计模型中。本研究的众多优势增加了广泛传播的可能性，并使研究面临过度解释和曲解的风险。

这些发现最可能引起的紧迫问题是成年人是否应增加大豆产品的摄入量。急于为公众消费制定饮食建议会扰乱对饮食观察研究的思考，最重要的是，饮食观察不是。纵向观察研究在研究设计的连续性方面相当先进，可以确定因果关系。⁵但是，饮食观察研究存在严重局限性，因此在做出因果关系陈述时应谨慎行事。这些关键的局限性包括由于难以从人们的食物中分离出饮食成分而错误地报告饮食摄入量，以及可能具有社会人口统计学特征来告知个人食物选择和其他相关健康行为的可能性。

饮食不实是很常见的，至少有两个原因。首先，人们不能可靠地编码与情感分离的长期记忆，并且在大多数情况下，诸如饮食之类的平凡行为不太可能引起情感。可以使用某些技术来减轻健忘，例如提供提示（食物模型），使用食物日记移动应用程序实时记录饮食摄入量，让受​​过训练的专业人员参与数据收集或使用受过训练的动机访谈技术。⁶但是，这些技术可能很耗时且占用大量资源，因此限制了它们在大型研究中的实际实施。其次，社会和人口科学领域的研究人员发现，当被问及时，人们倾向于以有利的眼光来代表自己。结果是一种信息偏见，根据研究领域的不同，也被称为“社会期望偏见”，“说谎偏见”或“印象管理”，这可能会产生错误的自我报告。⁷鉴于对体重控制的污名化，饮食行为极易出现故意误报的情况。

两种情况都将饮食暴露分类错误，并且在确实存在饮食时会掩盖关联（偏向零值），而在没有饮食存在时会错误地识别关联（远离零值时偏）。基于设备的评估提高了针对与健康相关的行为（如体育活动和睡眠）的暴露评估的准确性。相比之下，基于设备的饮食评估主要依赖于自我报告的投入。⁸对于长期和大规模的研究，设备仍然不切实际，并且不太可能取代食物频率调查表或饮食召回方法。

另一个重要因素是健康的（以及不健康的）行为聚集在一起，这可能会产生残留的混淆。与其他成年人相比，接受高质量饮食并坚持控制饮食的成年人通常与成年人保持相同的成年人，他们保持身体活跃，不吸烟，并拥有更多的教育和其他社会经济资源。结果，它们具有更好的整体代谢特征，并且比其他成年人更容易患冠心病。在本研究中，这些观察结果得到了证实，因为大豆摄入量最高的四分位数的参与者具有较少的冠心病危险因素。作者将这些模式纳入了他们的统计模型，但是剩余的混淆仍然很重要。借用另一个突出的例子，⁹与表明危害的主要临床试验相比。¹⁰

总之，这些局限性证明了谨慎的态度，避免仅根据观察性研究提出建议。最好的情况是，纵向观察研究报告的牢固，一致且生物学上合理的关联可以满足Bradford Hill爵士提出的推断因果关系的大多数标准。⁵但是，由于上述原因，它们仍然不足。尽管本研究的作者承认了这些局限性，但当公众消化和传播研究结果时，由公众利益驱动的媒体通常会忽略这些注意事项。

2019年发表的两项研究突显了另一种媒体偏向于涵盖针对流行或有争议食品的营养研究。该研究发表的关于鸡蛋和饮食中胆固醇与死亡率的关系的研究获得了4023 ²的Altmetric得分，而一项研究描述了植物性饮食获得的36的Altmetric得分^。11对媒体误解的合理关注以及公众急于阅读有关建议的标题，这些提示促使舆论领袖建议不要高调发表营养观察性研究。然而，盲目决定不发表或传播营养流行病学研究的研究结果中断了科学过程。相反，跨研究的一致发现应作为证明大规模随机对照试验费用合理的基准。

没有可靠的观察工作基础而直接跳入随机试验是不合理的。尽管随机试验是金标准，但由于难以测量和报告行为以及长期坚持饮食等复杂干预措施，因此它们也受到限制。值得注意的是，与对药物效果进行研究的药物研究相比，监测对饮食干预的依从性可能要困难得多。例如，在具有里程碑意义的DASH试验（停止高血压的饮食方法）中，要求参与者在研究者的观察下保存食物日记并在新陈代谢的厨房中进食。研究人员还测量了24小时的尿量，以确认反映了饮食习惯的尿液生物标志物。¹²显然，付出了巨大的努力才能取得有益于血压管理的结果，但如果没有这样的观察研究，这种支出是不合理的。

即使科学界和媒体以负责任的态度对待观察性营养研究的发现，临床医生仍处在与患者讨论行为改变的最前沿。绝大多数临床医生受时间限制或设备不足，无法潜水，而不是指南文件中建议的追求心脏健康饮食的建议。¹³营养咨询以及循证药物是冠心病原始，一级和二级预防的组成部分。在大多数情况下，与患者讨论营养流行病学研究的方法学局限性可能会造成混淆和适得其反。拖延讨论以发布实践指南和建议的建议无法在短期内满足患者的需求，并且错过了进行行为改变咨询的机会，尤其是当少于1％的美国人实现了理想的心血管健康饮食时（美国评估）心脏协会。¹⁴

负责任的方法是利用饮食观察研究的局限性之一，即健康行为的聚集。在就减少冠心病风险的大豆产品摄入量提供咨询意见时，突出大豆摄入量与其他饮食指南的潜在益处¹⁵可以利用做得好的观察研究的优势，而不会过度承诺特定行为改变的益处。尽管我们希望获得关于大豆摄入量是否可以降低冠心病风险的容易理解的信息，但提出或逆转当前建议所需的证据仍在研究中。通过出色的研究（如此处所述），我们可以更接近回答这个问题。

没有。

本文表达的观点不一定是编辑者或美国心脏协会的观点。

https://www.ahajournals.org/journal/circ