Food frequency dietary assessment: How bad is good enough?
Food frequency dietary assessment: How bad is good enough?
复制标题
DOI：
10.1093/aje/154.12.1087
复制
发表时间：
2001-12-15
影响因子：
5
通讯作者：
Byers, T
中科院分区：
文献类型：
--
作者：
Byers, T
本期杂志刊登了一系列关于饮食流行病学的论文。虽然这些论文可能无法帮助我们决定在接下来的节日食品季节选择哪些派对自助餐项目，但它们确实代表了饮食流行病学领域当前状况的一个横截面。这一领域包括越来越多的研究，使用不同的食物频率方法进行饮食评估，同时，研究继续严格审查同样的方法的有效性。本期（1-3）中的食物频率验证研究及其随附的评论（4，5）对建立关于正确使用食物频率方法进行膳食估计的共识的过程做出了值得注意的贡献。也许没有其他流行病学学科像饮食流行病学那样吸引了如此多的公众关注，同时也吸引了如此多的科学批评。这是因为这种风险既关系到公众的切身利益，又是出了名的难以衡量。一项小型的初步研究表明，西兰花的食用频率对乳腺癌生存率的影响可能会成为头条新闻，而其他重要主题的研究结果则不会成为头条新闻。围绕饮食流行病学研究结果差异的宣传导致了科学和公众对该领域的冷嘲热讽。一些过度的宣传可以归咎于流行病学家，但大多数是公众对饮食和健康信息的强烈需求的结果。由于我们已经学到了关于饮食和疾病的流行假设可能是无效的这一有时困难的教训，饮食测量的有效性已被公开质疑。实验营养科学家，谁受过训练，认为营养素作为精确定量的因素，往往嘲笑饮食流行病学的方法。当营养科学家思考流行病学的发现，将长期饮食与疾病联系起来时，他们是基于对诸如“你多久吃一次豌豆？”他们的怀疑是可以理解的。实验营养科学家和营养流行病学家之间的分歧似乎不仅是因为他们对暴露的精确度标准不同，而且还因为他们使用了不同的终点。虽然实验营养学可以精确地测量营养素暴露，但其结果，如实验动物的疾病或人类志愿者的短期生物标志物，通常与感兴趣的人类疾病无关。相比之下，流行病学饮食研究将相关疾病终点与相关饮食暴露联系起来，但根据实验营养科学的标准，饮食测量相当粗糙。流行病学家也怀疑食物频率自我报告支持病因推断的能力。人们不仅关注食物频率报告中的随机测量误差，而且还关注报告与疾病或疾病相关因素有关的饮食的偏差。这种担忧部分来自饮食流行病学中观察到的小规模关联，通常相对风险为1.5-1.9。饮食流行病学的挑战在于，由于不完善的饮食评估，很难知道1.5的相对风险是否应该被解释为一个被随机测量误差稀释的更大相对风险的指标，或者它是否是一个被偏倚或混淆夸大的更小相对风险的高估。我们对食物频率饮食评估类型的有效性有足够的了解吗？
This issue of the Journal features a diverse set of papers on dietary epidemiology. Although these papers might not help us decide which party buffet items to select in the coming days of the holiday food season, they do represent a cross section of the current status of the field of dietary epidemiology. This field includes a rapidly growing number of studies using variations of the food frequency method of dietary assessment and, at the same time, studies that continue to critically examine the validity of that very same method. The food frequency validation studies in this issue (1–3) and their accompanying commentaries (4, 5) are noteworthy contributions to the process of building a consensus about the proper use of the food frequency method of dietary estimation. There is perhaps no other epidemiologic discipline that has attracted as much public attention and, at the same time, as much scientific criticism as has dietary epidemiology. That is because the exposure is both of immediate interest to the public and notoriously difficult to measure. A small preliminary study suggesting an effect of the frequency of broccoli consumption on breast cancer survival might make headlines when even more definitive findings from studies on other important topics would not. Publicity surrounding variations in findings between dietary epidemiologic studies has led to both scientific and public cynicism about the field. Some excess publicity can be blamed on epidemiologists, but most has been a consequence of the high public appetite for information about diet and health. As we have learned the sometimes difficult lessons that popular hypotheses about diet and disease may well be null, the validity of dietary measurements has been openly questioned. Experimental nutritional scientists, who have been trained to consider nutrients as precisely quantitated factors, often scoff at dietary epidemiologic methods. When nutritional scientists ponder epidemiologic findings relating long-term diet to disease based on dietary estimates from such questions as “How often do you eat peas?” their skepticism is understandable. The disagreement between experimental nutritional scientists and nutritional epidemiologists appears to result from not only their different standards of precision for exposures but also their use of different endpoints. Although experimental nutrition can precisely measure nutrient exposures, their outcomes, such as disease in laboratory animals or short-term biomarkers in human volunteers, are often only tangentially related to the human diseases of interest. In contrast, epidemiologic dietary studies relate the relevant disease endpoints to the relevant dietary exposures, but the measures of diet are quite crude according to the standards of experimental nutrition science. Epidemiologists have also been skeptical about the ability of food-frequency self-reports to support etiologic inference. There has been concern about not only random measurement errors in food frequency reports but also biases in reporting diet related to disease or disease-linked factors. This concern comes in part from the small size of associations observed in dietary epidemiology, often with relative risks of the order of 1.5–1.9. The challenge in dietary epidemiology is that, because of imperfect dietary assessment, it is difficult to know whether a relative risk of 1.5 should be interpreted as an indicator of a much larger relative risk that has been diluted by random measurement error, or whether it is an overestimation of an even smaller relative risk that has been inflated by bias or confounding. Do we yet have enough understanding of the validity of the types of food frequency dietary assessments …