点击显示 收起
Cornell University, Division of Nutritional Sciences, N204 MVR, Ithaca, NY 14053, E-mail: tcc1{at}cornell.edu
Dear Sir:
The important paper by Hu et al (1) on findings from the Nurses' Health Study (NHS) deserves comment, especially now that these findings have been publicized in the popular media (2). In my opinion, several highly questionable interpretations of the data have been made. The study by Hu et al purports to show, in contrast with previous evidence, that increased consumption of animal protein is associated with decreased risk of ischemic heart disease (IHD), although this observation is only marginally significant.
The importance of this paper can hardly be overemphasized. It not only reports evidence mostly contrary to the existing literature, but also reveals an important shortcoming of this widely reported study, both in the experimental design and in the method of analysis. Dietary protein (as quintile means) ranged from 14.7% to 24.0% of energy and an overwhelming 80% of this protein was from animal sources; these findings suggest the consumption of a virtually carnivorous diet. This becomes even more evident in view of the relatively high intake of total fat (36–41% of energy, as quintile means) and the very low intake of fruit and vegetables and dietary fiber of the study cohort. Even the intake of protein in the lowest quintile was 50% more than recommended (3), perhaps close to 90–100% more than the requirement when the unusually high concentration of high-efficiency animal protein and the statistical construction of the recommended dietary allowances are taken into account.
This dietary experience contrasts sharply with the findings of the original international correlation studies (4, 5) that showed impressive associations between selected dietary factors and chronic degenerative diseases. The contrast between the diets of the cohorts in the international correlation studies and in the NHS can be illustrated by comparing the respective protein-fat associations in these cohorts. The correlation of total fat with animal protein in the international correlation study diets was 90–95%, whereas in the NHS dietary range, the correlation was small and nonsignificant (15%). Furthermore, both animal fat and animal protein, but not plant fat, were found to be tightly associated with breast cancer in the international correlation studies.
These earlier findings from the international correlation studies suggest that the incidence of chronic degenerative disease was associated with animal-based food consumption over a broad range of intakes—from very low amounts—at the expense of plant-based food consumption, and not necessarily with the consumption of any particular nutrient or nutrient group. The NHS dietary experience, in contrast, differs because of the uniformly high consumption of animal-based foods, thus severely limiting a meaningful investigation of the comprehensive effects either of this food group or of individual foods and related nutrients within this group. As for virtually every other study of Western subjects, the NHS therefore does not permit a discriminating analysis of the diet-disease associations originally observed in the international correlation studies; the necessary range of intake of these foods and their respective nutrients is missing. Not only is the detection of meaningful disease-related associations for individual foods and their indicator nutrients compromised, but the prospect of making paradoxical observations is increased and the investigation of the more comprehensive dietary effects is ignored.
Although the statistical method used in this study is popular and well accepted, it is also based on the highly unlikely assumption that the independent effects of individual nutrients are primarily and comprehensively responsible for the multiplicity of disease outcomes. This method originally was meant for testing the safety and efficacy of pharmaceuticals, not for evaluating the comprehensive effects of multiple dietary and nutritional components. In the study by Hu et al, for example, inclusion of various "known" risk factors in the analytic model cannot eliminate the problem of residual confounding, as is often implied. Also, as shown in Table 4 of the article, the group in the highest quintile of protein intake and at lowest risk for IHD also paradoxically smoked less, consumed less alcohol, exercised more, had lower body mass, and consumed more fruit, vegetables, dark bread, dietary fiber, folate, and polyunsaturated fat but less white bread, sweets, and desserts than the groups who consumed less protein—factors for which significant trends among quintiles of protein intake were seen and which are thought to be protective of IHD. In contrast, the high-protein consumers also had a longer history of hypertension and more family history of IHD, with each factor presumably contributing to increased IHD risk.
Within a dietary range in which one food group so predominates, it makes no sense to me that it is possible to reliably detect the so-called independent associations of individual constituents of this group when it can be expected that they share the same disease outcomes and when there are so many difficult-to-measure and interacting risk factor exposures. When will it be understood that it is the total diet and the aggregate and comprehensive effects of large food groups that make the greatest contribution to the maintenance of health and prevention of disease? The sort of reductionism embodied in the interpretation of data from this cohort runs the risk of severely misleading discourse on meaningful public health and public policy programs.
REFERENCES