This study comparing popular WLDs with the usual dietary intakes of Americans resulted in several important findings. First, popular WLDs promulgated over the past 10 years varied greatly in their macronutrient composition and micronutrient content. In addition, when taken as a whole, the WLDs varied substantially from the usual intake of Americans in several important ways. However, when compared on the basis of Euclidean distances, this study demonstrated that the usual diets of a representative sample of Americans were actually closer to the dietary recommendations of the 17 popular WLDs taken as a whole than the WLDs were to each other.
Our results indicate that the 17 diets included in this study varied widely with regard to many of the nutrient variables included in the analysis. For example, individual fatty acid contents (saturated, monounsaturated, and polyunsaturated) varied by at least a factor of 6, and most nutrients varied by at least a factor of 4 among the diets. Of interest (considering that these were all weight-loss diets), diets differed by a factor of 3 in energy content, varying from just over 750 kcal/d (Eat to Live, vegetarian for aggressive weight loss) to greater than 2200 kcal/d (Zone).
The mean nutrient composition of the WLDs was in greater agreement with published dietary guidelines[21,28] than mean intakes for individuals calculated from CSFII data. While mean intakes of polyunsaturated fat, vitamin E, and sodium from CSFII were very close to the mean values calculated from the WLDs, mean intakes of several nutrients associated with a higher risk of disease in previous studies (energy, saturated fat, cholesterol, and GL) were substantially higher in individuals based on CSFII data. Conversely, mean intakes of several nutrients associated with a reduced risk of disease (dietary fiber, folate, and calcium) were much lower in individuals compared with the WLDs.
Despite the apparent differences in nutrient composition of the WLDs considered together and nutrient intakes of individuals (and contrary to what we hypothesized), this study demonstrated that the usual diets of a representative sample of Americans were actually more similar to the WLDs taken as a whole than the WLDs were to each other. We originally speculated that based on Euclidean distances, the popular diets included in this study would be clustered at a considerable distance from a second cluster representing the usual intakes of Americans. However, this proved not to be the case. In fact, the probability that an individual’s diet had a distance from at least 1 of the WLDs that was greater than the distance between the 2 most dissimilar WLDs was less than 1%.
What are the possible ramifications of these findings? Because we initially speculated that the WLDs would form a cluster at some distance from the usual diets of Americans, we postulated that an individual seeking to lose weight could choose any one of the popular diets at random and (assuming one of the WLD diets was optimal for that person) be closer to consuming their optimal diet than if they stayed on their current diet. In other words, the choice of diets would be a far less critical factor than would simply adopting one of the diets.
The results of this study indicated otherwise. Because the distance between the usual diets of Americans and the WLDs considered together (6.04) was similar to the mean distance among the popular diets (6.12), the choice of a WLD may, in fact, be more critical than we hypothesized. Assuming that one of the diets is best for an individual, he/she might be more successful in losing weight by selecting a diet that is most dissimilar to his/her usual diet than in selecting one that is very similar. An individual trying to lose weight without success that is already ‘near’ (a small Euclidean distance from) one of these WLDs may be more successful by choosing a diet that is farther (a larger Euclidean distance) from his/her usual diet. There is evidence from previous studies that when individuals trying to lose weight adopt new dietary patterns that vary markedly from their spontaneous eating patterns, weight loss may be enhanced. When one couples this with data suggesting that WLDs varying radically in composition do not produce radically different weight losses on average, a subject (baseline diet)-by-treatment (recommended WLD) interaction seems quite plausible. That is, subjects eating, for example, in an Atkins-like way may do best on an Ornish-like diet, and vice versa. While acknowledging that it may be challenging for individuals to radically alter life-long dietary patterns, we believe this hypothesis deserves consideration for future research.
Within this article, we calculated Euclidean distances instead of the more traditional statistical measure of Mahalanobis distance. Our justification for this approach was 2-fold. First, all dietary variables were standardized prior to calculating distances. Mahalanobis distance is preferable when variables are measured in different units and have differing variances. However, since we standardized the variables prior to calculating distances, all variables were “unit-less” and the variances were set to 1 for all variables. Our second justification for using Euclidean distances was that we were not interested in the distances with regard to the association among the variables. Mahalanobis distance takes into account the correlation structure among the variables studied. For the purposes of this article, we were concerned with simple distances, not distances relative to a pattern of association. We did conduct a sensitivity analysis using Mahalanobis distances, and none of our results changed appreciably using this measure.
Our conclusions are limited by the absence of indication in the CSFII data regarding how many individuals were actively trying to lose weight, whether they were successful in doing so, or whether they were attempting to follow any particular dietary pattern. The impact of such data on our results, however, is likely to be limited, because individuals’ statements of intention to lose weight correlate poorly with adherence to weight-loss dietary patterns and with actual weight loss. It should also be noted (as discussed previously) that some of the diet books did not include sample menus. In those cases, we carefully constructed sample menus in strict accordance with the respective diet’s macronutrient recommendations, as presented in its book. Finally, we analyzed the WLDs as presented in the books rather than assessing dietary intake in persons on each diet. Our objective was to analyze the WLDs as recommended in each book, not how compliant persons following the diets might be.
In summary, based on Euclidean distances, the popular WLDs included in this study were more similar to the spontaneous diets of a representative sample of Americans than originally hypothesized. This could have important implications with regard to the choice of diets for individuals attempting to lose weight. Further studies are needed to determine the impact on weight loss of adopting diets of varying Euclidean distances from spontaneous eating patterns.