Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
European Journal of Clinical Nutrition (2002) 56, 37–43 ß 2002 Nature Publishing Group All rights reserved 0954–3007/02 $25.00 www.nature.com/ejcn ORIGINAL COMMUNICATION Lipid, protein and carbohydrate intake in relation to body mass index A Trichopoulou1*, C Gnardellis1, V Benetou1, P Lagiou1, C Bamia1 and D Trichopoulos1,2 1 Department of Hygiene and Epidemiology, University of Athens Medical School, Athens, Greece; and 2Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts, USA Objective: To examine whether the same amount of energy intake has different consequences on body mass index (BMI), depending on the source of energy from specific macronutrients. Design: Cross-sectional study, in the context of the European Prospective Investigation into Cancer and nutrition (EPIC). Setting: Communities all over Greece. Subjects: A total of 27 862 apparently healthy volunteers, men and women, ages 25 – 82 y. Interventions: None. Methods: Dietary information was collected through an interviewer-administered semi-quantitative food frequency questionnaire. In the context of a cross-sectional analysis, we calculated changes of BMI per increments of energy intake from protein, carbohydrates, saturated, polyunsaturated and monounsaturated lipids and ethanol, controlling for mutual confounding and other confounders, among all participants, and after exclusion of under-reporters and=or those on a diet. Results: Protein intake was positively associated with BMI. The association was evident when nutrients were not mutually adjusted for and increased after mutual adjustment among nutrients (b ¼ 0.80 kg=m2 per 418.4 kJ or 100 kcal increment, 95% confidence interval 0.55 – 1.06 for men, and b ¼ 1.59, 95% CI 1.30 – 1.88 for women), as well as after exclusion of underreporters and=or those on a diet. The effects of other macronutrients were less substantial or consistent. Conclusions: There is evidence indicating that protein intake is conducive to obesity. Moreover, our data suggest that neither saturated or monounsaturated lipids nor carbohydrates are likely to play a major role in increasing BMI over and beyond that indicated by their energy content. Sponsorship: EPIC-Greece is supported by the Europe Against Cancer Program of the European Commission and the Greek Ministry of Health. European Journal of Clinical Nutrition (2002) 56, 37 – 43. DOI: 10.1038=sj=ejcn=1601286 Keywords: protein; lipids; carbohydrates; body mass index; obesity Introduction A high body mass index (BMI) increases morbidity and mortality from several diseases, including cardiovascular *Correspondence: A Trichopoulou, Department of Hygiene and Epidemiology, University of Athens Medical School, 75 M. Asias Street, GR-115 27, Athens, Greece. E-mail: [email protected] Guarantor: A Trichopoulou. Contributors: AT is the principal investigator of the EPIC project in Greece and generated the working hypothesis. CG and CB undertook the statistical analyses. VB and PL are EPIC project managers, dealt with the biomedical and epidemiological issues and were responsible for the drafting of the paper. DT is the senior epidemiologist of the EPIC-Greece study. Received 7 March 1999; revised 30 May 2001; accepted 6 June 2000 diseases, diabetes mellitus and some forms of cancer (Garrow, 1986; Bray, 1996; World Cancer Research Fund & American Institute for Cancer Research, 1997; Seidell, 1998b). In addition, obesity is generally considered as an aesthetic drawback in most modern societies. Reduced energy intake and increased physical activity are the only established approaches to reducing BMI that can be sustained over long periods of time (Klesges et al, 1992; Nelson & Tucker, 1996; Rising et al, 1994; Schulz & Schoeller, 1994; De Groot & van Staveren, 1995; Parker et al, 1997; Trichopoulou et al, 2000). There is also an extensive literature, from both animal experiments and human studies, addressing the question whether specific energy-generating nutrients affect BMI differentially, after accounting for their energy content (Slattery et al, 1992; Stubbs et al, 1997; Macronutrient intake and BMI A Trichopoulou et al 38 Willett, 1998a; West & York, 1998). There appears to be no agreement among investigators, with the majority considering that energy from lipids is more conducive to an increased BMI than energy from other macronutrients (Astrup et al, 1997; Bray & Popkin, 1998; Hill et al, 2000), whereas a differential effect is not accepted by other investigators (Willett, 1998a; Seidell, 1998a). In the context of the Greek segment of the European Prospective Investigation into Cancer and nutrition (EPIC), we have undertaken a large cross-sectional study, relying on validated questionnaires for nutritional intake and physical activity, and standardised somatometric measurements. The objective of this study was to assess whether the same amount of energy intake has different consequences on BMI, calculated as the ratio of weight in kilograms over square of height in metres. depending on the source of energy from protein, carbohydrates, various forms of dietary lipids or ethanol. Methods A validated semi-quantitative food frequency questionnaire was administered, from 1994 to 1999, by specially trained interviewers to 27 862 apparently healthy adults, men and women, aged 25 – 82 y. The study subjects were recruited from regions all over Greece in order to participate in the Greek component of the EPIC study. EPIC is a multi-country prospective cohort study investigating the role of nutrition and other lifestyle and environmental factors in the aetiology of cancer and other chronic diseases (Riboli, 1992). Each participant signed an informed consent form before enrolment. Somatometric measurements concerning weight, height, sitting height, as well as waist and hip circumferences were also undertaken. Subjects were measured without shoes, lightly clothed and without any restrictive underwear. The semi-quantitative food frequency questionnaire was validated during the pilot phase of the study (Gnardellis et al, 1995; Katsouyianni et al, 1997). The questions covered the average frequency of consumption of approximately 150 food items and beverages over a period of 1 y. Standard portion sizes were used for the estimation of consumed quantities (Trichopoulou, 1992; Gnardellis et al, 1995). Nutrient intakes were calculated through a food composition database adapted to accommodate the characteristics of the Greek diet (Trichopoulou, 1992). Intakes were expressed as averages per day, taking into account seasonal variation (Katsouyianni et al, 1997). For this paper, only energy-generating nutrients (protein, carbohydrates, saturated, polyunsaturated and monounsaturated dietary lipids and ethanol) were considered. Occupational and leisure-time activities were assessed by the use of a special section of the lifestyle questionnaire, which was also used for the overall evaluation of physical activity level. Details have been given elsewhere (Trichopoulou et al, 2000). Briefly, the average time per day spent on European Journal of Clinical Nutrition job, household, sporting and other activities, as well as on sleep, was calculated. Each activity was assigned a MET value (the ratio of the metabolic rate associated with a given activity to the resting metabolic rate) recorded in published tables (Ainsworth et al, 1993). The time spent in each of the above activities was multiplied by the MET value of that activity, and all MET-hour products were summed to give a total MET-hour score for the day (Martinez et al, 1997). This score essentially corresponds to the amount of energy per kilogram of body weight expended by an individual during an average day. When energy intake is held constant, increasing energy intake from a particular nutrient presupposes reduction of energy intake from one or more of the remaining energygenerating nutrients. Because this complicates interpretation, we have considered as reliable only findings that were shown to be both robust in the different models and consistent in the various study subgroups (eg those defined by age, smoking status, extent of under-reporting etc). Multiple linear regression was used to evaluate the effects of energy-generating nutrients and ethanol on BMI, whilst adjusting mutually, as well as for possible confounding variables such as age (in specified age-groups as shown in Table 1, categorically), smoking habits (current smokers, exsmokers and never smokers, categorically), educational level expressed as years of schooling (in four groups as shown in Table 2, categorically) and energy expenditure (in 5 METhours per day increments, continuously). The large sample size assures the validity of the normal distribution approximation for statistical testing and estimation. Separate models were fitted for men and women (all non-pregnant), as well as for all participants and those who remained after exclusion of under-reporters (ie subjects with a calculated energy intake less than 1.14 of the basal metabolic rate; Schofield, 1985; Goldberg et al, 1991; Shetty et al, 1996; Gnardellis et al, 1998) and=or subjects on any kind of diet. The parameter coefficients for men and women were compared, when needed, through standard statistical procedures. Several sets of linear regression models were fitted, but the results were generally similar although estimates of the parameters may have occasionally varied. For clarity, we present here only results from the main two sets of these models. In the first set of models, the effects on BMI of energy intake from each energy-generating nutrient is shown, controlling for age, educational level (years of schooling), tobacco smoking, energy expenditure and total energy intake, but not for the other energy-generating nutrients. In each of these models, energy intake from a certain macronutrient was evaluated using standard methods as recommended by Willett and Stampfer (1986) and Willett (Willett, 1998b), but no attempt was made to adjust for energy intake from the remaining energy-generating nutrients. In the second set of models, energy intakes from all the energy-generating nutrients were simultaneously included, adjusting also for age, educational level (years of schooling), tobacco smoking and energy expenditure. In the later Macronutrient intake and BMI A Trichopoulou et al Table 1 39 Distribution of study participants, body mass index (BMI, in kg=m2) and tobacco smoking status, by age and gender Men Women BMI BMI Tobacco smoking (%) Tobacco smoking (%) Cut-offs of quartiles Age groups Cut-offs of quartiles n 1st 2nd 3rd Never smokers Past smokers Current smokers n 1st 2nd 3rd Never smokers Past smokers Current smokers 2072 4252 2398 2761 11483 25.2 25.7 26.0 25.6 25.6 27.3 27.8 28.4 28.1 27.9 29.7 30.3 31.0 30.7 30.4 24.3 19.9 27.9 29.8 24.8 18.6 29.9 37.6 46.9 33.6 57.1 50.1 34.5 23.3 41.7 2760 5669 4099 3851 16379 22.6 24.7 26.8 27.4 25.1 25.0 27.7 29.8 30.3 28.5 28.3 31.4 33.1 33.5 32.1 43.1 62.9 87.5 93.6 72.9 11.3 10.9 4.9 3.6 7.7 45.6 26.2 7.7 2.9 19.3 II. Excluding under-reporters and=or those on a diet < 40 1302 24.9 27.0 29.4 40 – 54 2349 25.5 27.7 30.0 55 – 64 1347 25.9 28.3 30.8 65 þ 1428 25.4 27.9 30.5 Total 6426 25.4 27.7 30.1 23.0 19.6 28.1 32.4 24.9 17.5 25.8 35.2 42.3 29.7 59.5 54.7 36.7 25.3 45.3 1615 3057 1897 1588 8157 22.2 24.6 26.8 27.2 24.6 24.2 27.5 29.7 30.0 27.9 27.0 30.8 32.9 33.0 31.4 44.7 65.3 88.3 94.3 72.2 10.3 9.3 4.1 3.4 7.2 45.0 25.4 7.6 2.3 20.6 I. All participants < 40 40 – 54 55 – 64 65 þ Total Table 2 Percentage distribution of study participants by age, gender and educational level (years of schooling) Men Educational level (years of schooling) (%)a Women Educational level (years of schooling) (%)b Age group 5y 6 – 11 y 12 y 13 y 5y 6 – 11 y 12 y 13 y I. All participants < 40 40 – 54 55 – 64 65 þ Total 0.2 1.5 21.1 36.3 13.7 16.7 36.9 54.1 56.1 41.4 23.2 12.9 7.5 3.4 11.3 59.9 48.7 17.2 3.9 33.4 0.4 4.0 44.6 56.2 25.8 32.5 48.9 36.6 38.3 40.6 28.4 19.7 10.9 3.6 15.2 38.6 27.3 7.4 1.5 18.2 those on a diet 18.7 23.4 42.3 12.1 56.7 7.1 55.7 2.9 43.5 11.3 57.7 44.2 11.7 2.5 30.9 0.6 3.9 47.6 57.9 23.9 34.9 53.4 37.5 37.2 42.9 26.9 18.2 8.7 2.7 14.7 37.6 24.4 5.9 1.5 18.3 II. Excluding under-reporters and=or < 40 0.2 40 – 54 1.4 55 – 64 24.4 65 þ 38.6 Total 14.3 a For 20 (0.2%) and 6 (0.1%) of the male participants in the upper and the lower panel, respectively, the educational level was unknown. For 42 (0.3%) and 20 (0.3%) of the female participants in the upper and the lower panel, respectively, the educational level was unknown. b models, total energy intake was not included to avoid collinearity — the so-called energy partition model (Willett, 1998b). All the analyses were performed with the STATA statistical package (Stata Corporation, 1999). In the Results section, all partial regression coefficients express changes in BMI for an increment of energy intake from the respective nutrient of 418.4 kJ or 100 kcal. Results Tables 1 and 2 show the distribution of study participants, BMI, tobacco smoking status and educational status (years of schooling), by age and gender, for all participants, and after exclusion of energy intake under-reporters and=or those on a diet of any type. The high prevalence of smoking among men and among younger women is well known for the Greek population (Table 1). There are, by design, more women than men in the study and about one half in both genders claims to be on a diet of some type and=or are considered as under-reporters of energy intake (Table 1). The level of education is, as expected, remarkably high in younger compared to older Greek people. Thus, among men, 59.9% of those younger than 40 y have more than 12 y of schooling, whereas, among those 65 y or older, the corresponding figure is 3.9% (Table 2). The increase of BMI until about the age of 65 (eg from a median of 27.3 and 25.0 kg=m2 among men and women, respectively, younger than 40 y, to a median of 28.4 and 29.8 kg=m2 among men and women, European Journal of Clinical Nutrition Macronutrient intake and BMI A Trichopoulou et al 40 respectively, 55 to 64-y-old) is in line with expectations. So is the association of under-reporting with increased BMI (Pvalue from t-test for unequal variances less than 0.001). Table 3 presents quartiles of total energy intake (in kJ) and percentage energy intake from specific nutrients by gender, for all participants, and after exclusion of energy intake under-reporters and=or those on a diet of any type. Again, the data are compatible to expectations, because Greeks are known to consume more than 40% of total energy intake from lipids, particularly in the form of olive oil (Trichopoulou et al, 1993). Note that the values in Table 3 denote the quartiles of energy intake from the different nutrients. Table 3 Cut-off points of quartiles of total energy intake (in kJ) and of percentage of energy intake from specific nutrients by gendera Men Quartiles Women Quartiles 1st 2nd 3rd 1st 2nd 3rd 7821.8 13.1 34.5 41.5 10.9 4.9 19.1 0.9 9597.3 14.1 38.3 45.0 12.6 5.5 21.8 3.2 11713.8 15.2 42.1 48.4 14.2 6.9 24.3 7.0 6280.6 13.3 36.5 43.8 11.3 5.1 20.1 0.0 7655.6 14.3 39.9 46.9 13.0 5.7 23.0 0.4 9266.2 15.4 43.4 49.9 14.7 7.5 25.3 1.3 II. Excluding under-reporters and=or those on a diet Energy 9339.6 10775.1 Protein (%) 13.1 14.1 Carbohydrates (%) 34.4 38.1 Lipids (%) 41.6 44.9 Saturated 11.4 12.8 Polyunsaturated 4.9 5.5 Monounsaturated 18.7 21.5 Ethanol (%) 1.3 3.7 12584.0 15.0 41.6 48.0 14.3 7.2 23.8 7.6 7483.7 13.3 36.5 44.5 11.9 5.2 20.0 0.0 8553.9 14.2 39.6 47.3 13.4 5.7 22.8 0.4 9966.6 15.1 42.8 49.9 15.0 8.1 25.0 1.6 I. All participants Energy Protein (%) Carbohydrates (%) Lipids (%) Saturated Polyunsaturated Monounsaturated Ethanol (%) a Note that the values denote the quartiles of energy intake from the different nutrients. Therefore, the figures in the columns do not add up to 100% and the sums increase, as they should, from quartile 1 to quartile 3. Table 4 Partial regression coefficients (b) (and 95% confidence intervals) expressing changes of body mass index (BMI) for an increment of 418.4 kJ or 100 kcal of energy intake from the specified nutrient, controlling for age, educational level (years of schooling), tobacco smoking, energy expenditure and energy intake (left side), or, controlling for age, educational level (years of schooling), tobacco smoking, energy expenditure and mutually among the nutrients (right side) Without mutual adjustment among nutrients Men I. All Protein Carbohydrates Saturated Polyunsaturated Monounsaturated Ethanol 0.76 7 0.09 0.29 0.18 0.06 7 0.07 II. Excluding under-reporters and=or Protein 0.83 Carbohydrates 7 0.06 Saturated 0.32 Polyunsaturated 0.22 Monounsaturated 0.001 Ethanol 7 0.07 a 0.55, 7 0.15, 0.15, 0.05, 7 0.02, 7 0.11, b Men 95% CI Women b 95% CI b 95% CI 0.98 7 0.03 0.42 0.31 0.15 7 0.03 1.35 7 0.27 0.12 0.56 0.08 7 0.83 1.08, 7 0.35, 7 0.05, 0.41, 7 0.03, 7 0.99, 1.62 7 0.18 0.30 0.71 0.18 7 0.67 0.80 7 0.09 7 0.18 0.13 0.03 7 0.03 a r ¼ 0.17 0.55, 1.06 7 0.14, 7 0.04 7 0.37, 0.01 0.004, 0.26 7 0.05, 0.11 7 0.08, 0.01 P < 0.0001 1.59 7 0.26 7 0.81 0.56 0.12 7 0.80 r ¼ 0.38 1.30, 1.88 7 0.32, 7 0.20 7 1.02, 7 0.61 0.41, 0.70 0.03, 0.21 7 0.95, 7 0.64 P < 0.0001 those on a diet 0.56, 1.10 7 0.13, 0.01 0.14, 0.50 0.06, 0.37 7 0.11, 0.11 7 0.12, 7 0.02 1.13 7 0.10 7 0.15 0.56 7 0.13 7 0.62 0.81, 7 0.19, 7 0.35, 0.40, 7 0.25, 7 0.79, 1.44 0.001 0.05 0.71 7 0.01 7 0.46 0.91 7 0.01 7 0.06 0.21 0.03 0.02 r ¼ 0.22 0.58, 1.23 7 0.08, 0.05 7 0.30, 0.18 0.05, 0.37 7 0.07, 0.14 7 0.03, 0.06 P < 0.0001 1.77 7 0.07 7 0.88 0.64 0.12 7 0.54 r ¼ 0.48 1.43, 2.11 7 0.14, 7 0.001 7 1.11, 7 0.65 0.48, 0.79 0.01, 0.22 7 0.70, 7 0.37 P < 0.0001 r is the multiple correlation coefficient. European Journal of Clinical Nutrition Women 95% CI b With mutual adjustment among nutrients Macronutrient intake and BMI A Trichopoulou et al 41 Therefore, the figures in the columns of Table 3 do not add up to 100% and the sums increase, as they should, from quartile 1 to quartile 3. Table 4 presents the partial regression coefficients expressing changes of BMI per 418.4 kJ (100 kcal) increments of energy intake from a particular nutrient, adjusting for age, educational level (years of schooling), tobacco smoking, energy expenditure and total energy intake (left side), or for age, educational level (years of schooling), tobacco smoking, energy expenditure and energy intake from the remaining energy-generating nutrients simultaneously (right side). The coefficients are presented when all participants have been considered (upper panel), as well as, after exclusion of under-reporters and=or those on a diet of any type (lower panel). The striking finding in this table is the significant positive association between protein intake and BMI. The association is evident when nutrients are not mutually adjusted for (ie left upper side of Table 4) and tends to increase after mutual adjustment for nutrients (right upper side of Table 4), as well as after the exclusion of underreporters and=or those on a diet of any type (lower panel of Table 4). The positive association of protein with BMI is also substantial, because the respective partial regression coefficients are considerably larger, in absolute terms, than those of all other nutrients. The lack of independence hinders formal statistical testing, but it is noteworthy that the confidence interval of the partial regression coefficient for protein does not overlap with those of the partial regression coefficients for the other macronutrients. There is also evidence that the protein partial regression coefficient is larger among women than among men (P 0.001 for the significance of interaction term between gender and protein intake in all models involving mutual adjustment among nutrients). With respect to carbohydrates, there is evidence of an inverse association that persists after mutual adjustment among nutrients. However, the association is not significant when under-reporters and=or those on a diet are excluded and, in all instances, the regression coefficients are, in absolute terms, 10 times smaller than those concerning protein. With respect to saturated lipids, the evidence is conflicting and, thus, unconvincing. There is evidence for a significant positive association when nutrients are not mutually adjusted for among men, but the association becomes inverse, and significantly so among women, after mutual adjustment among nutrients (P < 0.001). With respect to polyunsaturated lipid intake, there is evidence of a positive, statistically significant, but weak association with BMI, the evidence being generally weaker among men, compared to women. Regarding the intake of monounsaturated fat its effect, although generally not significant, appears to be positive and statistically significant among women, after adjustment for other macronutrients (right side, upper and lower panel of Table 4). When lipids are considered together, as total lipid intake, the respective regression coefficient is, as expected, a weighted average of the three constituent groups, but conveys no additional information. Finally, the evidence for ethanol intake appears compatible with an inverse association with BMI among women, whereas the association among men is weaker especially in regression models where all energy-generating nutrients are mutually adjusted for (right side of Table 4). Discussion The present study is based on data that are both observational and cross-sectional in nature, and thus has the drawbacks inherent in these approaches (Willett, 1998b; Trichopoulou et al, 2000). However, efficient analyses of cross-sectional data represent a valuable initial step in identifying relationships between diet and health-related outcomes. Moreover, intervention and prospective cohort studies have their own weaknesses. Thus, they can rarely be large enough to allow the documentation of other than very strong nutritional associations. Furthermore, they depend on long-term dietary compliance (intervention studies) or frequently repeated dietary measurements (observational cohort studies). Observational cross-sectional investigations can thus be useful, particularly when they are large and rely on strict protocols and standardised instruments and procedures. This study is one of the largest in this field and has relied on questionnaires that have been detailed, standardised, validated and administered by specially trained interviewers (Gnardellis et al, 1995; Katsouyianni et al, 1997; Pols, et al, 1997). The analysis of data from observational nutritional studies is not straightforward and several issues remain controversial. Prominent among them is the inability to specifically evaluate the role of changes in the intake of any particular nutrient alone, while controlling for total energy intake. It has been pointed out in the literature (Wacholder et al, 1994; Willett, 1998b) that assessment of differential effects of energy-generating nutrients on BMI (or other outcomes) through standard regression procedures is not straightforward. This is because keeping total energy intake constant and evaluating the consequences on BMI of increasing energy intake from a particular nutrient presupposes reduction of energy intake from one or more of the remaining energy-generating nutrients. Additional complications arise from the varying extent of misclassification in the reported food consumption and the estimated nutrient intake. Sources of misclassification may be gender or age-related or may be unidentifiable; moreover they may have different consequences depending on whether a particular nutrient is the focus of the analysis or serves as a confounder to be controlled for (Willett, 1998b). Given these problems and uncertainties surrounding nutritional observational data and their analysis on the one hand, and the difficulty of undertaking large long-term nutritional interventions on the other, it seems prudent to rely for aetiological inferences only on empirical associations European Journal of Clinical Nutrition Macronutrient intake and BMI A Trichopoulou et al 42 that are strong and consistent. Thus we have investigated the effect of protein, carbohydrates, various forms of dietary lipids and ethanol on BMI by means of different linear regression models, and we have considered as reliable only findings that were shown to be both (i) robust in the different models, (ii) consistent when under-reporters (Gnardellis et al, 1998) and=or participants on a diet of any type were excluded from the analysis, and (iii) consistent in subgroups defined by age, smoking status, etc (data not shown). Under these criteria and contrary to expectations, the strongest and most robust association in these data was the positive one between protein intake and BMI, after adjusting for demographic variables, tobacco smoking, total energy intake and physical activity. Given the very large sample size, the slight positive skewness of the distribution of BMI does not affect the validity of results relying on traditional analysis of variance and multiple regression and, in any case, log transformation of BMI generated essentially identical results. There were other findings of interest in the present investigation, although these were less consistent (Table 4). Thus, the positive association between protein intake and BMI is substantially and significantly stronger among women. Polyunsaturated lipids appear to be positively associated with BMI in both genders, but the association, although statistically significant, is generally much weaker than that concerning protein. Lastly, and again mostly among women, there is evidence that, for a given total energy intake, ethanol intake is inversely associated with BMI. Nevertheless, we are reluctant to embark on biological speculation on the basis of these results. The large size of the study can make results of minor magnitude and questionable importance statistically significant, and the apparent gender interaction that seems to magnify most findings among women could be attributed to more accurate reporting by women. The alternative, that in some way female metabolism is more discriminatory in its response to energy intake from different nutrients, does not have strong scientific foundations. In a parsimonious way, we interpret our findings as suggestive that protein intake is likely to increase BMI in both genders, whereas intake of carbohydrates, saturated lipids, monounsaturated lipids and ethanol are rather unlikely to affect BMI, when total energy intake and expenditure are kept constant. We cannot, however, reject the hypothesis that polyunsaturated lipid intake may increase BMI. The possibility that energy-generating nutrients may affect obesity in humans in ways over and beyond those accounted for by their energy content has been intensively investigated and several major reviews have evaluated the collective evidence. Most authors are inclined to believe that dietary lipid intake promotes obesity in humans (Astrup et al, 1997; Bray and Popkin, 1998; Hill et al, 2000) as it appears to do in laboratory animals (West & York, 1998). There are, however, other investigators who consider the evidence inconclusive, because of the inherent limitations in designing and analysing the studies addressing this complex issue European Journal of Clinical Nutrition (Willett, 1998a; Seidell, 1998a). The present investigation does not provide evidence that saturated or monounsaturated fat tend to increase obesity, when energy intake is accounted for. Moreover, the results of the present study are compatible with carbohydrates having no positive association with obesity, in line with the results reported by other investigators (Nelson and Tucker, 1996; Stubbs et al, 1997). Lastly, ethanol intake was not positively associated with obesity after controlling for energy intake; in fact, an inverse association was evident, particularly among women, possibly explained in terms of socio-cultural norms rather than metabolic mechanisms. In contrast, we found evidence that protein intake is strongly positively associated with obesity after controlling for energy intake in different ways. The finding is not entirely original (Slattery et al, 1992; Rolland-Cachera et al, 1995) and in fact, it has been hypothesised that protein intake may be the main underlying reason for the increasing prevalence of obesity in the Western world (McCarty, 2000). Nevertheless, the present study provides the strongest empirical evidence yet that protein intake may be a crucial determinant of obesity in humans. In conclusion, we have found evidence indicating that protein intake is conducive to obesity. No other association, positive or inverse, in these data was sufficiently strong and consistent to allow reliable biological speculation. Moreover, these data indicate that neither saturated or monounsaturated fat nor carbohydrates are likely to play a major role in increasing BMI over and beyond their energy content. Since both proteins and polyunsaturated lipids represent a relatively small fraction of total energy intake, it would appear that qualitative alterations of diet may have little effect on BMI at the individual level. On the other hand, at a population level, the relatively high protein and polyunsaturated lipid intake of Americans in the United States could explain, in part, the expanding obesity problem in this country and perhaps other countries as well. Acknowledgements The European Prospective Investigation into Cancer and nutrition (EPIC) is co-ordinated by the International Agency for Research on Cancer (IARC) and supported by the Europe Against Cancer Program of the European Commission. The Greek segment of the EPIC study is also supported by the Greek Ministry of Health. References Ainsworth BE, Haskell WL & Leon AS (1993): Compendium of physical activities: classification of energy costs of human physical activities. Med. Sci. Sports Exerc. 25, 71 – 80. Astrup A, Toubro S, Raben A & Skov AR (1997): The role of low-fat diets and fat substitutes in body weight management: what have we learned from clinical studies? J. Am. Dietet. Assoc. 97(Suppl 7), S82 – S87. Bray GA (1996): Obesity. In Present Knowledge in Nutrition, ed. EE Ziegler & LJ Filer, pp 19 – 32. Washington, DC: ILSI Press. Macronutrient intake and BMI A Trichopoulou et al Bray GA & Popkin BM (1998): Dietary fat intake does affect obesity! Am. J. Clin. Nutr. 68, 1157 – 1173. De Groot LC & van Staveren WA (1995): Reduced physical activity and its association with obesity. Nutr. Rev. 53, 11 – 13. Garrow JS (1986): Obesity. In Human Nutrition and Dietetics, ed. LSP Davidson, R Passmore & MA Eastwood, pp 465 – 479. Edinburgh: Churchill Livingstone. Gnardellis C, Trichopoulou A, Katsouyianni K, Polychronopoulos E, Rimm EB & Trichopoulos D (1995): Reproducibility and validity of an extensive semi-quantitative Food Frequency Questionnaire among Greek school teachers. Epidemiology 6, 74 – 77. Gnardellis C, Boulou C & Trichopoulou A (1998): Magnitude determinants and impact of under-reporting of energy intake in a cohort study in Greece. Public Hlth. Nutr. 1, 131 – 137. Goldberg GR, Black AE, Jebb SA, Cole TJ, Murgatroyd PR, Coward WA & Prentice AM (1991): Critical evaluation of energy intake data using fundamental principles of energy physiology: 1. Derivation of cut-off values to identify under-recording. Eur. J Clin. Nutr. 45, 569 – 581. Hill JO, Melanson EL & Wyatt HT (2000): Dietary fat intake and regulation of energy balance: implications for obesity. J. Nutr. 130(Suppl 2S), 284S – 288S. Katsouyianni K, Rimm EB, Gnardellis C, Trichopoulos D, Polychronopoulos E & Trichopoulou A (1997): Reproducibility and relative validity of an extensive semi-quantitative Food Frequency Questionnaire using dietary records and chemical markers among Greek school teachers. Int. J. Epidemiol. 26(Suppl 1), 118 – 127. Klesges RC, Klesges LM, Haddock CK & Eck LH (1992): A longitudinal analysis of the impacts of dietary intake and physical activity on weight change in adults. Am. J. Clin. Nutr. 55, 812 – 822. Martinez ME, Giovannucci E, Spiegelman D, Hunter DJ, Willett WC & Colditz GA (1997): Leisure-time physical activity, body size, and colon cancer in women. Nurses’ Health Study Research Group. J. Natl. Cancer Inst. 89, 948 – 955. McCarty MF (2000): The origins of Western obesity: a role for animal protein? Med. Hypoth. 54, 488 – 494. Nelson LH & Tucker LA (1996): Diet composition related to body fat in a multivariate study of 203 men. J. Am. Diet. Assoc. 96, 771 – 777. Parker DR, Gonzalez S & Derby CA (1997): Dietary factors in relation to weight change among men and women from two southeastern New England communities. Int. J. Obes. Relat. Metab. Disord. 21, 103 – 109. Pols MA, Peeters PHM, Cocke M, Slimani N, Bueno-de-Mesquita HB & Collette HJA (1997): Estimation of reproducibility and relative validity of questions included in the EPIC physical activity questionnaire. Int. J. Epidemiol. 26, Sl8l – Sl89. Riboli F (1992): Nutrition and Cancer: Background and rationale of the European Prospective Investigation into Cancer and Nutrition (EPIC). Ann. Oncol. 3, 783 – 791. Rising R, Harper IT, Fontvielle AM, Ferraro RT, Spraul M & Ravussin F (1994): Determinants of total daily energy expenditure: variability in physical activity. Am. J. Clin. Nutr. 59, 800 – 804. Rolland-Cachera MF, Deheeger M, Akrout M & Bellisle F (1995): Influence of macronutrients on adiposity development: a follow up study of nutrition and growth from 10 months to 8 y of age. Int. J. Obes. Relat. Metab. Disord. 19, 573 – 578. Schofield WN (1985): Predicting basal metabolic rate, new standards and review of previous work. Hum. Nutr. Clin. Nutr. 39C(Suppl 1), 5 – 41. Schulz LO & Schoeller DA (1994): A compilation of total daily energy expenditure and body weights in healthy adults. Am. J. Clin. Nutr. 60, 676 – 681. Seidell JC (1998a): Dietary fat and obesity: an epidemiologic perspective. Am. J. Clin. Nutr. 67(Suppl 3), 546S – 550S. Seidell JC (1998b): Societal and personal costs of obesity. Exp. Clin. Endocrinol. Diabetes 106(Suppl 2), 7 – 9. Shetty PS. Henry CJK, Black AE & Prentice AM (1996): Energy requirements of adults: an update on basal metabolic rates (BMRs) and physical activity levels (PALs). Eur. J. Clin. Nutr. 50(Suppl 1), S11 – S23. Slattery ML, McDonald A, Bild DE, Caan BJ, Hilner JE, Jacobs DR Jr & Liu K (1992): Associations of body fat and its distribution with dietary intake, physical activity, ethanol, and smoking in blacks and whites. Am. J. Clin. Nutr. 55, 943 – 949. Stata Corporation (l999): Intercooled Stata 6.0 for Windows 98=95=NT. College Station, TX: Stata Corporation. Stubbs RJ, Prentice AM & James WP (1997): Carbohydrates and energy balance. Ann. N.Y. Acad. Sci. 819, 44 – 69. Trichopoulou A (1992): Composition of Greek Foods and Dishes (in Greek and English). Athens: Athens School of Public Health. Trichopoulou A, Toupadaki N, Tzonou A, Katsouyanni K, Manousos O, Kada E & Trichopoulos D (1993): The macronutrient composition of the Greek diet: estimates derived from six case-control studies. Eur. J. Clin. Nutr. 47, 549 – 558. Trichopoulou A, Gnardellis CH, Lagiou A, Benetou V & Trichopoulos D (2000): Body mass idex in relation to energy intake and expenditure among adults in Greece. Epidemiology 11, 333 – 336. Wacholder S, Schatzkin A, Freedman LS, Kipnis V, Hartman A & Brown CC (1994): Can energy adjustment separate the effects of energy from those of specific macronutrients? Am. J. Epidemiol. 140, 848 – 855. West DB & York B (1998): Dietary fat, genetic predisposition, and obesity: lessons from animal models. Am. J. Clin. Nutr. 67, 505S – 512S. Willett W & Stampfer MJ (1986): Total energy intake: implications for epidemiologic analyses. Am. J. Epidemiol. 124, 17 – 27. Willett WC (1998a): Is dietary fat a major determinant of body fat? Am. J. Clin. Nutr. 67,(Suppl 3), 556S – 562S. Willett WC (1998b): Nutritional Epidemiology. New York: Oxford University Press. World Cancer Research Fund & American Institute for Cancer Research (1997): Energy and related factors. In Food Nutrition and the Prevention of Cancer: a Global Perspective, pp 366 – 375. World Cancer Research Fund and American Institute for Cancer Research. 43 European Journal of Clinical Nutrition