Curr Obes Rep DOI 10.1007/s13679-013-0089-0
ETIOLOGY OF OBESITY (MS WESTERTERP-PLANTENGA, SECTION EDITOR)
The Association Between Diet and Obesity in Specific European Cohorts: DiOGenes and EPIC-PANACEA Edith J. M. Feskens & Diewertje Sluik & Huaidong Du
# Springer Science+Business Media New York 2013
Abstract This review summarizes evidence from two projects embedded within the European Prospective Investigation into Cancer and Nutrition (EPIC) on the association between dietary factors and obesity risk, in particular change in weight and waist circumference. A total of 12 publications from DiOGenes and six from EPIC-PANACEA were reviewed. The results show that dietary fiber, especially cereal fiber, was inversely associated with weight or waist change, as well as fruit/ vegetable intake and the Mediterranean dietary pattern. Energy density and meat consumption were positively associated with the anthropometric changes, as was glycemic index with waist change. Clear associations with macronutrient composition were not observed. In additional studies, interactions with genetic polymorphism were investigated and shown to be present for protein intake and GI, although effect estimates were small. These interactions require replication. These results show that in European populations dietary factors are independently associated with weight/waist change. The findings provide further clues for the prevention of obesity. Keywords Cohort . Diet . Dietary patterns . Epidemiology . Macronutrients . Waistcircumference . Weightgain . Obesity . DiOGenes . EPIC-PANACEA
Introduction Epidemiological studies on diet and chronic disease have contributed considerably to the understanding of the role of E. J. M. Feskens (*) : D. Sluik Division of Human Nutrition, Wageningen University, PO Box 8129, 6700 EV Wageningen, The Netherlands e-mail:
[email protected] H. Du Clinical Trial Service Unit and Epidemiological Studies Unit (CTSU), Nuffield Department of Population Health, University of Oxford, Oxford OX3 7LF, UK
diet on the risk of cancer and cardiovascular diseases [1]. With the global increase of obesity and overweight, insight in the role of dietary factors as modifiable risk factors is important. Large cohort studies with high-quality dietary data are preferred to investigate associations with weight gain over time. The strength of a cohort study is that the exposure of interest has been assessed before the occurrence of the outcome. However, a major concern is the role of confounding. Confounding can be described as the effect of an extraneous factor that is mistaken for or mixed with the actual exposure effect, although the goal of most observational research is to estimate the actual exposure effect [2]. Regarding the risk of obesity, this means that cohort studies also need to measure main risk factors for obesity. Diet, physical activity and smoking are potential confounders: they are risk factors for obesity and potentially associated with dietary intake as well. Furthermore, independent of the dietary assessment method that is being employed, it is very likely that selective overor underreporting depending on BMI will take place, which is not a random, but a systematic error in dietary assessment. Evidence on this phenomenon has been obtained from studies using urinary nitrogen as marker of total protein intake [3, 4] and more recently by using the doubly labeled water technique [5]. Moreover, underreporting seems to be not only confined to total caloric intake, but is higher for foods containing fat or carbohydrate [6]. This hampers epidemiological studies on obesity and overweight, especially the cross-sectional studies, where diet and overweight are assessed simultaneously. In order to overcome these problems of confounding and misreporting large longitudinal studies are warranted. In these studies participants with baseline obesity can be excluded and the focus will then be on natural weight change in subjects with normal body weight at the start. Within Europe, two sub-studies of the European Prospective Investigation into Cancer and Nutrition (EPIC) have investigated the dietary determinants of obesity, addressing these methodological
Curr Obes Rep
problems of confounding and misreporting in their design and analysis: “Diet, Obesity and Genes” (DiOGenes) and the European Prospective Investigation into Cancer and Nutrition – Physical Activity, Nutrition, Alcohol Consumption, Cessation of Smoking, Eating Out of Home, and Obesity (EPIC-PANACEA). This review summarizes the published evidence from these cohorts on the association between dietary factors and obesity risk, in particular change in weight and waist circumference. The publications from DiOGenes and EPIC-PANACEA have addressed the associations of the following dietary factors and change in weight and waist: macronutrients, dietary fiber, energy density (ED), glycemic index (GI) and glycemic load (GL), as well as food groups and food patterns. In addition, attention will be paid to diet-gene interactions and weight change.
Methods Study Design and Population DiOGenes and EPIC-PANACEA DiOGenes and EPIC-PANACEA are sub-cohorts of the EPIC study. EPIC is a renown pan-European multicenter prospective cohort study including 519,978 participants, aged 35–70 years, from 23 study centers in ten countries, who were recruited from 1992 to 2000. Participants were predominantly recruited from the general population residing in a given geographic area (town, province, or country) [7]. DiOGenes is a multi-disciplinary European project targeting dietary components, genetic, and behavioral factors in the prevention of weight gain, funded by the European Union. Data from EPIC-participants from the following centers were included because they had a follow-up program including reassessment of anthropometry: Florence (Italy), Norfolk (United Kingdom; UK), Amsterdam, Maastricht and Doetinchem (The Netherlands), Potsdam (Germany), and Copenhagen and Aarhus (Denmark). Of the 146,543 men and women who took part in the baseline examination during 1992–1998, 102,346 (69.8 %) participated in the follow-up examination from 1998 to 2005. After applying the following exclusion criteria: missing information on diet and anthropometrics, unrealistic or implausible anthropometric measurements and energy intake, and presence of chronic diseases, the DiOGenes cohort included 89,432 participants (58 % women) [8]. In a later stage, the other EPIC-centers from France, Spain, Italy (Ragusa, Naples, Turin, Varese), UK (Oxford), The Netherlands (Utrecht), Greece, Germany (Heidelberg), Sweden, and Norway also provided follow-up data, based on self-report rather than measurements. This is known as EPIC-PANACEA, which aimed to study determinants of
body weight and subsequent weight change using a large sample of EPIC-participants with different dietary habits and lifestyles and a wide range of obesity prevalence. After excluding participants without dietary data, baseline anthropometric data or with extreme or implausible values of anthropometry and energy intake, EPIC-PANACEA included data of 373,803 (103,455 male and 270,348 female) EPIC-participants [9••]. Dietary Assessment Habitual diet over the past year was measured at baseline using validated, country-specific food frequency questionnaires (FFQs). In Italy, the Netherlands, Germany and Spain extensive quantitative dietary questionnaires with 300–500 food items were used. In Naples, semi-quantitative questionnaires were used and combined dietary methods of food records and questionnaires in Malmö [7]. To account for differences between national FFQs and to reduce potential measurement error, diet was also assessed with a highly standardized reference dietary measurement, a computerized 24-hour dietary recall in an age-stratified random sample of ca. 8 % of the whole cohort [10]. Anthropometric Measurements Baseline body weight (kg) and height (cm) were measured according to standardized procedures without shoes and light indoor clothing (Norfolk, Doetinchem, Amsterdam, and Maastricht) or underwear (Florence, Potsdam, Copenhagen, and Aarhus). In Oxford (UK), Norway, and France, anthropometric measurements at baseline were self-reported. Baseline waist circumference (cm) was measured either at the narrowest torso circumference or at the midpoint between the lower ribs and iliac crest [11]. At follow-up, weight and waist circumference were selfreported by the participants in follow-up questionnaires in all centers except in Norfolk (UK) and Doetinchem (Netherlands) where it was measured according to the baseline protocol. Covariate Assessment Information on lifestyle was collected through self-administered questionnaires. Questions covered age, gender, physical activity, education level, smoking, menopausal status, and use of hormone replacement therapy. Information on health status (cancer, CVD, diabetes) was collected using either questionnaires or disease registries. Physical activity level was indexed into five categories (inactive, moderately inactive, moderately active, active, or unknown) based on occupational and recreational activities. Information on smoking status (never, former, or current smoker) was collected via selfadministered questionnaires at baseline and at follow-up.
Curr Obes Rep
Results Table 1 gives an overview of the 18 studies included in this review on diet and weight or waist change published up to November 2013. We included 12 publications on the DiOGenes cohort and six publications from EPIC-PANACEA. Macronutrients Protein Protein intake may be the macronutrient least vulnerable to specific over-or underreporting by overweight or obese people. Interesting hypotheses have recently been put forward regarding beneficial effects of protein on weight loss, as protein is considered to increase thermogenesis and satiety more than other macronutrients [12]. However, the analyses of Halkjaer et al. in the DiOGenes cohort show that a higher intake of total and animal protein was associated with subsequent weight gain in both genders. The pooled estimate for a 150-kcal/d higher intake was 52.9 g/y for total protein and 56.22 g/y for animal protein. The association was strongest among women; a 150-kcal/d higher total protein intake was associated with a weight gain of 77.46 g/y (95 %CI: 35.27, 119.65) in women and 28.87 g/y (95 %CI: −0.84, 58.58) in men [13]. Moreover, the association was mainly attributable to protein from red and processed meat and poultry, rather than from fish or dairy. There was no overall association between intake of plant protein and subsequent changes in weight. No clear overall association between intakes of total protein or any of the subgroups, and changes in waist circumference was present. More recently, Vergnaud et al. confirmed this observation in EPIC-PANACEA [14]. A diet with more than 22 % of energy from protein were associated with a 23-24 % higher risk of becoming overweight or obese in normal weight and overweight subjects at baseline, compared to diets with up to 14 energy % protein. Furthermore, a 5 % higher proportion of protein at the expense of fat or carbohydrates was positively associated with weight gain after 5 years. At the expense of carbohydrate, weight gain was 247 g (95 %CI: 160, 334) in men and 338 g (95 %CI: 296, 480) in women; at the expense of fat, weight gain was 275 g (95 %CI: 184, 366) in men and 397 g (95 %CI: 303, 491). These associations were observed for both plant and animal protein. At first sight, these results of observational studies are in contrast to the findings from intervention studies. The DiOGenes multicenter randomized dietary intervention study randomized 773 participants to one of five weightmaintenance diet of 26 weeks low and high in protein and GI after an initial weight loss period. This trial showed that a high- or moderately-high protein diet is better in maintaining weight loss [15].
However, in defining a diet as low- or high in protein the absolute intake of protein is key, not the relative proportion [12]. It appears that the effects of dietary protein on body weight depend on energy balance. In a condition of weightloss energy intake is decreased and subjects are losing bodyweight. Normally, this implies the risk of increased hunger, decreased energy expenditure, together with a decrease in fat free body mass [12]. These phenomena are prevented by keeping the absolute protein intake at the original level of 0.8–1.2 g/kg body weight/d [16], thus increasing energy % of protein [12]. In case of energy balance under controlled conditions, with an increase in protein intake at the cost of carbohydrate and fat intake, fat free body mass increases at the expense of body fat even without additional exercise, thus keeping body weight stable [17]. When subjects are overfed, thus keeping a positive energy balance under controlled conditions, they gain more weight at higher protein intakes compared to the lower protein intake [18]. In this situation weight gain consists mainly of gain of fat free mass. The smallest weight gain was observed in the low protein diet (5 energy-%), which corresponds more or less with a normal absolute protein intake in a weight stable population [18]. The observational longitudinal effects of protein intake related to weight gain in epidemiological studies correspond to these overfeeding studies. It appears that subjects who increased their protein intake from energy balance onward gained more weight than subjects who did not increase their protein intakes. Fatty Acids Low fat diets have long been thought to be optimal for overweight prevention and weight loss [19]. In general, diets with a higher energy percentage of fat result in a higher total energy intake, given that fat is the most energy-dense macronutrient. However, it is not clear whether the fat content of the diet, with or without adjustment for total energy intake, influences weight gain. Within DiOGenes, no significant association between fat intake (amount or type) and weight change was shown after adjustment for anthropometric, dietary and lifestyle factors and follow-up period [20]. Per 1 g/d energy-adjusted fat intake, the difference in mean annual weight change was 0.90 g/y (95 % CI: −0.54, 2.34 g/y) for men and −1.30 g/y (95 % CI: −3.70, 1.11 g/y) for women. When participants with overweight or obesity at baseline were excluded, similar results were obtained. This suggests that selective underreporting of fat intake did not play an important role. Moreover, fat intake was not associated with weight change in EPIC -PANACEA: a 5 % higher proportion of fat at the expense of carbohydrates was not associated in men, or in women [14]. In the Nurses’ Health Study percentage of calories from animal, saturated, and trans- fatty acids were positively
89,432
89,432
89,432
89,432
11,921
89,432
Du, 2009a [8]
Du, 2009b [34]
Du, 2010 [28••]
Du, 2011 [43]
Forouhi, 2009 [20]
N
Buijsse, 2009 [35]
DiOGenes
First author, year
Weight change and waist circumference change
Weight change
Outcome variable
Baseline fat intake (total, saturated, polyunsaturated, and monounsaturated fats)
Weight change
123 single nucleotide Weight change polymorphisms (SNPs) from 15 candidate genes in the hypothalamic pathway
Total fiber, cereal fiber and fruit Weight change and waist and vegetable fiber intake circumference change
Dietary glycemic index (GI) and Weight change and waist glycemic load (GL) circumference change
Dietary energy density (ED): energy intake (kcal) from foods divided by the weight (g) of foods
Fruit and vegetable intake
Exposure variable
• No significant associations were found between the 123 SNPs and weight change or risk of being a ‘weight gainer’ • Carriers of the minor allele SNP rs7180849 in the neuromedin b gene (NMB) had more pronounced weight gain at a higher glycemic index of the diet
• Pooled estimate for a 10-g/d higher total fiber intake was −39 g/y (95 %CI: −71, −7 g/y) for weight change and −0.08 cm/y (95 %CI: −0.11, −0.05 cm/ y) for waist circumference change • Pooled estimate for a 10-g/d higher cereal fiber intake was −77 g/y (95 % CI: −127, −26 g/y) for weight change and −0.10 cm (95 %CI: −0.18, −0.02 cm/y) for waist circumference change • Pooled estimate for a 10-g/d higher fruit and vegetable fiber intake was not significant for weight change and −0.08 cm/y (95 %CI: −0.15, −0.01 cm/y) for waist circumference change
• GI was not significantly associated with weight change. • GI was associated with waist change: pooled estimate for a 10-unit GI difference was 0.26 cm/y (95 % CI: 0.20, 0.33 cm/y) • GL was not significantly associated with weight and waist change.
• ED was not significantly associated with weight change. • For 1 kcal/g ED, annual waist circumference change was 0.09 cm/y (95 %CI: 0.01, 0.18 cm/y). • ED was stronger positively related to weigth and waist circumference change in subjects with baseline BMI 22 en% protein were related to a 23-24 % higher risk of becoming overweight or obese
• Fruit and vegetable intake was not associated with weight change overall, but in those who quitted smoking during follow-up • In women, interactions were observed for age, BMI, smoking status, and prudent dietary pattern scores.
• A 100-kcal/day increase in meat consumption was associated with a higher weight gain: pooled estimate was 25 gram/year (95 %CI: 19, 31 gram/year). • The association was seen in men and women, normal weight and overweight, smokers and nonsmokers, and for red meat, poultry and processed meat
• A 2-point increase of the Mediterranean diet score (0– 18 points) was associated with a 5-year weight change or −0.04 kg (95 %CI: −0.07, −0.02 kg) (pooled estimate) • A 2-point increase in the Mediterranean diet score was also associated with the odds of becoming overweight or obese in 5 years: pooled OR 0.97 (95 %CI: 0.95, 0.99).
• In men, eating at restaurants was nonsignificantly positively associated with weight gain. • No association was seen between eating at restaurants and weight changes, controlling for energy intake
Age, BMI at enrolment, menopausal status, highest • A 10-gram/day higher consumption of fish was educational level achieved, smoking status, physical associated with a weight gain of 5.70 gram/year activity level, total energy intake, indicator variable (95 %CI: 4.35, 7.06) in women. for plausibility of reported energy intake Mixed• These associations in women were significant for lean effect linear regression with random effects on and fatty fish consumption intercept took into account clustering of data within • The odds ratio of becoming overweight in 5 years was the cohorts. significantly higher with a higher consumption of fish in women, but not in men
Adjustments
Curr Obes Rep
Curr Obes Rep
associated with weight change [21]. However, the associations were weak, but stronger in overweight rather than normal weight women. Especially the latter observation may indicate that selective misreporting may have affected the findings, although this cannot be directly established. In conclusion, results from DiOGenes and EPIC-PANACEA show that fat intake is not associated with weight gain when adjusted for energy. Carbohydrates In DiOGenes, energy-adjusted carbohydrate intake was not associated with change in waist circumference for a given BMI (WCBMI) [22]. The association between total carbohydrate intake and subsequent weight change was studied in EPIC-PANACEA [14]. Vergnaud et al. showed that a higher proportion of carbohydrate in the diet, at the expense of protein, was inversely associated with weight gain: weight change was −174 g (95 %CI: −240, −109) in men and −307 g (95 %CI: −359, −254) in women. The associations were stronger for rich in fiber carbohydrates. Concluding, evidence on the role of carbohydrate in obesity warrants further study. Energy Density Foods high in fat and sugar are generally more energy-dense. They are also more palatable and as a result may stimulate over-eating [23]. Experimental studies also suggest that people tend to eat a similar volume of food to feel satiated, and thus, consuming energy-dense foods could cause passive over-eating [24]. Long-term experiments have however indicated that unrestrained eaters compensate a lower energy density by increasing the amount eaten [25]. Some observational cross-sectional studies have shown a positive association between ED and obesity, but there are concerns about reverse causality [26]. In DiOGenes, ED was calculated as the energy intake (kcal) from foods divided by the weight (gram) of foods [8]. Note that 1 gram of fat provides 37.7 kJ, while protein and carbohydrate each provide less, 16.7 kJ per gram. Drinks (including water, tea, coffee, juice, soft drinks, alcoholic drinks, and milk) were not included in the calculation in this study. Mean ED was 7.1 kJ/g and varied across study centers. As expected, a lower ED was associated with a lower intake of sugar and fats and a higher intake of fruits and vegetables. After adjustment for baseline anthropometrics, demographic and lifestyle factors, follow-up duration and energy from beverages, ED was not associated with weight change. In overweight subjects a non-significant inverse association was found. In contrast, ED was significantly associated with change in waist circumference. For 4.2 kJ/g ED, the annual weight change was −42 gram (95 %CI: −112, 28) and annual waist circumference change was 0.09 cm (95 %CI: 0.01, 0.18). The
association for waist circumference was stronger in participants with normal body weight at baseline; per 4.2 kJ/g ED, change in waist was 0.17 cm/year (95 %CI: 0.09, 0.25). In an additional analysis of the DiOGenes data by Romaguera et al., the association between ED and change in waist was shown to be independent of change in BMI [22]. A 4.2 kJ/g greater ED predicted a change in WCBMI of 0.09 cm (95 %CI: 0.02, 0.13) in men and 0.15 cm (95 %CI: 0.09, 0.21) in women. Selective underreporting could explain the absence of an overall association between ED and weight change and the tendency for an inverse one in obese participants. Results for change in waist circumference were less affected. We can only speculate that selective reporting is more dependent on body weight than on waist circumference, possibly because people are more aware of their weight rather than their waist. Dietary Fiber Dietary fibers are defined as the edible parts of plant foods that are resistant to digestion and absorption in the small intestine and are completely or partially fermented in the large intestine [27]. Dietary fiber may contribute to in body-weight control through various physiologic mechanisms, for example increasing satiation and satiety, slowing down digestion, and increasing release of gut hormones such as cholecystokinin (CCK) and glucagon-like peptide 1 (GLP1) [27]. In DiOGenes, Du et al. has shown that a higher intake of dietary fiber was associated with lower gain in body-weight and waist circumference [28••]. Total dietary fiber was inversely associated with subsequent change in weight and waist circumference: for a 10-g/d higher total fiber intake, the pooled estimate was −39 g/y (95 %CI: −71, −71 g/y) for weight change and −0.08 cm/y (95 %CI: −0.11, −0.05 cm/y) for waist circumference change. Cereal fibers were more strongly related to change in weight and waist: a 10-g/d higher fiber intake from cereals was associated with −77 g/y (95 %CI: −127, −26 g/y) weight change and −0.10 cm/y (95 %CI: −0.18, −0.02 cm/y) waist circumference change. Fruit and vegetable fiber was associated with changes in waist circumference, but not with body weight change. Furthermore, a higher fiber intake was inversely associated with a change in WCBMI in women: per 10 gram increase, change in WCBMI was −0.06 cm/y (95 %CI: −0.08, −0.03) [22]. From these cohorts, fiber intake appears a promising candidate to reduce the risk of weight gain. The results also agree with those from the Nurses’ Health Study from the United States. Liu et al. showed that women with a higher increase in fiber intake gained less body weight over 12 years [29]. Glycemic Index and Glycemic Load The concept GI was developed in the early 1980s by David Jenkins. It is a quantitative measure of carbohydrate quality
Curr Obes Rep
based on the blood glucose response after consumption. GL was defined later to capture the entire blood glucose-raising potential of dietary carbohydrates. It is calculated as the product of GI and the total available carbohydrates [30]. It has been suggested that low GI or GL diets can help to prevent body weight gain and stimulate weight loss. The low blood glucose and insulin response after consuming a low GI or GL meal could stimulate satiation and satiety and may thus lead to a decrease in energy intake. A low GI or GL diet could regulate fuel partitioning in the way of reducing fat storage as well as limit the decrease of resting metabolic rate under energy restriction [31, 32]. Within DiOGenes, GI and GL were based on an extensive approach involving experts including Prof. T. Wolever and Prof. J. Brand-Miller to obtain a good GI composition table for this European cohort [33]. Across cohorts, mean dietary GI and GL were 57 and 134, respectively [34]. The associations of GI and GL with subsequent changes of weight and waist circumference were heterogeneous across centers. Pooled estimates showed a non-significant association of GI and GL with change in weight and a positive significant association with change in waist. With every 10-unit increase in GI, waist circumference significantly increased by 0.19 cm/yr (95 %CI: 0.11, 0.27) and with every 50-unit increase in GL, waist circumference increased by 0.07 cm/yr (95 %CI: −0.04, 0.17). In subsequent analyses by Romaguera et al., a 10-unit increase in GI was positively associated with change in WCBMI in both men (0.07 cm/y [95 %CI: 0.03, 0.12]) and women (0.06 cm/y [95 %CI: 0.03, 0.10]). Moreover, a 50-unit increase in GL was associated with a change in WCBMI of 0.09 cm/y (95 %CI: 0.01, 0.17) in women [22]. These findings suggest that a higher dietary GI could accelerate waist circumference gain; or in other words, consuming a low GI diet may protect against the long-term development of abdominal obesity. However, given that only a small effect was observed, further studies are needed to confirm this finding. Foods and Food Patterns The role of specific foods and food patterns in weight gain has been studied in both DiOGenes and EPIC-PANACEA. Buijsse et al. showed within DiOGenes that a 100-gram higher fruit and vegetable intake was associated with a weight change of −14 g/y (95 %CI: −19, −9) [35]. In those who stopped smoking during follow-up, weight change was larger than in the others (−37 g/y; p for interaction