Anda di halaman 1dari 11

bs_bs_banner

O R I G I N A L A RT I C L E

DOI: 10.1111/j.1467-3010.2012.01977.x

Associations between fat, sugar and other macronutrient intakes in the National Diet and Nutrition Survey
G. W. Horgan* and S. Whybrow
*Biomathematics & Statistics Scotland, Aberdeen, UK; Food, Consumer Behaviour and Health Research Centre, Faculty of Health and Medical Sciences, University of Surrey, Guildford, UK

Summary

Observational studies show an inverse relationship between fat and sugar within diets, described as the fat-sugar seesaw. It has been suggested however, that this is a consequence of expressing macronutrient intakes as percentages of energy intake rather than by the dietary choices made by individuals. To test this we examined the associations between macronutrient intakes in the diets of adults (n = 1724) participating in the UK National Diet and Nutrition Survey (NDNS) and within the same adults across different days of the week. Pearsons correlations were calculated between the macronutrient intakes from fat, total sugar, intrinsic sugars, non-milk extrinsic sugars (NMES), non-sugar carbohydrate, protein and alcohol. Energy intakes relative to estimated basal metabolic rate (BMR) were calculated to partially account for differences in energy requirements. Pearsons correlations also examined associations between the same macronutrients in foods from the Nutrient Databank (ND), used to analyse dietary intakes in the NDNS. Correlations between fat and sugar(s) were calculated in ve ways: (i) percentage energy between individuals; (ii) amount (g) between individuals; (iii) amount (g) relative to BMR between individuals; (iv) within individuals over seven days; and (v) between food items in the ND. Negative correlations were obtained between mean daily percentage energy intakes of fat and, total sugars, NMES and intrinsic sugars (all P < 0.001). However, when mean daily macronutrient intakes were expressed in weight (g)/day, these were all positively correlated (all P < 0.001). Mean estimated correlations between macronutrient intakes (g/day) for each individual across the days of the week were also positive, indicating a lack of fat-sugar seesaw effect. Within the ND, the correlation between fat and total sugar (g/100g of food) was weakly positive (P = 0.006). Only when examining the correlation between fat and sugar(s) between individuals in percentage terms (% energy) was the fat-sugar seesaw evident; in all other methods the correlations between fat and sugar(s) were positive. Examination of the effects of using percentage energy values to describe the macronutrient composition demonstrated that the fat-sugar seesaw is only an inevitable mathematical consequence, rather than the result of dietary choice. Comparing diet composition in percentage terms alone can therefore be misleading. It is only when examined as absolute values (weight), as well as percentage contributions to energy, that the interrelationships between macronutrients can be studied fully.

Correspondence: Dr Stephen Whybrow, Researcher, Food, Consumer Behaviour and Health Research Centre, Faculty of Health and Medical Sciences, University of Surrey, Guildford GU2 7TE, UK. E-mail: s.whybrow@surrey.ac.uk; E-mail: g.horgan@abdu.ac.uk

2012 The Authors Journal compilation 2012 British Nutrition Foundation Nutrition Bulletin, 37, 213223

213

214

G. W. Horgan and S. Whybrow

Keywords: fat, fat-sugar seesaw, macronutrient intakes, NDNS, non-milk extrinsic sugar, sugars

Introduction
Dietary intake data suggest that fat and sugar have an inverse relationship. Many observational studies show that, in diets where a large percentage of the reported energy intake is provided by fat, a small percentage of energy in the diet comes from sugar(s) (Baghurst et al. 1994; Parnell et al. 2008). McColl (1988) described this as the fat-sugar seesaw. Furthermore, one of these studies noted that consumers with a high sugar, low fat intake tended to have a lower body mass index than low sugar, high fat consumers (Bolton-Smith & Woodward 1994). This led to the suggestion that sugar energy displaces fat energy from the diet and that, as dietary fat is more energy dense and potentially more likely to encourage weight gain, high sugar intakes could protect against obesity (Gibney et al. 1995). Dietary fat has an energy density of 37 kJ/100 g, which is more than twice that of sugar (16 kJ/100 g) (Holland et al. 1991). Therefore, any increase or decrease in the amount (weight) of sugar in the diet will have a smaller effect on the percentage energy from macronutrients (and total energy) than a corresponding increase or decrease in the amount of fat. The concept of the fat-sugar seesaw has been used to suggest that the healthy eating targets of reducing dietary sugar and fat cannot be achieved simultaneously (Drummond et al. 1996). This led to the conclusion that dietary advice would be more effective if focused on fat reduction, while setting no limits on the consumption of carbohydrates, including sugar (Drummond et al. 1996; Gibson 1997). Because sugar is combined with fats in the manufacture of many processed foods, an alternative argument is that high sugar foods can act as a vehicle for dietary fat (Emmett & Heaton 1995; Macdiarmid et al. 1998). Furthermore, Macdiarmid et al. (1995) cast doubt on the fat-sugar seesaw concept, observing that it depends on whether macronutrients are calculated and analysed as percentages of energy intake or weight in grams. Central to the evidence for a fat-sugar seesaw is the inverse relationship between intakes of fat and sugar in peoples diets. Daily dietary intakes show variation, both between individuals and on an individual basis (e.g. de Castro 1991). These variations tend to produce a wide range of different correlations between various

macronutrient intakes in individuals, perhaps suggesting that people respond differently (i.e. that they exhibit individual differences) to alterations in the macronutrient composition of the diet. Having said that, however, each correlation is poorly estimated from just seven daily observations (such as when taken from 7-day food intake records); therefore, it is possible that much of the variation in correlation is due to sampling rather than differences in the dietary behaviours of people. Moreover, sampling distribution of the correlation coefcient is complex, usually being described through approximations, such as Fisher (1922) and Konishi (1978). To test the concept of the fat-sugar seesaw, this study examined correlations between dietary macronutrients, between individuals and on an individual basis over a 7-day period using dietary data from the National Diet and Nutrition Survey (NDNS) for adults, which were collected in 2000 and 2001 (Henderson et al. 2003). We also examined relationships between the macronutrient compositions of foods in the Food Standard Agencys (FSAs) Nutrient Databank. The Nutrient Databank includes energy and nutrient information on some 7000 foods, and was used by the NDNS researchers to calculate energy and nutrient intakes from the weights of foods and drinks recorded by the participants.

Methods
Collection of the NDNS data used here took place between July 2000 and June 2001 and included a nationally representative sample of adults aged 1964 years from England, Wales and Scotland (Henderson et al. 2003). Food intake records, using the 7-day weighed food intake method, were provided by n = 1724 people. In the current analysis, total energy intake was divided into seven components: fat, total sugar, intrinsic sugars, non-milk extrinsic sugars (NMES), non-sugar carbohydrates, protein and alcohol. Here total sugar refers to all intracellular and extracellular dietary sugars and milk sugars, whether these were added to foods or are naturally occurring. Intrinsic sugar is that contained within the cell walls of foods (such as in fruits) and milk sugars (e.g. lactose). Non-milk extrinsic sugars are those not contained within the cell walls, such as sugar from honey and table sugar, but excluding milk sugars. Non-sugar

2012 The Authors Journal compilation 2012 British Nutrition Foundation Nutrition Bulletin, 37, 213223

Fat and sugar relationship in the NDNS

215

carbohydrate was calculated as total available carbohydrate minus total sugar, and comprises available complex carbohydrate (starch, dextrins and glycogen) and oligosaccharides, and does not include bre. The percentage contribution to mean daily energy intake (% energy) for each macronutrient was calculated using the energy equivalent values for each macronutrient and the mean daily intake weight (g) of each macronutrient. The energy equivalent values used were 16 kJ/100 g for non-sugar carbohydrate, intrinsic sugar, NMES and total sugar, 17 kJ/100 g for protein, 37 kJ/ 100 g for fat and 29 kJ/100 g for alcohol (Holland et al. 1991). The percentage energy contribution to mean daily energy intake was calculated for each macronutrient as: % energy(macronutrient) = 100 [(weight(g) of macronutrient energy equivalent value of macronutrient(kJ/g))/mean daily energy intake(kJ)]. Mean daily energy intake was calculated from the sum of the products of the individual macronutrient intakes (g) the individual energy equivalent values (kJ/g). Between-subject associations between the macronutrient intakes were examined by calculating mean daily macronutrient intakes from the 7-day food intake records from each of the n = 1724 subjects. To try to account for differences in energy requirements (at least partially), energy intake relative to estimated basal metabolic rate (BMR) was calculated using the Schoeld (1985) equations. The resulting mean relative daily macronutrient intake values were examined for statistically signicant associations using Pearsons productmoment-correlation coefcient. Within-subject associations between the macronutrient intakes were also examined to compare them with between-subject associations and to explore the possibility that the relationships between macronutrient intakes varied from one subject to the next. To do this, associations among the macronutrient intakes were calculated separately for each of the n = 1724 subjects using their individual daily macronutrient intake values from the 7-day food records. This produced 1724 estimates of the associations between each pair of macronutrients (e.g. fat and total sugar), each of which was estimated from seven values of daily macronutrient intake taken from one subject. Each pair of macronutrients was analysed for statistically signicant associations using Pearsons product-moment-correlation coefcient. A sampling distribution was calculated to test how much of the variation in the within-subject associations between fat and total-sugar intake was a consequence of sampling rather than a true subject effect. The sampling distribution was calculated by estimating from a large number (n = 10 000) of simulated samples of size 7

(representing the 7 days of the food records) with the true simulated correlation set equal to the mean of the observed correlations calculated from the dietary data from the NDNS. Correlations between the different macronutrients were examined, using Pearsons product-moment-correlation coefcient, in the FSAs Nutrient Databank of foods and beverages that was used to estimate energy and nutrient intakes in the NDNS survey. Nutrient concentrations per 100 g of food product were provided for n = 7374 foods and beverages. A small proportion (approximately 5%) of these was expressed as smaller amounts (e.g. per 1 g). The majority of these foods were dietary supplements (such as vitamin and mineral supplements) and were excluded from the current analysis. Therefore, the number of food items that were included in the analysis of the Nutrient Databank was n = 6990. Associations between alcohol and the other macronutrients are not reported here because there are very few alcohol-containing foods. This study did not consider the saturated fatty acid (SFA) content of the diet, specically when looking at relationships between macronutrients (especially sugars) or upon examination of the foods in the Nutrient Databank. Because the SFA content of the diet is more strongly related to some diseases, such as cardiovascular disease, total fat was considered a more important variable for analysis with regard to energy balance, and for this reason we examined associations with total fat rather than SFAs. All analyses were performed using the R statistical computing package (Version 2.14.0; R Development Core Team 2010).

Results
Mean daily energy and macronutrient intakes according to each gender and for the total sample are shown in Table 1.

Between-subject associations
Figure 1 shows a scatter plot matrix of individual means of the absolute macronutrient intakes (g/day). These are all positively correlated (P < 0.001). Even when absolute intakes are expressed relative to estimated BMR, the correlations remain positive (correlations not reported here). When the macronutrients were expressed as their relative contribution to mean daily energy intakes (i.e. percentage energy), there was evidence of a fat-sugar seesaw, in that diets with high percentage energy intakes from fat had low percentage energy intakes from sugar. Correlations of percentage energy intake from fat vs. percentage energy intake from

2012 The Authors Journal compilation 2012 British Nutrition Foundation Nutrition Bulletin, 37, 213223

216

G. W. Horgan and S. Whybrow

Table 1 Mean daily energy and macronutrient intakes


Women 958 n Energy (MJ) Energy (kcal) Energy intake: BMR CHO (g) Fat (g) Protein (g) Alcohol (g) Total sugar (g) Intrinsic sugar (g) NMES (g) CHO (% energy) Fat (% energy) Protein (% energy) Alcohol (% energy) Total sugar (% energy) Intrinsic sugar (% energy) NMES (% energy) Mean ( SD) 8.03 (2.52) 1921(603) 1.37 (0.46) 232.2 (78.5) 71.1 (27.0) 72.9 (23.3) 15.6 (21.4) 100.7 (46.9) 38.4 (20.2) 62.3 (39.8) 46.5 (7.1) 32.6 (6.0) 15.7 (3.1) 5.3 (6.7) 20.0 (6.6) 7.9 (3.9) 12.1 (6.2) Men 766 Mean ( SD) 8.22 (2.65) 1967(634) 1.07 (0.36) 236.5 (81.4) 73.6 (29.1) 76.0 (31.1) 14.6 (19.8) 102.0 (47.8) 38.3 (20.2) 63.7 (40.9) 46.3 (7.4) 32.8 (6.4) 16.0 (3.7) 4.9 (6.3) 19.8 (6.7) 7.7 (3.9) 12.1 (6.4) All 1724 Mean ( SD) 8.11 1940 1.24 234.1 72.2 74.3 15.2 101.3 38.3 62.9 46.4 32.7 15.8 5.1 19.9 7.8 12.1 (2.58) (617) (0.44) (79.8) (28.0) (27.1) (20.7) (47.3) (20.2) (40.3) (7.2) (6.2) (3.4) (6.5) (6.6) (3.9) (6.3)

seven separate daily intake records. The mean correlation between the weights (g/day) of fat and total sugar obtained was positive (r = 0.24, n = 12 068, P = < 0.001). This differed signicantly from zero, thus indicating that there were more positive than negative correlations between fat and total sugar (g/day). Figure 2 also shows the theoretical distribution of correlations between two variables estimated from 10 000 simulated samples of size 7 (corresponding to the seven daily food intake records), when the true correlation between the two variables is r = 0.24. The t between the theoretical distribution of correlations (line in Fig. 2) and the observed distribution of correlations (bars in Fig. 2) from the subjects data is remarkably close.

Associations within the Nutrient Databank


Figure 3 illustrates a scatter plot of the amount (g/100 g) of fat and total sugar in each of the n = 6990 food items in the Nutrient Databank that were included in the analysis. The correlation between fat and total sugar in the food items of the Nutrient Databank is very weakly positive (r = 0.03, n = 6990, P = 0.006) and many of the food items are concentrated in the bottom left of the plot. This association is also illustrated in log scale format for visual emphasis (Fig. 3b). Two clusters can be seen within this plot, one smaller cluster of foods containing substantially more of both fat and sugar than those in the other larger cluster. Closer inspection of the Nutrient Databank revealed that items in this smaller cluster were mostly the more processed and composite foodstuffs, whereas the larger cluster consisted more of raw food ingredients. However, even within the smaller cluster, which contained foods with a sugar content of at least 10 g/100 g, there was a weak, but positive, correlation (r = 0.09, n = 1539, P < 0.001) between fat and sugar, whereas this association was negative in the larger cluster (r = -0.18, n = 5451, P < 0.001). Table 3 shows the level of associations among fat, sugars, carbohydrate and protein in the n = 6990 foods in the Nutrient Databank. Fat showed a signicant positive correlation with total sugar, NMES, carbohydrate and non-sugar carbohydrate, and a signicant negative correlation with intrinsic sugar. Protein showed a signicant negative correlation with total sugar, intrinsic sugar, NMES, carbohydrate and non-sugar carbohydrate. Thus, fat had a positive correlation with all macronutrients (except intrinsic sugar) whereas protein had a positive correlation with fat and a negative correlation with all other macronutrients.

BMR, basal metabolic rate; CHO, carbohydrate; NMES, non-milk extrinsic sugar; SD, standard deviation; % energy, percentage energy (see Methods).

the various sugars were (r = -0.40, n = 1724, P < 0.001) for total sugar; (r = -0.22, n = 1724, P < 0.001) for NMES; and (r = -0.33, n = 1724, P < 0.001) for intrinsic sugars.

Within-subject associations
Table 2 shows the mean within-subject correlations between the weights (g/day) of the various macronutrients of the reported diet as calculated from each subjects seven separate daily intake records. All correlation values in Table 2 are positive and statistically signicant (P < 0.001), indicating not only a lack of fat-sugar seesaw in the diets of individuals, but the opposite. The positive correlations between the macronutrients (Table 2) show that (on average) on days when subjects consumed more of one macronutrient than they did on other days, they also consumed more of all of the other macronutrients. For example, on days when subjects consumed more fat than they did on average on other days, they also consumed more intrinsic sugar, more NMES and more total sugar, rather than less, as would be predicted by the fat-sugar argument. Figure 2 shows a histogram of the distribution of within-subject correlations between the weights (g/day) of fat and total sugar, calculated from each individuals

2012 The Authors Journal compilation 2012 British Nutrition Foundation Nutrition Bulletin, 37, 213223

Fat and sugar relationship in the NDNS

217

100 200 300 400

50

100

150

200 100 150 200

Fat

100 200 300 400

Sugar

Non-sugar CHO

150

200

50

100

Protein

50

100 150 200

100

200

300

Figure 1 Scatter plot matrix of individual means (g/day) of the macronutrient components. To reduce clutter, axes are marked at the top and bottom, and left and right, on alternate plots. CHO, carbohydrate.

Discussion
Principal ndings
In this study of the correlations between fat and sugar(s) in the diets of NDNS subjects, and the FSAs Nutrient Databank, which was used to calculate energy and

nutrient intakes from the weights of foods and drinks recorded by the subjects, the correlations between fat and sugar(s) were calculated in ve ways: (1) between individuals in percentage terms (percentage energy); (2) between individuals in absolute amounts (g); (3) between individuals in amount (g) relative to BMR; (4) within individuals over 7 days; and (5) between food

2012 The Authors Journal compilation 2012 British Nutrition Foundation Nutrition Bulletin, 37, 213223

100

200

300

50

218

G. W. Horgan and S. Whybrow

Table 2 Within-subject correlations for the macronutrient components of the diet (g/day)
Fat Fat Total sugar Intrinsic sugar Non-milk extrinsic sugars Carbohydrate Non-sugar carbohydrate Protein r = 1.00 r = 0.24 r = 0.15 r = 0.23 r = 0.42 r = 0.41 r = 0.50 Total sugar Intrinsic sugar NMES CHO Non-sugar CHO Protein

r = 1.00 r = 0.51 r = 0.85 r = 0.68 r = 0.20 r = 0.19

r = 1.00 r = 0.11 r = 0.42 r = 0.20 r = 0.24

r = 1.00 r = 0.58 r = 0.16 r = 0.13

r = 1.00 r = 0.79 r = 0.36

r = 1.00 r = 0.36

r = 1.00

For all correlations, n = 12 068, P < 0.001. NMES, non-milk extrinsic sugar; CHO, carbohydrate.

Fat-sugar correlations

(a)
100 90

0.8

80 70
Total sugar (g/100g)

0.6

60 50 40 30 20 10

Frequency

0.2

0.4

0 0 20 40
Fat (g/100g)

60

80

100

(b)
0.0

100
1.0 0.5 0.0 Correlation 0.5 1.0

10
Total sugar (g/100g)

Figure 2 Histogram of within-individual correlations for fat and total-sugar intake (g/day), calculated for each subject. The theoretical sampling variation for a population with a constant correlation is also shown as the superimposed smooth curve.

1 0.01 0.1 1 10 100

items in the Nutrient Databank. Only when examining the correlation between fat and sugar(s) between individuals in percentage terms (percentage energy) was the fat-sugar seesaw evident; in all other methods, the correlation between fat and sugar(s) was positive. Results showed that in a free-living human population, macronutrient intakes do not exhibit a fat-sugar seesaw effect when expressed in absolute terms (i.e. weight in g). There was no inverse relationship between any one macronutrient and another, and specically there was no inverse relationship between fat and intrinsic sugar, NMES or total sugar, but rather there was a

0.1

0.01
Fat (g/100g)

Figure 3 (a) Fat and sugar amount per 100 g in all items in the National Diet and Nutrition Survey food composition database. (b) The same plot on a log scale.

2012 The Authors Journal compilation 2012 British Nutrition Foundation Nutrition Bulletin, 37, 213223

Fat and sugar relationship in the NDNS

219

Table 3 Food database correlations in the macronutrient composition of foods (expressed as g/100 g of food)
Fat Fat Total sugar Intrinsic sugar NMES CHO Non-sugar CHO Protein r = 1.00 P < 0.001 r = 0.03 P = 0.006 r = -0.02 P = 0.041 r = 0.04 P < 0.001 r = 0.09 P < 0.001 r = 0.10 P < 0.001 r = 0.17 P < 0.001 Total sugar Intrinsic sugar Extrinsic sugar CHO Non-sugar CHO Protein

r = 1.00 P < 0.001 r = 0.38 P < 0.001 r = 0.96 P < 0.001 r = 0.72 P < 0.001 r = 0.07 P < 0.001 r = -0.24 P < 0.001

r = 1.00 P < 0.001 r = 0.11 P < 0.001 r = 0.27 P < 0.001 r = 0.02 P = 0.086 r = -0.12 P < 0.001

r = 1.00 P < 0.001 r = 0.70 P < 0.001 r = 0.07 P < 0.001 r = -0.22 P < 0.001

r = 1.00 P < 0.001 r = 0.74 P < 0.001 r = -0.20 P < 0.001

r = 1.00 P < 0.001 r = -0.06 P < 0.001

r = 1.00 P < 0.001

For all correlations, n = 6990. CHO, carbohydrate; NMES, non-milk extrinsic sugar.

mild positive one. In most cases, this lack of inverse relationship was not a consequence of dietary behaviour; rather it is possibly connected to the macronutrient composition of specic food items that the study participants chose to consume. Given that mean withinsubject correlation between absolute intakes (i.e. weight in g) of fat and total sugar was positively associated, it can be concluded that fat and sugar do not displace each other in absolute terms (i.e. gram for gram); at least this was the case in free-living individuals self-reporting their habitual diets. Our results indicate that on days when more of one macronutrient (e.g. fat) is consumed compared with other days, then there is also a tendency to consume more of the other macronutrient [in this case sugar(s)] as well.

Comparison with other studies


Numerous epidemiological studies have shown an inverse relationship between sugar and fat consumption when macronutrients are expressed as their percentage contribution to total energy intake (e.g. Baghurst et al. 1994; Bolton-Smith & Woodward 1994; Gibney et al. 1995; Alexy et al. 2003; Dwyer et al. 2003); indeed, such relationships were observed in the current analysis. Furthermore, when Australian adults were divided into quintiles (fths) of percentage energy intake from fat, Baghurst et al. (1994) found signicant negative trends of percentage energy intake from fats and sugars. Baghurst et al. (1994) grouped sugars differently than in the current analysis, but they reported signicant negative trends of percentage energy intake from fats and

simple sugars (monosaccharides and disaccharides whether occurring naturally or added to foods and equivalent to the sum of intrinsic sugars and NMES reported in the current paper), natural sugars (all sugars other than those added to processed foods) in both men and women, and for added sugars (all sugars that had been added to processed foods or added by the consumer) in men, but not women. Calculating absolute intakes as weight in g of sugars from the published values of Baghurst et al. (1994) shows that the weight of simple sugars, natural sugars and added sugars for men, but not women, also decreases with quintiles of increasing percentage energy from fat. Thus, the data presented by Baghurst et al. (1994) appear to show a fat-sugar seesaw even when sugar is expressed as weight and fat as a percentage of energy intake, at least in men. When the NDNS data used in the current analysis were divided up into quintiles of percentage energy from fat using the same methods as Baghurst et al. (1994), similar relationships to those reported by Baghurst et al. (1994) were apparent; weight of intrinsic sugars (g), total sugars (g) and percentage energy from NMES decreased with increasing quintiles of percentage energy intake from fat (values not shown here), suggestive of a fat-sugar seesaw. This is an apparent contradiction to the results of the regression analysis, conducted on the same data, showing positive correlations between fat and sugar(s) (Table 2), and contesting the fat-sugar seesaw; the weight of fat and sugar(s) in the diet appears to have contradictory relationships, being negative when intakes of sugar(s) are considered against quintiles of the percentage energy intake from fat, yet positive

2012 The Authors Journal compilation 2012 British Nutrition Foundation Nutrition Bulletin, 37, 213223

220

G. W. Horgan and S. Whybrow

when considered against intakes of fat expressed as weights (g). This apparent paradox is a result of the seesaw in percentage energy from fat and absolute amount of sugar (g) being an artefact rather than a real dietary effect; this is simply a weaker version of the correlation, expressed as percentage energy from fat and percentage energy from sugar that arises from the mere operation of taking percentages instead of using weight in the analysis. Furthermore, if sugar, fat, protein and non-sugar carbohydrate are independent variables (i.e. no seesaw effect), then sugar (g) and the percentage energy from fat will be negatively correlated. These will even be negatively correlated when there is a weak positive correlation between sugar (g) and fat (g). The reason for this is that expressing one (or more) of the macronutrients as a percentage, whether percentage contribution to energy or total weight, introduces a dependency between the macronutrients. Consider, for example, a range of theoretical foods that comprise only fat and sugar in variable amounts. When the macronutrient composition of these foods is expressed as a percentage of total energy content of the food, there must be a perfect negative correlation (or seesaw) between percentage energy from fat and percentage energy from sugar, because the total must equal to 100% and foods that have a high percentage energy from fat must have a low percentage energy from sugar, and vice versa. If further macronutrients (non-sugar carbohydrate and protein) are introduced into the model, the same effect remains in operation, albeit less strongly. The mechanism for this is that foods that have a high percentage energy from fat must have a low percentage energy from the sum of the remaining macronutrients (in this case sugar, non-sugar carbohydrate and protein). At 37 kJ/100 g, the energy density of dietary fat is approximately 2.3 times that of sugar (16 kJ/100 g) (Holland et al. 1991). This has the effect of strengthening the correlation between fat and sugar when one, or indeed both, are expressed as their percentage contribution to total energy; the association would still be negative if fat and sugar had the same energy densities because the different energy densities of fat and sugar are constants and have a simple scaling effect. The above negative relationships between percentage energy from fat and weight (g) of sugar(s) are evident even in dietary data fabricated from random numbers emphasizing that the fat-sugar seesaw is a mathematical artefact rather than a dietary effect. This can be seen by using the random number function of a spreadsheet program to generate fat, sugar, non-sugar carbohydrate and protein values for theoretical foods and plotting percentage energy from fat against weight

(g) of sugar. The authors may be contacted for such a spreadsheet. That fat and sugar, or even fat and carbohydrate, are reciprocally related in percentage terms in the diet is almost inevitable because they are the main energyproviding macronutrients in typical Western style diets. The contribution of the other main macronutrient (protein) to daily energy intakes tends to show less variation. When expressed as absolute amounts (g/100 g), there is a positive relationship between fat and sugar in peoples diets (Emmett & Heaton 1995; Macdiarmid et al. 1995; Drewnowski et al. 1997), and as reported here and elsewhere (Stubbs et al. 2001) the relationship between fat and sugar in ready-to-eat foods in food composition tables is far less evident (see below). In the present study, clear positive relationships were evident among the macronutrients (g) in the recorded dietary intake data. When macronutrient intakes were expressed as weight (g), and energy requirements were taken into account, albeit approximately by calculating intake relative to estimated BMR, the correlations between pairs of macronutrients were all positive. Thus, as expected, people with higher energy requirements tended to achieve the necessary higher energy intakes by consuming more of all the macronutrients and not by discernibly selecting more of one and less of another as a seesaw effect would suggest. The estimated correlations between fat and total sugar varied greatly between subjects (Fig. 2). For example, for some people (i.e. those with positive correlations between fat and total sugar), on the days on which they consumed more fat they also consumed more total sugar, whereas for others (i.e. those with negative correlations between fat and total sugar) whenever they consumed more fat they consumed less total sugar. Thus, at face value, Figure 2 could be taken to suggest that there are different dietary behaviour phenotypes, one in which the fat-sugar seesaw operated in their diet selection and the other in which it does not. However, this could all be explained by sampling variation. The t between the theoretical range of correlations (line in Fig. 2) and the observed range of correlations (bars in Fig. 2) in the subjects is remarkably close. Thus, it is not possible to infer variation at all in within-subject macronutrient correlation.

Associations within the Nutrient Databank


In the analysis when the contributions of fat and sugar were plotted against each other (Fig. 3b), there appeared to be two main clusters of foods stemming from within the Nutrient Databank. Closer inspection

2012 The Authors Journal compilation 2012 British Nutrition Foundation Nutrition Bulletin, 37, 213223

Fat and sugar relationship in the NDNS

221

revealed that the relationship between fat and sugar tended to differ between the two cluster groups, such that, foods that made up the smaller cluster were typically the more processed foods, which tend to have relatively large amounts of both sugar and fat. Commercially available snack foods tend to be mixtures of fats and sugars (e.g. chocolate), or fats and starches (e.g. biscuits). They tend to be low in protein and have a low moisture content (Whybrow et al. 2007). Commonly within the food supply chain, sugar acts as a vehicle for fat by increasing the palatability of highfat foods (Emmett & Heaton 1995), whereas lessprocessed foods (such as fruits) and raw food ingredients (such as margarine, cooking oils, our and table sugar) tend to be composed predominantly of fat, starch or sugar. Another group of foods that can be high in fat with little or no sugar, but with a relatively large amount of protein, is meat (and some sh). In processed foods, sugar(s) tend to be positively associated with fat, while in raw foods/ingredients they tend not to be. The correlations between fat and sugar in the two clusters are consistent with this; being positive in the smaller cluster and negative in the larger cluster. Overall, these clusters may be reected in the associations in the overall diet, depending on how much the diet is comprised of processed foods or lessprocessed foods. Accordingly, there may be a different relationship between fat and sugar in the context of the overall diet dependent upon whether or not the sugar comes from processed or raw foods. However, the overall relationship between fat and sugar(s) in the diets of individuals in the NDNS was positive. In this study, a seesaw effect was not evident from the assessment of the dietary behaviour of free-living subjects not undergoing any specic dietary interventions. Having said that, a seesaw effect may arise in subjects attempting to modify their weight by altering their diet. This is because, while weight loss diets have many strategies, reducing fat or carbohydrate intake is often suggested. Under these circumstances, individuals might compensate by increasing their intake of other macronutrients, in absolute terms, relative to what they were consuming beforehand. However, this does not appear to be the case for ad libitum low-fat (Carmichael et al. 1998) or low-carbohydrate diets (Samaha et al. 2003), although more rigorous manipulations, such as Atkins type diets, which severely reduce the amount of carbohydrate consumed in the diet, do result in an increase in the intake of other macronutrients (Gardner et al. 2007). The overall conclusion seems to be that compensation is weak, while supplementing the diet with either fat or sugar does not produce macronutrient-specic

compensation (Raben et al. 2002; Mazlan et al. 2006; Whybrow et al. 2007). The relationship between fat and sugar may also be different in populations whose diets comprise very few processed foods. It has been suggested that a decrease in the consumption of extrinsic sugars could also bring about a decrease in fat consumption because of the positive association between these two nutrients in many processed foods (Emmett & Heaton 1995). The lack of association between fat and sugar in lessprocessed foods suggests that in diets with few processed foods, a decrease in sugar intake may lead to an increase in fat intake. In part, this would depend on how raw foods and ingredients are combined in the diet. Comparing dietary composition in percentage terms alone can be misleading. It is a feature of compositional data to introduce negative correlations between macronutrients, because when any set of independent variables is expressed as a percentage of the total, these percentages are more likely than not to be negatively correlated. In such circumstances, the correlation is a statistical artefact, which gives little real information about how the original variables relate to one another, and only when dietary intakes are examined as absolute values (g), as well as percentages, can the interrelationships between the different macronutrients be studied fully.

Strengths and limitations of the study


One issue that casts a shadow over all dietary analysis surveys (such as the NDNS used here) is the misreporting of food intake. For example, it is well known that people tend to report lower intakes than they have consumed, to eat less than they habitually do and even to change their pattern of consumption (Macdiarmid & Blundell 1998; Stubbs et al. 2003). With regard to under-reporting, Emmett and Heaton (1995) found that while sugar and fat intakes were positively related when people reporting low energy intakes were included in the analysis, the relationship became non-signicant when low energy reporting was accounted for. The authors attributed this to biased misreporting of certain foods. Moreover, misreporting is often clearly evident when dietary intakes are implausibly low for someone in assumed energy balance (Black 2000). Unfortunately, however, there is no satisfactory way to remove misreporting effects from a dietary survey. Omitting subjects with lower reported intakes only causes bias and higher reported intakes are also affected by misreporting, because they are more likely in those with higher activity

2012 The Authors Journal compilation 2012 British Nutrition Foundation Nutrition Bulletin, 37, 213223

222

G. W. Horgan and S. Whybrow


Baghurst KI, Baghurst PA & Record SJ (1994) Demographic and dietary proles of high and low fat consumers in Australia. Journal of Epidemiological and Community Health 48: 2632. Black AE (2000) Critical evaluation of energy intake using the Goldberg cut-off for energy intake: basal metabolic rate. A practical guide to its calculation, use and limitations. International Journal of Obesity 24: 111930. Bolton-Smith C & Woodward M (1994) Dietary composition and fat to sugar ratios in relation to obesity. International Journal of Obesity 18: 8208. Carmichael HE, Swinblirn BA & Wilson MR (1998) Lower fat intake as a predictor of initial and sustained weight loss in obese subjects consuming an otherwise ad libitum diet. Journal of the American Dietetic Association 98: 359. De Castro JM (1991) Weekly rhythms of spontaneous nutrient intake and meal pattern of humans. Physiology & Behavior 50: 72938. Drewnowski A, Henderson SA, Shore AB et al. (1997) The fatsucrose seesaw in relation to age and dietary variety of French adults. Obesity Research 5: 5118. Drummond S, Kirk TR & De Looy A (1996) Are dietary recommendations for dietary fat reduction achievable? International Journal of Food Sciences and Nutrition 47: 2216. Dwyer JT, Michell P, Cosentino C et al. (2003) Fat-sugar see-saw in school lunches: impact of a low fat intervention. Journal of Adolescent Health 32: 42835. Emmett PM & Heaton KW (1995) Is extrinsic sugar a vehicle for dietary-fat? Lancet 345: 153740. Fisher RA (1922) On the probable error of a coefcient of correlation deduced from a small sample. Metron 1: 332. Gardner CD, Kiazand A, Alhassan S et al. (2007) Comparison of the atkins, zone, ornish, and LEARN diets for change in weight and related risk factors among overweight premenopausal women: the A TO Z weight loss study: a randomized trial. Journal of the American Medical Association 297: 96977. Gibney MJ, Sigman-Grant M, Stanton JL Jr. et al. (1995) Consumption of sugars. American Journal of Clinical Nutrition 62: 178S 94S. Gibson S (1997) Obesity: is it related to sugar in childrens diets? Nutrition & Food Science 97: 1847. Henderson L, Gregory J, Irving K et al. (2003) The National Diet and Nutrition Survey: Adults Aged 19 to 64 Years. Vol. 2: Energy, Protein, Carbohydrate, Fat and Alcohol Intake. London: The Stationery Ofce. Holland B, Welch AA, Unwin ID et al. (1991) The Composition of Foods, Cambridge, The Royal Society of Chemistry and Ministry of Agriculture, Fisheries and Food. Konishi S (1978) An approximation to the distribution of the sample correlation coefcient. Biometrika 65: 6546. Macdiarmid J & Blundell J (1998) Assessing dietary intake: who, what and why of under-reporting. Nutrition Research Reviews 11: 23153. Macdiarmid JI, Cade JE & Blundell JE (1995) Extrinsic sugar as vehicle for dietary fat. Lancet 346: 6967. Macdiarmid JI, Vail A, Cade JE et al. (1998) The sugar-fat relationship revisited: differences in consumption between men and women of varying BMI. International Journal of Obesity 22: 105361.

levels. For these reasons, we did not attempt any adjustment or editing of the data to deal with misreporting. To check whether misreporting might have inuenced our results, we looked for associations with misreporting indicators (i.e. reported energy intake relative to estimated energy requirements) and found none; there was no signicant association between within-subject fatsugar correlation and energy intake relative to estimated energy requirements. The NDNS data used here was collected over a decade ago. Accordingly, the macronutrient composition of some processed foods will have changed over that time, particularly the fat and sugar content of foods. Such changes can inuence the strength of associations between fat and sugar in the foods people choose to consume and can also inuence overall diet. These are, however, unlikely to alter the direction of the associations.

Conclusions
In the diets of free-living populations, macronutrient intakes do not show a seesaw relationship when expressed in absolute terms (i.e. g/100 g). There is no indication that people strictly choose between fats and sugar, or that higher intakes of one macronutrient are associated with lower intakes of the other, as might be interpreted from the fat-sugar seesaw argument. Comparing diet composition in percentage terms alone can be misleading. It is only when dietary intakes are examined as absolute values, as well as relative to their contribution to energy, that the interrelationships between the different macronutrients can be studied fully.

Acknowledgements
This work was supported by The Scottish Governments Rural and Environment Science and Analytical Services Division (RESAS).

Conict of interest
The authors have no conict of interest.

References
Alexy U, Sichert-Hellert W & Kersting M (2003) Associations between intake of added sugars and intakes of nutrients and food groups in the diets of German children and adolescents. British Journal of Nutrition 90: 4417.

2012 The Authors Journal compilation 2012 British Nutrition Foundation Nutrition Bulletin, 37, 213223

Fat and sugar relationship in the NDNS


Mazlan N, Whybrow S, Horgan G et al. (2006) Effects of increasing increments of fat and sugar-rich snacks into the diet on energy and macronutrient intake in lean and overweight men. British Journal of Nutrition 96: 596606. Mccoll KA (1988) The sugar-fat seesaw. British Nutrition Foundation Nutrition Bulletin 13: 1149. Parnell W, Wilson N, Alexander D et al. (2008) Exploring the relationship between sugars and obesity. Public Health Nutrition 11: 8606. R Development Core Team 2010. R: a language and environment for statistical computing. R Foundation for Statistical Computing. Vienna, Austria. Raben A, Vasilaras TH, Moller AC et al. (2002) Sucrose compared with articial sweeteners: different effects on ad libitum food intake and body weight after 10 wk of supplementation in overweight subjects. American Journal of Clinical Nutrition 76: 7219.

223

Samaha FF, Iqbal N, Seshadri P et al. (2003) A low-carbohydrate as compared with a low-fat diet in severe obesity. New England Journal of Medicine 348: 207481. Schoeld WN (1985) Predicting basal metabolic rate, new standards and review of previous work. Human Nutrition: Clinical Nutrition 39: 541. Stubbs RJ, Mazlan N & Whybrow S (2001) Carbohydrates, appetite and feeding behavior in humans. Journal of Nutrition 131: 2775S81S. Stubbs RJ, Oreilley L, Fuller Z et al. (2003) Detecting and Modelling Mis-Reporting of Food Intake in Adults. Food Standards Agency: London. Whybrow S, Mayer C, Kirk TR et al. (2007) Effects of two-weeks mandatory snack consumption on energy intake and energy balance. Obesity Research 15: 67385.

2012 The Authors Journal compilation 2012 British Nutrition Foundation Nutrition Bulletin, 37, 213223

Anda mungkin juga menyukai