Original Communication Energy density, energy intake and ... - Nature

European Journal of Clinical Nutrition (2001) 55, 349±359 ß 2001 Nature Publishing Group All rights reserved 0954±3007/01 $15.00 www.nature.com/ejcn

Original Communication Energy density, energy intake and weight status in a large free-living sample of Chinese adults: exploring the underlying roles of fat, protein, carbohydrate, ®ber and water intakes JD Stookey1 1

Department of Nutrition, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, USA

Objective: This paper uses observational data to explore what dietary constituents might be responsible for associations between energy density, energy intake and weight status among free-living individuals. Design: Cross-sectional data on 5783 Chinese adults (aged 20 ± 59 y) from the 1991 China Health and Nutrition Surveys were used to test for associations between energy density, energy intake and weight status, controlling for age, sex, height, activity level, smoking status, urban residence and income. Pearson correlation coef®cients were calculated to identify nutrient intake correlates of energy density. Replacing energy density with its nutrient correlates (3 day mean estimates of fat, protein, ®ber and water intakes) in the models predicting energy intake and overweight status, the independent effects of speci®c nutrients were investigated. Results: Energy density was positively and signi®cantly associated with energy intake and overweight status. Energy density was signi®cantly correlated with every nutrient examined, although the correlations were weak for most variables except water intake. Only water intake behaved consistently across analyses. It was negatively correlated with energy density, negatively, signi®cantly and independently associated with energy intake as well as overweight status. Despite positive associations with energy intake, fat and protein intake were not signi®cant predictors of overweight status. Fiber intake was strongly and positively associated with overweight status. Conclusions: Of the nutrients examined, only water intake appeared to explain the effects of energy density on energy intake and overweight status. Descriptors: energy density; overweight; water intake; China European Journal of Clinical Nutrition (2001) 55, 349±359

Introduction Although high-fat diets have been linked with hyperphagia as well as overweight status and obesity, the causal mechanisms underlying this relationship have not been elucidated. In recent reviews of the clinical and=or epidemiologic literature, several authors have concluded that hyperphagia related to high-fat diets cannot be attributed simply to the fat content of the diet (Poppitt, 1995; Prentice & Poppitt, 1996; Bray & Popkin, 1998; Willett, 1998). Poppitt (1995) writes `recent evidence from both animal and human studies has shown that over-consumption on a *Correspondence: J Dunmeyer Stookey, Carolina Population Center, CB No. 8120 University Square, University of North Carolina at Chapel Hill, 123 W. Franklin St, Chapel Hill, NC 27516-3997, USA. E-mail: [email protected] Guarantor: JD Stookey. Contributor: Hypothesis, data analysis and manuscript preparation all completed by JD Stookey. Received 19 July 2000; revised 27 November 2000; accepted 6 December 2000

high-fat diet is not due to the fat content per se, but rather in combination with the high energy content. It is only high-fat, high-energy foods, not high-fat foods alone that lead to hyperphagia'. Efforts to pinpoint the causal factor(s) have been confounded by the effects of energy density and nutrient colinearity. Prentice & Poppitt (1996), in their review of all longterm ( > 24 h) diet manipulation studies, conclude that changes in energy density appear to be crucial to the control of energy intake. When the energy density of all experimental foods is manipulated, the obligatory change in energy density is directly related to change in energy intake. `When selecting an energy-dense diet, individuals readily consume a suf®cient weight of food to maintain a high energy intake. However, when selecting a low energydense diet, individuals appear unable to eat suf®cient weight of food to maintain a high intake and consequently energy intake falls. The management of weight maintenance and=or weight loss may be achieved by eating a diet of low energy density (p 169)'. Results from existing clinical studies of covertly manipulated energy density

Energy density, energy intake and overweight status JD Stookey

350

(Spiegel, 1973; Garrow et al, 1978; Stubbs et al, 1998; Stubbs et al, 1995a,b; Lissner et al, 1987; Kendall et al, 1991) show a positive association between energy density and weight change. If energy density consistently predicts energy intake and weight change, what then is the underlying mechanism? Identifying the mechanism(s) underlying the consistently observed effects of energy density on energy intake and weight change is complicated by the amorphous nature of energy density. Unlike speci®c nutrients such as vitamin C, the structure of energy density is not a constant. Energy density, the energy content per weight or volume of food, has many determinants (energy intake, total bulk, volume or weight), and can re¯ect a spectrum of diets of different compositions. The above referenced studies of energy density have involved qualitatively different experimental diets, produced by the addition or subtraction of water, arti®cial sweeteners, thickeners, and=or butterfat to liquid diets, or by altering the fat, pasta or vegetable content of solid diets. Although not reported in most studies, the intended dietary manipulation (eg subtraction=addition of fat) necessarily caused shifts in the macro-, micro- and nonnutrient pro®les of the experimental diets. Lissner et al (1987) report shifts in macronutrient composition from 30 ± 35%, 12% and 55% of energy from fat, protein and carbohydrate, respectively, on the medium-fat diet to 15 ± 20%, 13% and 70% of energy on the low-energy dense diet, to 45%, 10% and 42% of energy on the high-energy dense diet, respectively. Using thickening agents in their dietary manipulations, Stubbs et al (1998) found signi®cant changes in water and ®ber intakes, as well as protein, fat and carbohydrate intakes. Since any of the component alterations could ultimately be responsible for the observed effects, it is not clear what mechanism mediates the effects of energy density. Studies that alter energy density by altering only one dietary component at a time are needed to tease apart the effects. This paper uses observational data to explore what dietary constituents might be responsible for associations between energy density, energy intake and weight status among freeliving individuals. Nutrient correlates of energy density are included together in multivariable models to evaluate each nutrient effect holding the other nutrients constant. Methods Data Cross-sectional data from the 1991 China Health and Nutrition Surveys (CHNS) for adults aged 20 ± 59 y were used for this analysis. The CHNS was designed to study factors affecting individual food choice, nutrient intake and nutritional status, with a focus on dietary patterns associated with nutritional de®cit or excess. The sampling frame and survey methodology have been described in detail previously (Popkin et al, 1995; Paeratakul et al, 1998). Trained Chinese interviewers obtained detailed individual-level diet data via 24 h recall for three consecu-

European Journal of Clinical Nutrition

tive days as well as household-level information on changes in food inventory. The household-level data was used to estimate individual consumption of edible oil (Guo et al, 1999), as well as to cross-check the diet recall data. In addition to the dietary data, information was available on age (y), gender, income, urban=rural residence (according to the Chinese census de®nition), smoking status (past, never), height, weight and physical activity. Body weight and height were measured on each individual by trained health workers. Physical activity was determined based on regular daily occupations, and recorded as a multilevel variable (very light, light, moderate, heavy). Very light activity was characterized by working in a sitting position or as an of®ce worker; light activity as work in a standing position; moderate activity as work carried out by drivers or electricians; and heavy activity as the work of farmers. The activity variable was designed by the Chinese Nutrition Society to re¯ect total energy expenditure, and intended for use in calculating the Chinese RDAs. The variable signi®cantly predicts energy intake and weight status among Chinese adults (Paeratakul et al, 1998). The CHNS follows human subjects approval procedures that have been approved both by the University of North Carolina School of Public Health and Chinese Academy of Preventive Medicine human subjects protection committees. The 1991 food composition table (FCT) for China (INFH-CAPM, 1991) was used to calculate mean daily energy and nutrient intakes from the food consumption data for each individual. Energy density was calculated by dividing the reported mean energy intake by the total grams consumed. Energy intakes from all of the foods and beverages included in the 1991 FCT were included in the calculation of total energy intake. As the FCT and associated CHNS datasets only include a few beverages (milk, coconut juice, sugarcane juice, spirits, beer, wine, champagne and brandy), the energy intake estimates in this paper speci®cally include energy from foods, milk, coconut juice, sugarcane juice and alcohol. The mean water and ®ber (non-starch polysaccharide) intakes were expressed per unit of energy intake to distinguish between diets high and low in these nutrients. Body mass index (BMI) was calculated as weight (kg)=height (m2). Subjects were classi®ed as normal or overweight using the World Health Organization BMI cutpoints of 18.5 and 25.0 (WHO Expert Committee, 1995). Undernourished subjects (BMI < 18.5) and subjects with missing data were excluded from the study sample (n 5783). Statistical analyses Effect of energy density on energy intake and weight status. Separate ordinary-least squares (OLS) regression models were ®t to evaluate the relationship between energy density (kJ=g) and mean daily energy intake, and energy density and total amount of foods consumed. These models controlled for age, gender, income, urban=rural residence, smoking status, activity level and height. Height was included in the models as a proxy for body-size-related dietary requirements.


Crude and multivariable logistic regression models were ®tted to test for a positive association between energy density and overweight status. Again, age, gender, income, urban=rural residence, smoking status, activity level and height were included in the multivariable models as covariates. Energy intake was not included in these models, since it is intermediate along the hypothesized causal pathway between energy density and overweight status. In epidemiologic analyses, it is considered inappropriate to control for intermediate variables (Rothman, 1986). In an additional model, the total amount of foods consumed, a potentially independent factor regulating energy intake, was also controlled. If determined by visual=learned portion size cues or limitations of stomach size or gastric emptying (Hunt, 1980), the total amount of food consumed may lead to overweight by a causal pathway independent of energy density. Given that the amount consumed correlates with energy density, this potentially independent determinant might confound the association between energy density and weight status. Holding constant the quantity consumed, the additional model focuses on the effect of qualitative differences in the composition (energy density) of reported diets. The total amount consumed was not included in models predicting energy intake, since energy intake is the product of this covariate and energy density, (g consumed)*(kJ=g consumed). Nutrient ± energy density intercorrelations. Given that previous researchers have reported shifts in fat, protein, carbohydrate, ®ber and water intakes with changes in energy density (Lissner et al, 1987; Stubbs et al, 1998), these nutrients were identi®ed as potential correlates and=or determinants of energy density. Each of these nutrients contributes to energy intake or to the bulk, weight or volume of the food consumed. Poppitt (1995) has shown that the fat content of foods listed in the McCance & Widdowson food composition table varies directly with energy density. Figure 1 illustrates that the fat, carbohydrate, protein, ®ber and water contents of foods listed in the 1991 Food Composition Table for China also vary with energy density. To determine to what extent the overall nutrient intakes (the nutrient content of the whole diet as distinct from the nutrient contents of speci®c foods) linearly covary with energy density in this sample, Pearson correlation coef®cients were calculated. Although the following analyses focus on fat, protein, carbohydrate, ®ber and water intakes, we realize that these nutrients may serve as proxy measures for other nutrient correlates of energy density. Macro- and micronutrients are known to correlate with energy intake and the quantity consumed Ð greater food consumption leading to higher intakes in general (Willett & Stampfer, 1998a,b). Nutrients and non-nutrients that are correlated with energy density may be represented by one or more of the selected nutrients, eg fat-soluble nutrients by fat intakes, and water-soluble vitamins, minerals and electrolytes by water intake. To get a sense of between-nutrient intercorrelations,

Pearson correlation coef®cients were also calculated for other available nutrient intakes. In recognition that high-fat, high-energy (HFHE) diets have been an important and controversial focus of research on appetite regulation and weight control (Bray & Popkin, 1998; Willett, 1998), the nutrient pro®les of observed HFHE diets were examined and compared with diets lower in fat and energy. Diets with fat and energy intakes above the 75th percentile for this sample (32% of energy and 13 080 kJ, respectively) were compared with diets medium in fat and energy (MFME: 25 ± 75th percentile for both nutrients) and diets low in fat and energy (LFLE: < 25th percentile for both nutrients).

351

Effects of nutrient correlates of energy density on energy intake and overweight status. Next, the OLS and logistic regression models with energy density as the main exposure (described above) were re-®tted with energy density replaced by its nutrient correlates. Fat (percentage of energy), protein (percentage of energy), ®ber (g=kJ) and water (g=kJ) intakes were entered as independent variables predicting energy intake and overweight status. Carbohydrate intake was included in these models by default as the reference category. The models are essentially what Willett and Stampfer (1998b) refer to as `multivariate nutrient density models', which are useful for the study of the effects of several nutrients simultaneously. These models examine the effects of speci®c nutrients holding other nutrients constant. Model speci®cation was chosen so as not to violate model assumptions as well as retain clear interpretation of the results. The nutrients were entered as untransformed continuous variables, since the nutrient-outcome bivariate distributions did not violate the model assumptions for linearity. All nutrients were rescaled so that their minimum value was zero and the reference category for these models was diets with the lowest percentage energy from fat and protein, lowest ®ber and water content. The reference category, thus, consists of diets high in carbohydrate, low in protein, fat, ®ber and water. Explicitly, a one-unit increase in fat or protein intake in these models necessarily implies a shift in diet composition of a 1% decrease in energy from carbohydrate. Since ®ber and water intakes are not energy-yielding nutrients, a one-unit increase in either of these nutrients does not imply a shift in energy from carbohydrate. Model coef®cients for water and ®ber intake re¯ect the comparison between diets high in carbohydrate, low in protein, fat, water and ®ber and diets high in carbohydrate, low in protein and fat, but higher in water or ®ber. Age, gender, income, urban=rural residence, smoking status, activity level and height were included in the models as covariates. Lastly, to replicate previous analyses of fat intake and overweight status where nutrient covariates were not held constant, protein, fat and water intakes were dropped from the full model predicting overweight status. The Stata statistical program (StataCorp, 1999) was used for all of the above analyses. European Journal of Clinical Nutrition


352

Results Descriptive sociodemographic and dietary intake data are shown in Table 1. Overweight subjects consumed signi®cantly less energy and total grams of food. The diets of overweight subjects were higher in energy density

Figure 1

with a P-value of 0.06. On average, overweight subjects had higher fat, protein and ®ber (as percentage of energy) intakes, and signi®cantly lower carbohydrate and water intakes than normal-weight subjects. Consistent with current knowledge, overweight status was associated with older age, female sex, urban residence,

Correlations between fat, protein, carbohydrate, ®ber and water contents and energy density for foods in the 1991 China Food Composition Table.



higher income, non-smoking status, and reduced activity level.

as kcal=g, the crude likelihood of overweight increased by 18% (OR 1.18, 95% CI 0.99 ± 1.39). Adjustment for age, gender, urban residence, income, smoking status, activity level and height increased the OR to 1.06 (1.01 ± 1.10) when in kJ units, and 1.26 (1.05 ± 1.50) when energy density was expressed in kcal=g units. Energy density appeared to be even more strongly associated with overweight status after also controlling for the total amount of foods consumed (OR 1.09; CI 1.04 ± 1.05 per unit kJ=g or OR 1.44, CI 1.17 ± 1.76 per unit kcal=g).

Energy density as predictor of energy intake and overweight status The mean (s.d.) energy density of foods consumed by this sample of free-living Chinese was 10.7 (1.8) kJ=g (or 2.6 (0.4) kcal=g). Energy density was positively and signi®cantly associated with energy intake. Adjusting for age, gender, urban residence, income, smoking status, activity level and height, a one unit increase in energy density (kJ=g) was associated with an increase in energy intake of 305.7 (18.2) kJ (see Table 2, model 1). To further illustrate this effect, the difference in predicted energy intake for subjects at the 25th and 75th percentiles of the energy density distribution (9.5 and 11.9 kJ, respectively) for this sample was 734 kJ. Energy density was also negatively and signi®cantly associated with the total amount of foods consumed (g), controlling for the same covariates (model 2). The results from sex-speci®c models did not differ (data not shown). Crude and adjusted odds ratios (OR) were calculated to estimate the effect of increasing energy density on overweight status (see Table 2, models 3 and 4). For a one unit increase in energy density (kJ=g), the crude likelihood of overweight increased by 4% (OR 1.04, 95% CI 1.00 ± 1.08). For a one unit increase in energy density expressed

353

Correlations between energy density and speci®c nutrients Energy density was signi®cantly correlated with the fat, protein, carbohydrate, ®ber and water content of the diets consumed (see Table 3). While total fat intake was positively correlated with energy density, the other four nutrients were negatively correlated. The strongest of these correlations (r 0.91) was observed between energy density and water intake (see Figure 2). The remaining nutrient-energy density correlations ranged between 0.10 and 0.20. Not surprisingly, Table 3 also indicates that fatsoluble vitamin E and cholesterol covaried with fat intake, and water-soluble calcium, potassium and vitamin C correlated with water intake. Table 4 shows the nutrient pro®les of diets varying in fat and energy composition. The HFHE diets differed signi®cantly from the lower fat and energy diets in several respects. HFHE diets were

Table 1 Characteristics of adults aged 20 ± 59 y participating in the 1991 China Health and Nutrition Surveys (n 5783) Total Mean (s.d.) Energy (kJ) Total amount of food consumed (g) Energy density (kJ=g) Fat (percentage of energy) Protein (percentage of energy) Carbohydrate (percentage of energy) Fiber (g) Fiber (g=100 kJ) Water (g) Water (g=kJ) Age (y) Weight (kg) Height (m) BMI (kg=m2) Income (Yuan=y) Female Very low activity Low activity Moderate activity Heavy activity Ever smoker Urban resident

11511.4 1095.4 10.7 24.8 11.7 62.9 12.1 0.11 531.1 0.047 37.2 56.8 159.9 22.2 710.0

(2759.8) (292.5) (1.8) (10.3) (2.1) (11.2) (6.0) (0.05) (207.5) (0.02) (10.8) (8.6) (8.3) (2.7) (518.2)

(%) 53.4 11.6 16.7 18.2 53.5 36.2 33.8

Normal weight (n 4981) Mean (s.d.) 11551.2 1101.2 10.7 24.5 11.6 63.2 12.1 0.10 533.6 0.047 36.5 54.9 160.1 21.4 690.1

(2756.8) (293.5) (1.8) (10.3) (2.1) (11.3) (6.0) (0.05) (208.2) (0.02) (10.7) (7.0) (8.2) (1.7) (511.8)

Overweight (n 802) Mean (s.d.) 11265.0 1059.3 10.8 26.4 12.0 61.2 12.4 0.11 515.1 0.047 41.7 68.5 158.8 27.1 833.5

(%) 51.6 10.5 15.2 17.8 56.5 37.6 31.2

(2767.7)* (283.9)* (1.8){ (10.1)* (2.0)* (10.7)* (5.9) (0.05)* (202.6)* (0.02) (10.5)* (8.0)* (8.9)* (2.3)* (540.1)*

(%) 64.6 18.5 25.8 20.7 35.0 27.7 49.5

*Signi®cantly different from normal weight value at the 0.05 level. Almost signi®cantly different from normal weight value (P < 0.10). Energy intakes were converted from kilocalories to kilojoules using the 4.184 conversion factor.

{



354 Table 2 Associations between energy density, energy intake, the total quantity consumed and likelihood of overweight status in free-living Chinese adults Outcome

Independent variables Intercept Energy density (kJ=g) Age (y) Female Height (m) Low activity Moderate activity Heavy activity Ever smoker Urban residence Income Total amount consumed (g) Adjusted R2

Model 1 Energy intake (kJ) b (s.e.)

Model 2 Total amount consumed (g) b (s.e.)

OR (95% CI)

OR (95% CI)

7178.1 (438.5) 305.7 (18.2)* ÿ14.2 (3.3)* ÿ1032.2 (113.0)* 25.5 (5.4)* ÿ20.9 (127.2) 259.8 (126.8)* 950.6 (126.8)* 212.1 (100.8)* ÿ620.9 (88.3)* ÿ0.02 (0.08)

1823.4 (42.0) ÿ76.2 (1.7)* ÿ1.3 (0.3)* ÿ99.3 (10.8)* 2.3 (0.5)* ÿ0.03 (12.2) 24.5 (12.1)* 86.8 (12.1)* 21.5 (9.7)* ÿ60.0 (8.5)* ÿ0.002 (0.007)

1.06 (1.01 ± 1.10)* 1.04 (1.04 ± 1.05)* 1.55 (1.19 ± 2.02)* 1.00 (0.99 ± 1.02) 0.96 (0.75 ± 1.23) 0.94 (0.72 ± 1.22) 0.52 (0.39 ± 0.68)* 0.77 (0.61 ± 0.98)* 1.31 (1.08 ± 1.60)* 1.00 (1.00 ± 1.00)*

1.09 (1.04 ± 1.05)* 1.05 (1.04 ± 1.05)* 1.62 (1.24 ± 2.11)* 1.00 (0.99 ± 1.01) 0.96 (0.75 ± 1.23) 0.93 (0.72 ± 1.20) 0.50 (0.38 ± 0.65)* 0.76 (0.60 ± 0.97)* 1.34 (1.11 ± 1.64)* 1.00 (1.00 ± 1.00)* 1.00 (1.00 ± 1.00)*

17.4

32.5

Model 3 Model 4 Likelihood of overweight status

Continuous variables were rescaled so that the reference category is male, rural residents, aged 20, with the lowest income, activity levels and dietary energy density. Models 1 and 2 were ordinary-least squares regression models. Models 3 and 4 were logistic regression models. *Signi®cantly different from the reference category value (P < 0.05). Energy intakes were converted from kilocalories to kilojoules using the 4.184 conversion factor.

signi®cantly higher in energy density, fat, protein, vitamin E and cholesterol, and lower in carbohydrate, ®ber, water, vitamin C and sodium. Speci®c nutrients replacing energy density in models predicting energy intake and overweight status Models 1 ± 4 were re®t with fat, protein, ®ber and water intake variables instead of energy density as the main exposure (see Table 5). Results are presented for both sexes together, as the results of sex-speci®c models did not differ and sex interactions in the combined model were not statistically signi®cant. Multicollinearity was not observed in the multivariate nutrient density models (variance in¯ation factors < 10). Models 1 and 5 (Tables 2 and 5) explained a similar proportion of the variation in energy intake (adjusted R2 17%). Controlling for age, gender, height, activity, smoking status, urban residence, income, fat and protein intakes (as percentage of energy) were independently, positively and signi®cantly (P < 0.10 for protein) associated with energy intake. Water and ®ber intakes were negatively and signi®cantly (P < 0.10 for ®ber) associated with energy intake. European Journal of Clinical Nutrition

In both models 7 and 8, holding the other nutrient intakes constant, water intake was strongly, negatively and signi®cantly associated with the likelihood of overweight status. Expressed in g=kJ units, the estimated magnitudes of effect were huge. Expressed as g=kcal, the effects of water intake on risk of overweight status were OR 0.14 (95% CI 0.04 ± 0.42) and OR 0.06 (95% CI 0.01 ± 0.21) for models 7 and 8, respectively. A 1 g=kcal lower water intake was associated with a 16-fold increase in risk of overweight status. Subjects at the 25th percentile of the water intake distribution (with water intakes of 0.15 g=kcal) had a 45% greater risk of being overweight than subjects with water intakes at the 75th percentile (0.24 g=kcal). Contrary to expectation, ®ber intake (g=100 kJ food consumed) was strongly and positively associated with risk of overweight status. Controlling for the other nutrients and covariates, fat and protein intakes were not signi®cantly associated with the odds of being overweight in this sample. When protein, ®ber and water intake were dropped from the full model, the effect of fat intake on overweight status remained non-signi®cant (OR 0.99, 95% CI 0.98 ± 1.00, P 0.08).


355

Figure 2 Correlations between daily intakes of fat, protein, carbohydrate, ®ber and water and overall dietary energy density for adults participating in the 1991 China Health and Nutrition Surveys. European Journal of Clinical Nutrition


356 Table 3

1 2 3 4 5 6 7 8 9 10 11 12

Pearson correlation matrix of energy density with selected nutrient intakes

Energy density Fat (percentage of energy) Protein (percentage of energy) Carbohydrate (percentage of energy) Fiber (g=100 kJ) Water (g=kJ) Vitamin E (mg=kJ) Cholesterol Vitamin C (g) Sodium (mg) Calcium (mg) Potassium (mg)

1

2

3

4

5

6

7

8

9

10

11

12

1.00 0.17* ÿ0.15* ÿ0.13* ÿ0.17* ÿ0.91* 0.14* ÿ0.07* ÿ0.65* ÿ0.04* ÿ0.45* ÿ0.54*

1.00 0.12* ÿ0.94* ÿ0.41* 0.10* 0.54* 0.49* ÿ0.08* 0.02* 0.10* ÿ0.15*

1.00 ÿ0.26* 0.11* 0.13* 0.07* 0.45* ÿ0.07* 0.005 0.32* 0.39*

1.00 0.39* ÿ0.13* ÿ0.49* ÿ0.54* 0.10 ÿ0.002 ÿ0.13* 0.09*

1.00 0.07* ÿ0.06* ÿ0.27* 0.10* 0.13* 0.19* 0.61*

1.00 0.05* 0.20* 0.68* 0.07* 0.49* 0.50*

1.00 0.21* ÿ0.11* 0.05* 0.15* ÿ0.02

1.00 ÿ0.07* ÿ0.02 0.12* 0.01

1.00 0.07* 0.33* 0.40*

1.00 0.09* 0.11*

1.00 0.39*

1.00

*Statistically signi®cant pearson correlation coef®cient (P < 0.05).

Table 4 Nutrient pro®les associated with diets differing in fat and energy content reported by free-living Chinese adults in the 1991 China Health and Nutrition Survey Diet patterns de®ned by fat and energy intake Low fat, low energy (n 306) Mean (s.d.) Energy density (kJ=g) Fat (percentage of energy) Protein (percentage of energy) Carbohydrate (percentage of energy) Fiber (g=100 kJ) Water (g=kJ) Vitamin E (mg=kJ) Cholesterol (mg=kJ) Vitamin C (g=kJ) Calcium (mg=kJ) Sodium (mg=kJ) Potassium (mg=kJ)

9.8 12.0 11.3 76.7 0.14 0.051 0.0027 0.0052 0.011 0.038 0.7 0.18

(1.8) (2.8) (1.6) (3.1) (0.06) (0.021) (0.0011) (0.0098) (0.008) (0.014) (0.3) (0.05)

Medium fat, medium energy (n 1586) Mean (s.d.) 10.6 24.1 11.7 63.6 0.11 0.048 0.004 0.016 0.009 0.039 0.6 0.16

(1.7)* (4.5)* (2.1)* (6.0)* (0.05)* (0.016)* (0.002)* (0.017)* (0.006)* (0.016) (0.3)* (0.04)*

High fat, high energy (n 427) Mean (s.d.) 11.5 38.1 12.0 48.4 0.08 0.045 0.004 0.027 0.008 0.036 0.6 0.15

(1.7)* (4.7)* (2.3)* (6.5)* (0.04)* (0.013)* (0.002)* (0.021)* (0.004)* (0.013) (0.4)* (0.03)*

*Signi®cantly different from low-fat low energy diet (P < 0.05). Energy intakes were converted from kilocalories to kilojoules using the 4.184 conversion factor.

Discussion This study contributes information about nutrient intake(s) that might mediate the consistently observed effects of energy density on energy intake and weight status. To brie¯y summarize the results: in this sample of Chinese adults, energy density was positively and signi®cantly associated with energy intake and overweight status. Energy density was signi®cantly correlated with every nutrient we examined, although the correlations were weak for most variables except water intake. When energy density was replaced by its nutrient correlates in models predicting energy intake, fat and protein intake appeared positively associated with energy intake, while ®ber and water intakes were negatively associated with energy intake. Despite their signi®cant or almost signi®cant associations with energy intake, fat and protein intake were not signi®cant predictors of overweight status. Contrary to expectation, given the negative correlation with energy density and energy intake, ®ber intake was strongly and European Journal of Clinical Nutrition

positively associated with overweight status. Water intake was the only nutrient to behave consistently across analyses. It was negatively correlated with energy density, and negatively, signi®cantly and independently associated with energy intake as well as overweight status. In an attempt to understand how dietary energy density exerts its effects, we took advantage of detailed multifaceted data from the CHNS to examine the big picture of energy density in a single sample of adults Ð how energy density covaries with the quantity consumed, various nutrient intakes and weight status in a single sample of adults. To date, this is the only study in free-living individuals of the effects of energy density on both energy intake and weight status that also controls for relevant covariates. Prentice & Poppitt (1996) report the existence of associations between energy density and energy intake in data from the Cambridge Family Food Survey and the MRC National Survey of Health and Development, but do not show these results, and explicitly state that the associations were not adjusted for activity or body size. There is no


357 Table 5 Associations between determinants of energy density, energy intake, the total quantity consumed, and likelihood of overweight status in free-living Chinese adults Outcome

Independent variables Intercept Fat (percentage of energy) Protein (percentage of energy) Fiber (g=100 kJ) Water (g=kJ) Age (y) Female Height (m) Low activity Moderate activity Heavy activity Ever smoker Urban residence Income Total amount consumed (g) Adjusted R2

Model 5 Model 6 Energy intake (kJ) Total amount consumed (g) b (s.e.) b (s.e.) 10907.3 (418.4) 32.2 (4.2)* 31.0 (17.2){ ÿ1507.9 (781.6){ ÿ30497.8 (1966.4)* ÿ13.4 (3.3)* ÿ1037.2 (113.4)* 25.1 (5.4)* 45.6 (127.2) 331.4 (126.8)* 1130.1 (130.5)* 228.9 (100.4)* ÿ701.7 (88.7)* ÿ0.13 (0.08){ 17.9

685.7 (40.0) ÿ1.9 (0.4)* 6.0 (1.6)* ÿ77.7 (74.4) 8213.1 (188.0)* ÿ1.2 (0.3)* ÿ99.6 (10.9)* 2.4 (0.5)* 1.5 (12.2) 28.5 (12.1)* 106.1 (12.5)* 17.4 (9.6){ ÿ66.9 (8.5)* ÿ0.02 (0.007)* 33.2

Model 7 Model 8 Likelihood of overweight status OR (95% CI)

OR (95% CI)

1.00 1.00 (0.99 ± 1.01) (0.99 ± 1.01) 1.02 1.02 (0.98 ± 1.06) (0.98 ± 1.06) 53.7 57.2 (8.93 ± 322.7)* (9.4 ± 349.0)* 0.00023 5.7e ± 06 (1.99 ± 0.03)* (2.3 ± 0.001)* 1.04 1.04 (1.04 ± 1.05)* (1.04 ± 1.05)* 1.44 1.50 (1.10 ± 1.88)* (1.15 ± 1.97)* 1.00 1.00 (0.98 ± 1.01) (0.98 ± 1.01) 0.94 0.94 (0.73 ± 1.21) (0.73 ± 1.21) 0.92 0.91 (0.70 ± 1.19) (0.70 ± 1.18) 0.49 0.47 (0.37 ± 0.66)* (0.35 ± 0.62)* 0.76 0.76 (0.60 ± 0.97)* (0.60 ± 0.96)* 1.39 1.43 (1.14 ± 1.69)* (1.17 ± 1.74)* 1.00 1.00 (1.00 ± 1.00)* (1.00 ± 1.00)* 1.00 (1.00 ± 1.00)*

Continuous variables were rescaled so that the reference category is male, rural residents, aged 20, with the lowest income, activity levels and dietary energy density. Models 1 and 2 were ordinary-least squares regression models. Models 3 and 4 were logistic regression models. *Signi®cantly different from the reference category value (P < 0.05). Energy intakes were converted from kilocalories to kilojoules using the 4.184 conversion factor.

mention of age, smoking or social class. These authors report ®nding no signi®cant relationship between energy density and energy intake in two of the community data sets they examined. Marti-Henneberg et al (1999) found a signi®cant correlation between energy density and BMI in adult males, but not in females in Spain. Given that the estimated effect of energy density on weight status changed by as much as 22% (from OR 1.18 to 1.44) when controlling for covariates in this study, it seems that known risk factors for overweight status may confound unadjusted effect estimates. Cox and Mela (2000) report correlations between BMI and energy density for some methods of calculating energy density, but not others. Confounding bias or differences in the types of data used to calculate energy density may explain divergent results across studies.

As energy density increased, the grams of total food consumed decreased for this sample population. The results suggest that, holding all other variables constant, the total amount of food consumed decreased with increasing fat content, but increased with increasing water and protein content. These results are not entirely consistent with the clinical ®nding of no signi®cant difference in weight or volume of food consumed with experimental diets of varying energy density (Prentice & Poppitt, 1996). The apparent compensation in the amount consumed, while not large enough to eliminate the effect of energy density on energy intake, might be related to portion size cues or to regulatory mechanisms. Subjects consuming higher-fat foods may have consciously attempted to eat less. Indeed, the signi®cantly lower energy intakes and total grams European Journal of Clinical Nutrition


358

consumed among overweight subjects may indicate underreporting for this group. Although a possible source of error, under-reporting bias should not materially alter the present results. If subjects under-reported consistently across all foods, they would have under-reported both the numerator and denominator of the energy density variable (both the energy and the grams consumed), and the error would cancel out. If, on the other hand, overweight subjects disproportionately underreported foods higher in energy density, the effect would be to weight the energy density for the whole diet towards lower-energy-density foods, reducing the observed energy density, and thereby weakening any association between energy density and overweight status. Energy density signi®cantly predicted overweight status after adjusting for known covariates, including the total amount of foods consumed. Age, gender, activity, body size, socio-economic differentials, portion size, stomach size- or gastric-emptying-related factors do not explain away this association. Energy density appeared to be associated with energy intake and weight status in this large free-living population as it does in more restrictive clinical settings. As expected, all of the nutrient intakes examined were signi®cantly correlated with dietary energy density. Energy density directly covaried with fat intake and indirectly covaried with water and ®ber intakes. Considering that protein contributes to total energy intake and is positively correlated with fat intake, the negative correlation between protein intake and energy density was unexpected. The weak correlations between energy density and all of the nutrients, except water, were also unexpected. These unexpected results may be related to true aspects of Chinese food consumption patterns (ie protein may be consumed with energy-dilute foods such as vegetables), to measurement errors or limitations of the food composition table. The between-nutrient correlations and between-diet comparisons of nutrient pro®les illustrate the complex nature of this nutritional epidemiologic problem. Energy density covaries with a spectrum of nutrients, making it dif®cult to pinpoint which nutrient is responsible for the observed effects. Nutrients covary with other nutrients, again making it dif®cult to interpret the effect of a particular nutrient. Is one nutrient acting as proxy for another? In this paper, fat-soluble nutrients were found to covary with fat intake, and water-soluble nutrients to covary with water intake. Thus, theoretically, before ruling out potential roles for such nutrients, it would be premature to conclude from the present results that fat or water intakes mediate the effects of energy density. The models in this paper essentially examine the effects of water and water-soluble nutrient intakes holding constant fat and fat-soluble nutrient intakes, protein and ®ber. Consistent with the literature on HF, HE diets (Poppit, 1995, Prentice & Poppitt, 1996; Bray & Popkin, 1998; Willett, 1998), high energy intakes were not attributable to fat and fat-soluble nutrient intakes in this study.


Given the strong positive association observed between ®ber intake and overweight status, it is unlikely that ®ber intakes explain the protective effects of energy-dilute diets. Water intake and its correlates are left as the most likely candidates for further research. Clinical studies of water incorporated into foods and satiety provide support for a key role for dietary water intake (Rolls et al, 1999). The results of this study ®t well with the data on water incorporated into foods. The water intake variable used in these analyses was calculated from data on foods and energy-yielding beverages consumed over a 3-day period for the purpose of studying energy intake, nutritional de®cits or excess. This data was carefully cross-checked against changes in household food inventory. Despite the detailed quality of this data, information on water intake as a beverage may be lacking since this non-energy yielding nutrient was not a focus of this survey. The only beverages included in the survey were coconut juice, sugarcane juice, milk and alcoholic beverages. The water intake variable used for these analyses is more representative of food water intake than of total water intake. According to results from Rolls et al (1999), the available variable may be the biologically relevant exposure, however. These authors found that consuming foods with a high water content more effectively reduced subsequent energy intake than drinking water with food. Although a preload of soup decreased subsequent energy intake, a casserole containing the same ingredients (type and amount) that was served with water did not affect satiety. Cox and Mela (2000) also report that energy density calculated from all food, milk and alcohol (the same components as the variable in this study), excluding other non-alcoholic beverages was signi®cantly associated with obesity. Energy density appears to in¯uence the regulation of energy intake and weight control even under covert conditions when portion size cues or learned behaviors are circumvented (Spiegel, 1973; Garrow et al, 1978; Stubbs et al, 1998; Stubbs et al, 1995a,b; Lissner et al, 1987; Kendall et al, 1991). The underlying biological mechanism must therefore involve regulatory processes. Although speculative, the tight regulation of water intake, hydration status, and water metabolism may be involved in regulating dietary intake. Unlike fat intake, which has been shown to be poorly regulated (see Prentice & Poppitt, 1996), water intake, hydration status and water metabolism are tightly controlled at several levels by osmoregulatory mechanisms, the renin ± angiotensin system, intracellular signaling cascades, and gene transcription (Yancey et al, 1982; McManus & Churchwell, 1994; Haussinger, 1996; Burg et al, 1996). That water metabolism is a function of osmolytes may explain why water incorporated into foods=solution has a different effect from water intake alone. Future studies might examine how energy density and energy intake vary with vasopressin levels. In conclusion, this observational study explored the effects of nutrient correlates of energy density on energy intake and overweight status among Chinese adults. Of the nutrients examined, only dietary water intake appeared to


explain the effects of energy density on energy intake and overweight status. The generalizability of these ®ndings to other populations may be limited given the relatively high carbohydrate and low fat intakes prevalent in China. Longitudinal studies are needed to con®rm the associations observed in this cross-sectional study.

Acknowledgements ÐFunding for parts of the project design, data collection and computerization has been provided by the Chinese Academy of Preventive Medicine (CAPM), the Carolina Population Center (CPC) of the University of North Carolina at Chapel Hill (UNC-CH), and the National Institutes of Health (NIH) (R01-HD30880 and R01-HD38700). Funds for the research reported in this article were provided by NIH. This article is part of a collaborative research project between the CAPM, directed by Zhai Fengying and a group from UNC-CH and CPC, directed by Barry M Popkin. Special thanks go to Barry Popkin for his encouragement and support.

References Bray GA & Popkin BM (1998): Dietary fat intake does affect obesity! Am. J. Clin. Nutr. 68, 1157 ± 1173. Burg MB, Kwon ED & Kultz D (1996): Osmotic regulation of gene expression. F.A.S.E.B. J. 10, 1598 ± 1606. Cox DN & Mela DJ (2000): Determination of energy density of freely selected diets: methodological issues and implications. Int. J. Obes. Relat. Metab. Disord. 24, 49 ± 54. Garrow JS, Durrant ML, Mann S, Stalley SF & Warwick PM (1978): Factors determining weight loss in obese patients in a metabolic ward. Int. J. Obes. 2, 441 ± 447. Guo X, Popkin BM, Mroz TA & Zhai F (1999): Food price policy can signi®cantly reduce the negative effects of the nutrition transition in China. J. Nutr. 129, 994 ± 1001. Haussinger D (1996): Regulation of cell function by level of hydration. Naturwissenschaften 83, 264 ± 271. Hunt JN (1980): A possible relation between the regulation of gastric emptying and food intake. Am. J. Physiol. 239, G1 ± G4. INFH-CAPM (Institute of Nutrition and Food Hygiene, Chinese Academy of Preventive Medicine) (1991): The Food Consumption Tables. People's Medical Publication House. Kendall A, Levitsky DA, Strupp BJ & Lissner L (1991): Weight loss on a low-fat diet: consequence of the imprecision of the control of food intake in humans. Am. J. Clin. Nutr. 53, 1124 ± 1129. Lissner L, Levitsky DA, Strupp BJ, Kalkwarf HJ & Roe DA (1987): Dietary fat and the regulation of energy intake in human subjects. Am. J. Clin. Nutr. 46, 886 ± 892.

Marti-Henneberg C, Capdevila F, Arija V, Perez S, Cuco G, Vizmanos B & Fernandez-Ballart J (1999): Energy density of the diet, food volume and energy intake by age and sex in a healthy population. Eur. J. Clin. Nutr. 53, 421 ± 428. McManus ML & Churchwell KB (1994): Clinical signi®cance of cellular osmoregulation. In Cellular and Molecular Physiology of Cell Volume Regulation, ed. K Strange, pp 63 ± 77. Boca Raton, FL: CRC Press. Paeratakul S, Popkin BM, Ge K, Adair LS & Stevens J (1998): Changes in diet and physical activity affect the body mass index of Chinese adults. Int. J. Obes. Relat. Metab. Disord. 22, 424 ± 432. Popkin BM, Paeratakul S, Zhai F & Ge K (1995): Dietary and environmental correlates of obesity in a population study in China. Obes. Res. 3(Suppl), 135s ± 143s. Poppitt SD (1995): Energy density of diets and obesity. Int. J. Obes. Relat. Metab. Disord. 19(Suppl 5), S20 ± S26. Poppitt SD & Prentice AM (1996): Energy density and its role in the control of food intake: evidence from metabolic and community studies. Appetite 26, 153 ± 174. Prentice AM & Poppitt SD (1996): Importance of energy density and macronutrients in the regulation of energy intake. Int. J. Obes. Relat. Metab. Disord. 20(Suppl 2), S18 ± S23. Rolls BJ, Bell EA & Thorwart ML (1999): Water incorporated into a food but not served with a food decreases energy intake in lean women. Am. J. Clin. Nutr. 70, 448 ± 455. Rothman KJ (1986): Modern Epidemiology. Boston, MA: Little, Brown and Company. Spiegel TA (1973): Caloric regulation of food intake in man. J. Comp. Physiol. Psychol. 84, 24 ± 37. StataCorp (1999): Stata Statistical Software: Release 6.0. College Station, TX: Stata Corporation. Stubbs RJ, Harbron CG, Murgatroyd PR & Prentice AM (1995a): Covert manipulation of dietary fat and energy density: effect on substrate ¯ux and food intake in men eating ad libitum. Am. J. Clin. Nutr. 62, 316 ± 329. Stubbs RJ, Ritz P, Coward WA & Prentice AM (1995b): Covert manipulation of the ratio of dietary fat to carbohydrate and energy density: effect on food intake and energy balance in free-living men eating ad libitum. Am. J. Clin. Nutr. 62, 330 ± 337. Stubbs RJ, Johnstone AM, Harbron CG & Reid C (1998): Covert manipulation of energy density of high carbohydrate diets in `psuedo free-living' humans. Int. J. Obes. Relat. Metab. Disord. 22, 885 ± 892. WHO Expert Committee (1995): Physical Status: the Use and Interpretation of Anthropometry. WHO Technical Report Series 854. Geneva: World Health Organization. Willett WC (1998): Is dietary fat a major determinant of body fat? Am. J. Clin. Nutr. 67(Suppl) 556S ± 562S. Willett W & Stampfer M (1998a): Nature of variation in diet. In Nutritional Epidemiology, 2nd edn, ed. W Willett, pp 33 ± 49. New York: Oxford University Press. Willett W & Stampfer M (1998b): Implications of total energy intake for epidemiologic analyses. In Nutritional Epidemiology, 2nd edn, ed. W Willett, pp 273 ± 301. New York: Oxford University Press. Yancey PH, Clark ME, Steven C, Hand R, Bowlus D & Somero GN (1982): Living with water stress: evolution of osmolyte systems. Science 217, 1214 ± 1222.

359
