A solid body of evidence exists from experimental investigations and intervention studies for the association of dietary energy density (DED) with weight gain, increased energy intake (EI) and obesity( Reference Yao and Roberts 1 – Reference Ledikwe, Rolls and Smiciklas-Wright 3 ). Body weight status( Reference Vernarelli, Mitchell and Hartman 4 ), BMI( Reference Howarth, Murphy and Wilkens 5 ), waist circumference( Reference Murakami, Sasaki and Takahashi 6 ), obesity and the metabolic syndrome( Reference Mendoza, Drewnowski and Christakis 7 ) have all been found to be positively associated with DED in cross-sectional studies, providing evidence that the observations extend to free-living populations. A relationship between DED and the nutritional quality of the diet has also been highlighted; lower DED has been reported to be associated with higher dietary quality in Irish children and teenagers, American adults, Spanish adults and elderly Spanish adults( Reference Ledikwe, Blanck and Khan 8 – Reference Schroder, Vila and Marrugat 11 ) and with intakes of foods and nutrients related to better dietary quality in Mediterranean, Australian and American adults( Reference Cuco, Arija and Marti-Henneberg 12 – Reference Rolls and Bell 15 ). DED has also been proposed as a potential proxy for dietary quality( Reference Patterson, Warnberg and Poortvliet 16 ), and low DED has been reported to be associated with a trend towards a healthier lifestyle, including lower alcohol consumption, smoking and physical inactivity, in a representative sample of Spanish adults( Reference Schroder, Covas and Elosua 10 ).
To date, there have been no estimations of DED in an Irish population. As DED is associated with dietary quality and with obesity, it is worth estimating DED values for the population. This will allow tracking of secular trends and enable comparison with those of other countries. It is also worth distinguishing the dietary components that influence DED in free-living populations to explain differences in DED estimates and to ascertain potential targets for public health messages to lower DED estimates. The National Children's Food Survey (NCFS), the National Teens’ Food Survey (NTFS) and the National Adult Nutrition Survey (NANS) together form a nationally representative sample of the Irish population aged 5–90 years that provides fully disaggregated habitual food consumption data at the level of the individual from which calculation of such estimates is possible.
The aims of the present study were to calculate and present DED estimates for a nationally representative sample of the Irish population and to identify dietary determinants of DED in a free-living population.
Methods
Study sample and dietary intake assessment
The NCFS (2003–4), the NTFS (2005–6) and the NANS (2008–10) were carried out to establish databases of habitual food and drink consumption in representative samples of Irish children, adolescents and adults aged 5–12, 13–17 and 18–90 years, respectively. A 7 d weighed food record was used to collect dietary intake data from 594 children (293 boys and 301 girls), a 7 d semi-weighed food record was used to collect dietary intake data from 441 adolescents (224 boys and 217 girls), and a 4 d semi-weighed food record was used to collect dietary intake data from 1500 adults (760 females and 740 males). Analyses of dietary intake data were carried out using WISP© (Tinuviel Software), which contains McCance and Widdowson's The Composition of Foods, 6th edition( 17 ), 5th edition( Reference Holland, Welch and Unwin 18 ) plus supplemental volumes( Reference Holland, Brown and Buss 19 – Reference Chan, Brown and Lee 28 ) and the Irish Food Composition Database( Reference Black, Ireland and Moller 29 ). Ethical approval was obtained from St James’ Hospital and Federated Dublin Voluntary Hospitals Joint Research Ethics Committee for the NCFS and the University College Cork Clinical Research Ethics Committee of the Cork Teaching Hospitals for the NTFS and the NANS. Written consent was obtained from the participants or from the parents/guardians for participants aged < 18 years. A full description of methodologies is given elsewhere (http://www.iuna.net).
Estimation of dietary energy density
DED was estimated as kJ/g and was calculated using three different methods by taking the following into account: (1) energy and weight of the total diet; (2) energy and weight including only foods and excluding all beverages (i.e. all solid foods and liquid-like foods such as yogurts and soups included and all beverages excluded); (3) energy and weight of all foods and solids in beverages (values for solids in beverages were calculated by removing the water portion of each food code).
Estimation of dietary energy density accounting for under-reporters for energy
DED was estimated both including and excluding under-reporters for energy (UR). For each child and teenager, BMR was predicted from standard equations using body weight and height( Reference Schofield 30 ). Minimum EI cut-off points, calculated as multiples of BMR( Reference Torun, Davies and Livingstone 31 ), were used to identify UR. In the study sample, 32 % of children and 64 % of teenagers were classified as UR.
In adults, body composition was measured using the Tanita body composition analyser BC-420MA, which calculated BMR through multiple regression analysis using fat-free mass values (Tanita Limited). EI is expressed as a ratio of BMR (EI:BMR). The cut-off limits proposed by Goldberg et al. ( Reference Goldberg, Black and Jebb 32 ) to classify individuals as UR were used. The cut-off point relevant to the NANS was 1·10; 30 % of the participants had an EI:BMRest below this value.
Statistical analyses
Statistical analyses were conducted using SPSS© version 15.0 for Windows™ (SPSS, Inc.). DED was estimated and expressed as mean, median and standard deviation by sex and age group. An independent-samples t test and a two-way between-groups ANOVA were conducted to evaluate differences in DED estimates by sex and age group.
The dietary determinants of DED were identified by categorising the participants by tertile of energy, macronutrient and food group intakes into low-, medium- and high- or non-, low- and high-consumer groups of each. The participants were categorised into consumers and non-consumers for some less consumed food groups. A one-way ANOVA or an independent-samples t test was used to test for significant differences (P< 0·05) in means across consumer groups. Significant differences between the groups were determined using ANOVA followed by Tukey's honestly significant difference or Hochberg's GT2 post hoc test where appropriate.
Post hoc tests were chosen depending on similarity between sample sizes and homogeneity of variance between the groups (as determined by Levene's test). The effect size of the significant differences between consumer groups was calculated as follows: η2= sum of squares between groups/total sum of squares (values ≥ 0·01 indicate a small effect, ≥ 0·06 a medium effect and ≥ 0·14 a large effect, as classified by Cohen( Reference Cohen 33 )).
Food intake patterns that may explain the difference in DED estimates between sexes and with age were estimated by examining food group intakes across age groups in males and females separately. For this purpose, the participants were categorised into three age groups: 5–18; 19–50; 51–90 years. ANOVA was used to test for significant differences (P< 0·05) in mean intakes across age groups. Significant differences between the groups were determined using ANOVA followed by either Tukey's honestly significant difference or Hochberg's GT2 post hoc test. A one-way between-groups ANCOVA was used to test for significant differences between sexes adjusted for age.
Secular trends and generational effects
The DED estimates of a subsample from the NANS (those aged 18–64 years, n 1274) were compared with estimates calculated for the participants of the North/South Ireland Food Consumption Survey (NSIFCS), 2001 (n 1379). The NSIFCS was carried out 10 years before the NANS and included a representative sample of adults aged 18–64 years living in Ireland. Dietary intake data were collected using a 7 d food record. A full description of the survey design and methodology of the NSIFCS has been published previously( Reference Harrington, Robson and Kiely 34 ). Differences in DED estimates calculated for the NANS and NSIFCS were examined using a one-way between-groups ANCOVA. This comparison also allowed to assess the possibility of a generational effect on trends observed between DED and age groups in the cross-sectional analysis.
Results
For each of the three methods, DED estimates were inversely associated with age group (P linear trend < 0·001) and females had lower mean DED estimates than males (P <0·001) (Table 1). DED estimates were lower for females in all age groups (P <0·001).
Table 1 Dietary energy density (DED; kJ/g) estimates for the Irish population, aged 5–90 years, by age and sex (n 2535) (Number of participants and medians; mean values and standard deviations)
* Mean value was significantly different from that of males as calculated using an independent-samples t test (P< 0·05).
† P linear trend across age groups as calculated using two-way ANOVA.
Mean DED estimates calculated excluding UR (see online supplementary Table S1) were similar to those calculated for the total population and trended similarly by sex and across age groups. Exclusion of UR resulted in estimates similar to those calculated for the total population regardless of the calculation method used.
Determinants of DED did not vary by the calculation method. Determinants are thus only presented for DED calculated by including energy and weight of all foods and solids in beverages.
Higher mean DED estimates were found to be associated with higher intake of energy (kJ) and percentage of total energy (%TE) from fat, carbohydrates and total sugar (P <0·001) and lower intakes of protein and dietary fibre (P <0·001) (Table 2). Higher consumption or being a consumer of certain foods was found to be both positively and inversely associated with DED. DED estimates were significantly lower in those with higher intakes (g/10 MJ) of vegetables, fruit, potatoes (boiled, baked or mashed), fresh meat, brown bread, fish, vegetable and pulse dishes, eggs, pulses, soup, cooked breakfast cereals and nuts (P <0·001). Mean DED estimates were significantly higher in those with higher intakes (g/10 MJ) of white bread, chips, ready-to-eat breakfast cereals (RTEBC), processed meat, sugar-sweetened beverages (SSB), savoury snacks, chocolate and non-chocolate confectionery, savouries, alcohol and potato products (P <0·001). Variation in the intakes of vegetables, fruit and SSB (g/10 MJ) was found to be associated with the largest differences in mean DED estimates between consumer groups (η2: vegetables 0·32, fruit 0·21 and SSB 0·20). Differences in the intakes (g/10 MJ) of potatoes, fresh meat, brown bread, bananas, fish, white bread, chips, RTEBC, and chocolate and non-chocolate confectionery also contributed to the differences in DED estimates, but to a lesser extent (η2: 0·06–0·13).
Table 2 Dietary energy density (DED) estimates by tertile of mean daily energy intake (kJ), macronutrient intakes (percentage of total energy, %TE), and dietary fibre and food group intakes (g/10 MJ/d) in the Irish population (n 2535) (Number of participants, mean values and standard deviations)
↑ and ↓ , the direction of change in DED as the dietary intakes increase from low to high; 
, a non-directional association of DED with dietary intakes; –, no significant difference in DED with change in dietary intake.
a,b,cMean values with unlike superscript letters were significantly different between the consumer groups (P< 0·05).
* As calculated using a one-way ANOVA with Tukey's honestly significant difference or Hochberg's GT2 post hoc, or an independent-samples t test.
To explain the lower DED estimates obtained for females and the inverse association of DED with age, energy-adjusted intake patterns of macronutrients and dietary fibre and of the food groups that contributed to the largest variation in mean DED estimates were examined across age groups, split by sex (Table 3), and between sexes, adjusted for age (Table 4). Higher intakes of protein, total sugar (%TE) and dietary fibre (g/10 MJ) and lower intakes of carbohydrates (%TE) were found to be associated with older age groups (P <0·001). The intakes (g/10 MJ) of vegetables, fruit, fish, potatoes, fresh meat and brown bread were significantly higher (P <0·001) in older age groups in males and females, while the intakes (g/10 MJ) of SSB, chocolate and non-chocolate confectionery, RTEBC, savoury snacks, chips and white bread were significantly lower (P <0·001). The intakes of fat (P =0·007), carbohydrates, total sugar (%TE) and dietary fibre (g/10 MJ) (P <0·001) were higher in females than in males, while those of energy (kJ) (P <0·001) and protein (%TE) (P= 0·003) were lower. The intakes of vegetables, fruit, chocolate confectionery and savoury snacks were significantly higher in females than in males (P <0·001) and those of potatoes (P =0·020), RTEBC (P =0·001) and white bread (P <0·001) were significantly higher in males than in females.
Table 3 Dietary energy density (DED) estimates and mean daily energy intake (kJ), macronutrient intakes (percentage of total energy, %TE), and dietary fibre and food group intakes (g/10 MJ) in the Irish population by age group split by sex (n 2535) (Mean values and standard deviations)
↑ and ↓ , the direction of change in DED as the nutrient or food group intakes increase from low to high; 
, a non-directional association of DED with nutrient or food group intakes; –, no significant difference in DED with change in dietary intake.
a,b,cMean values with unlike superscript letters were significantly between the age groups within a sex (P< 0·05).
* As calculated using a one-way ANOVA.
Table 4 Dietary energy density (DED) estimates and mean daily energy intake (kJ), macronutrient intake (percentage of total energy, %TE), and dietary fibre and food group intakes (g/10 MJ) in the Irish population by sex adjusted for age (n 2535) (Mean values with their standard errors)
* As calculated using a one-way ANCOVA.
The DED (calculated including energy and weight of all foods and solids in beverages) estimates of the NSIFCS participants were also inversely associated with age group (P <0·001) and females had lower DED estimates than males (P <0·001). Mean DED was 8·50 (sd 1·55) kJ/g for males and 7·10 (sd 1·59) kJ/g for females. Comparison of DED estimates of the sample from the NSIFCS and the comparable subsample from the NANS, split by age and sex, is shown in Fig. 1. The mean DED estimate obtained for women, adjusted for age, in the NSIFCS did not differ significantly from that obtained for the subsample in the NANS (P =0·427). However, DED estimates obtained for men in the NSIFCS were significantly lower than those obtained for men in the NANS (P <0·001); over the 10 years, differences were apparent in only two age groups: 41–45 (P =0·015) and 56–60 (P =0·019) years.
Fig. 1 Comparison of dietary energy density (DED) estimates (kJ/g) calculated including solids in beverages for the Irish adult population aged 18–64 years, by age and sex ((a) females and (b) males) in the North/South Ireland Food Consumption Survey (
) and the National Adult Nutrition Survey (
). Error bars: ± 2 sd.
Discussion
In the nationally representative free-living Irish population aged 5–90 years included in the present study, DED estimates were higher for males than for females and were inversely associated with age group. Little difference was observed in DED estimates calculated for adults between the two surveys carried out 10 years apart. The intakes of fruit, vegetables and SSB were the major dietary determinants of DED in this population.
Many methods have been proposed for the calculation of DED and there is little consensus as to which is the most appropriate method for free-living populations( Reference Cox and Mela 35 – Reference Ledikwe, Blanck and Khan 37 ). Part of the difficulty lies in the high dependency of DED on the amount of water in the diet. The treatment of beverages is therefore important, as the inclusion of beverages has a lowering effect on the resulting DED estimates( Reference Ledikwe, Blanck and Khan 37 ). Variation in the associations of dietary and anthropometric factors with DED, depending on the method of computation, has been acknowledged( Reference O'Connor, Walton and Flynn 9 , Reference Kant and Graubard 14 , Reference Cox and Mela 35 ). Johnson et al. ( Reference Johnson, Wilks and Lindroos 38 ) proposed calculating DED by excluding drinks and including energy contributed by drinks as a covariate in analyses as the most reliable method for testing the relationship between DED and weight gain in free-living humans. The energy derived from beverages may be an important factor in the causation of obesity, in particular, from SSB, which have been reported to be independently associated with obesity( Reference Malik, Pan and Willett 39 ). In the present study, nationally representative estimates of DED calculated using three different calculation methods were evaluated. Although the DED estimates differed, the determinants and age–sex trends did not. As such, the primary analysis was focused on DED estimates calculated using a novel method: including energy and weight of all foods and solids in beverages. This approach aims to remove the bias that including beverages can introduce in DED estimates while taking into account the contribution of beverages to EI and as such has potential as a standardised method for further publications.
There was little difference in mean DED estimates on including and excluding UR. Trends in DED estimates across sexes and age groups were constant irrespective of accuracy in reporting energy. Other studies have also reported energy under-reporting as having no substantial effects on DED estimates( Reference Schroder, Covas and Elosua 10 , Reference Patterson, Warnberg and Poortvliet 16 ). Thus, the trends in DED estimates with population demographics and associations of dietary intake with DED were investigated by including all participants.
It is difficult to compare these estimates with published estimates from other populations as the methods used for calculating DED vary considerably, as discussed above. This further highlights the need for a standardised method. The DED estimates calculated for the Irish population using the food only method were found to be remarkably similar to those calculated for a sample from Northern Ireland( Reference McCaffrey, Rennie and Kerr 40 ), higher than those for a sample from Spain( Reference Schroder, Covas and Elosua 10 ) and lower than those for a sample from the USA( Reference Ledikwe, Blanck and Khan 37 ). Some consistent trends were observed across the data; females have been reported to have lower DED estimates in child and adult groups in a number of studies( Reference Mendoza, Drewnowski and Christakis 7 , Reference Schroder, Covas and Elosua 10 , Reference Kant and Graubard 14 , Reference Patterson, Warnberg and Poortvliet 16 , Reference Ledikwe, Blanck and Khan 37 ); similarly, Irish DED estimates were lower for females than for males. In keeping with the Irish data, a decrease in DED with age has been observed; Mendoza et al. ( Reference Mendoza, Drewnowski and Christakis 7 ) reported a peak in DED at the age of 7–8 years and a decline thereafter in US children, while Kant & Graubard( Reference Kant and Graubard 14 ) reported a lower mean DED in US adults aged >50 years than in those aged ≤ 50 years, and Ledikwe et al. ( Reference Ledikwe, Blanck and Khan 37 ) found that those aged ≥ 70 years had significantly lower DED estimates than other age groups.
DED is primarily determined by the water and fat contents of the diet( Reference Stubbs, Ferres and Horgan 41 ). Studies have shown a positive association of EI with DED( Reference Rolls 2 ) and associations of lower DED with higher fat, added sugar and dietary fibre intakes and lower protein and carbohydrate intakes( Reference Kant and Graubard 14 , Reference Ledikwe, Blanck and Kettel Khan 42 ). Fruit and vegetables are the food groups most commonly associated with lower DED in the literature( Reference Ledikwe, Blanck and Khan 8 , Reference Kant and Graubard 14 , Reference Patterson, Warnberg and Poortvliet 16 ). In the present study, higher intakes of energy (kJ), fat (%TE), carbohydrates (%TE) and total sugar (%TE) were found to be associated with higher DED estimates, while higher intakes of protein (%TE) and dietary fibre (g/10 MJ) were found to be associated with lower DED estimates. Variation in the intakes of fruit, vegetables and SSB had the greatest effect on mean DED estimates, followed by that in the intakes of potatoes, fresh meat, brown bread, white bread, chips, RTEBC, and chocolate and non-chocolate confectionery. Higher intakes of fruit, vegetables, potatoes, fresh meat and brown bread were associated with lower DED estimates, while higher intakes of SSB, white bread, chips, RTEBC, and chocolate and non-chocolate confectionery were associated with higher DED estimates.
In an attempt to explain the lower DED estimates obtained for females and the inverse association of DED estimates with age group, energy-adjusted intake patterns of nutrients and the food groups that explained large amounts of variance in DED estimates in the population as a whole were examined across sexes and age groups. As DED decreased across age groups, intakes of all dietary components examined that were associated with higher DED estimates in the total population decreased and those associated with lower DED estimates increased. In particular, the intakes of fruit and vegetables increased and those of RTEBC decreased considerably across age groups, with the intakes of vegetables in 5- to 18-year-olds more than doubling when compared with those in the 51–90 years age group and those of RTEBC decreasing 3-fold. The disparity in DED estimates between sexes is less clearly explained by differences in consumption patterns. Females consumed significantly more fat, carbohydrates, total sugar, dietary fibre, fruit, vegetables, chocolate confectionery and savoury snacks and less energy, protein, potatoes and RTEBC when compared with males. Of particular interest is that females consumed 1·5 times more fruit and vegetables combined when compared with males.
The downward trend in DED estimates with age group should not be interpreted as longitudinal variations in DED, as the present analysis is based on cross-sectional data and generational effects are possible. However, to examine the potential generational effect, estimates calculated for adults from the NSIFCS were compared with those calculated for a subsample, comparable for age and sex, of adults from the NANS. Little difference was observed in Irish DED estimates over time, adding credibility to the presence of an inverse association between DED and age group.
In conclusion, differences in food choices, in particular, fruit, vegetables and RTEBC, explain the variation in DED estimates by age group and sex. The intakes of fruit, vegetables and SSB appear to have the greatest influence on DED in the Irish population and are potential targets for public health interventions. DED estimates calculated for Irish adults have not changed appreciably in the last 10 years. The DED estimates calculated in the present study are the first nationally representative estimates for an Irish population and are reported by sex and age group to serve as a baseline for future studies and public health interventions.
Supplementary material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S0007114514003420
Acknowledgements
The authors thank the participants and their families for taking part in the surveys.
The present study was supported by the Irish Government under the National Development Plan (2000–6) and the Department of Agriculture, Fisheries & Food under the Food for Health Research Initiative (2007–12).
The authors’ contributions are as follows: L. O'C carried out the analysis, drafted the manuscript and is responsible for the integrity of the data; J. W. is the co-ordinator of the national nutrition surveys and contributed to the manuscript content; A. F. is a principal investigator of the national nutrition surveys and provided critical input for the analysis. All authors critically reviewed the manuscript and approved the final version submitted for publication.
None of the authors has any conflicts of interest to declare.
