Journal of Human Nutrition and Dietetics
PUBLIC HEALTH NUTRITION AND EPIDEMIOLOGY Choosing the best method to estimate the energy density of a population using food purchase data W. L. Wrieden,1 J. Armstrong,2 A. S. Anderson,3 A. Sherriff4 & K. L. Barton3 1
School of Pharmacy and Life Sciences, Robert Gordon University, Aberdeen, UK School of Health and Life Sciences, Glasgow Caledonian University, Glasgow, UK 3 Centre for Public Health Nutrition Research, University of Dundee, Dundee, UK 4 University of Glasgow Dental School, Glasgow, UK 2
Keywords energy density, food consumption, household purchase data, nutrient intake, obesity. Correspondence W. L. Wrieden, School of Pharmacy & Life Sciences, Robert Gordon University, Riverside East (N548), Garthdee Road, Aberdeen AB10 7GJ, UK. Tel.: +44 (0) 1224 262856 E-mail: [email protected] How to cite this article Wrieden W.L., Armstrong J., Anderson A.S., Sherriff A. & Barton K.L. (2015) Choosing the best method to estimate the energy density of a population using food purchase data. J Hum Nutr Diet. 28, 126–134 doi:10.1111/jhn.12227
Abstract Background: Energy density (ED) is a measure of the energy content of a food component or diet relative to a standard unit of weight. Widespread variation in ED assessment methodologies exist. The present study aimed to explore the feasibility of calculating the ED of the Scottish diet using UK food purchase survey data and to identify the most appropriate method for calculating ED for use in the development of a Scottish Dietary Goal that captures any socioeconomic differences. Methods: Energy density was calculated using five different methods [food; food and milk; food, milk and energy containing (non-alcoholic) beverages; food, milk and all non-alcoholic beverages; and all food and beverages]. ED of the Scottish diet was estimated for each of the ED methods and data were examined by deprivation category. Results: Mean ED varied from 409 to 847 kJ 100 g–1 depending on the method used. ED values calculated from food (847 kJ 100 g–1) and food and milk (718 kJ 100 g–1) were most comparable to other published data, with the latter being a more accurate reflection of all food consumed. For these two methods, there was a significant gradient between the most and least deprived quintiles (892–807 and 737–696 kJ 100 g–1 for food and food and milk, respectively). Conclusions: Because the World Cancer Research Fund recommendations are based on ED from food and milk, it was considered prudent to use this method for policy purposes and for future monitoring work of the Scottish Diet to ensure consistency of reporting and comparability with other published studies.
Introduction Energy density (ED) is a measure of the energy content of a food component or diet relative to a standard unit of weight. As such, it is significantly affected by the composition of the food or diet, notably the water, fibre and macronutrient content. ED is a particularly valuable measure because it can provide a single composite measure for diet, which is indicative of the type of foods that make up a diet, and thus provide a measure of one aspect of the quality of the diet that is particularly useful in the 126
context of obesity. ED is defined by the World Cancer Research Fund (WCRF; World Cancer Research Fund/ American Institute for Cancer Research, 2007) as ‘the amount of energy per unit weight of foods or diets. The units of measure are kilojoules (kJ) or kilocalories (kcal) per 100 g (g–1)’. Although this definition for ED is widely accepted, the method by which ED is calculated and reported varies. The increasing interest in using ED as an indicator of the quality of the diet in relation to the risk of obesity (Johnson et al., 2009; Vergnaud et al., 2009; Perez-Escamilla et al., 2012) and chronic disease is ª 2014 The British Dietetic Association Ltd.
W. L. Wrieden et al.
hampered by the lack of clarity over the best method to use. In 2007, the WCRF (World Cancer Research Fund/ American Institute for Cancer Research, 2007) formulated a public health goal stating that the average ED of the overall diet be reduced towards 523 kJ 100 g–1 (125 kcal 100 g–1). This recommendation was based on ED calculated from food and milk as described by Prentice & Jebb (2003), although a number of other methods have been reported with a wide range of values for dietary ED (Cox & Mela, 2000; Kant & Graubard, 2005; Ledikwe et al., 2005; Hartline-Grafton et al., 2009). Clearly, there is a need for a priori decision to be made on the most appropriate method to use so that relevant values can be attained and compared between population groups (Johnson et al., 2009). One of the most widespread uses of dietary ED, and a factor in considering validity of the measure, is in relation to body mass index (BMI). Some studies have found positive associations between ED and BMI (Martı-Henneberg et al., 1999; Cox & Mela, 2000; Stookey, 2001; Kant & Graubard, 2005; Ledikwe et al., 2005; Greene et al., 2006; Iqbal et al., 2006; Bes-Rastrollo et al., 2008; Savage et al., 2008; Hartline-Grafton et al., 2009; Vernarelli et al., 2011); however, this may depend on the method used to calculate ED. Kant & Graubard (2005) compared three methods used to calculate ED. They found that ED based on food only and food and energy containing beverages were modest predictors of BMI. Cox & Mela (2000) compared six methods and found that obese subjects had diets with a significantly higher ED than lean subjects when ED was based on food and milk. Conversely, they did not find the same relationship in this sample when ED was based on food alone. Hartline-Grafton et al. (2009) reported that ED from food was related to BMI. However, their definition of food included liquids consumed as foods (e.g. smoothies and milk on cereal). They found no association between ED from beverages alone and BMI, thus illustrating the importance in clarifying the exact method used. Ledikwe et al. (2005) compared eight methods and reported that men had a higher dietary ED than women and that ED decreased with age. It was suggested this may be a result of an under-reporting of high ED foods and/or over-reporting of low ED foods (e.g. fruit and vegetables) by women and older individuals. Ledikwe et al. (2005) also found differences in ED of the diets by race/ethnicity for most calculation methods, which they suggest may be partly a result of different cultural norms regarding alcohol. Associations have also been found between income and the ED of the diet (Nichele et al., 2005; Kant & Graubard, 2007; Ricciuto & Tarasuk, 2007) and this is unsurprising considering the well established inverse relationship between ED and energy cost (Drewnowski et al., 2004; Drewnowski & ª 2014 The British Dietetic Association Ltd.
Estimating energy density from purchase data
Darmon, 2005; Maillot et al., 2007; Monsivais & Drewnowski, 2007; Waterlander et al., 2010). Many of the aforementioned ED calculations were performed on individual diet records collected by 24-h dietary recall (Kant & Graubard, 2005; Ledikwe et al., 2005; Hartline-Grafton et al., 2009) or weighed food diaries (Cox & Mela, 2000; Prentice & Jebb, 2003), which include detailed information on the type and weight of foods consumed and the mode of consumption (e.g. milk on cereal or milk as a drink). The primary purpose of the present study to identify the best method of estimating the ED of the Scottish Diet. There are no surveys based on individual diet records in the UK with a sufficiently sized Scottish sample. However, studies by Nichele et al. (2005) and Ricciuto & Tarasuk (2007) have successfully estimated ED using food purchase data in France and Canada, respectively. Wrieden et al. (2006) reviewed surveys that could be used to monitor the diets of the Scottish population on an annual basis for a Working Group on Monitoring Scottish Dietary Targets, concluding that the UK Expenditure and Food Survey (EFS) [renamed in 2008 the UK Living Costs and Food (LCF)] was the only survey with a sufficient sample size to monitor dietary guidelines over time in the Scottish population. It was recognised, by the Food Standards Agency Scotland and the Scottish Government, that it was important to have dietary ED as part of the monitoring process. Therefore, the present study explored the feasibility of estimating the ED of the Scottish diet using EFS/LCF food purchase data. The EFS/LCF food purchase data are currently the primary data source for monitoring progress towards the Scottish Dietary Targets (Wrieden et al., 2013). The EFS/ LCF is a continuous household (HH) purchase survey (of HHs selected randomly on an annual basis) in the UK commissioned jointly by the Office for National Statistics (ONS) and the Department for Environment and Rural Affairs (Defra). As part of the survey, individuals within each HH complete a detailed 14-day diary of all food and beverages purchased for consumption both in and out of the home from which mean food and nutrient consumption per person can be derived. The diaries provide valuable data on average population intakes, appropriate for population level goals, for specific food groups and nutrients. Using this dietary information, a robust methodology has been developed using the EFS/LCF data to monitor food and nutrient intakes in Scotland over the period 2001 to 2009 and this work continues (Wrieden et al., 2013). An important aspect of this work is determining trends in socioeconomic differences in the diet to inform research and policy in relation to health inequalities. The present study aimed to: (i) to explore the feasibility of calculating the ED of the Scottish diet using UK 127
W. L. Wrieden et al.
Estimating energy density from purchase data
food purchase survey data and (ii) to identify the most appropriate method for calculating ED for use in the development of a Scottish Dietary Goal that will capture any socioeconomic differences. Materials and methods The present study used food purchase data from the Scottish sample of the UK EFS (2001–2007) and the LCF [2008–2009; Department of the Environment & Rural Affairs (Defra), 2012] to estimate HH level dietary consumption data. This is a continuous but cross-sectional survey of HHs in the UK. Data for each year, in raw form, were obtained from the UK Data Archive (UK Data Archive, 2012). Variables on the sampling methodology of the EFS/LCF HHs and the variable for deprivation categories [Scottish Index of Multiple Deprivation (SIMD)] (Scottish Government, 2012) were obtained from the UK ONS (ONS, 2010) and Scottish Neighbourhood Statistics (Scottish Neighbourhood Statistics, 2012), respectively.
The published literature was reviewed to determine established methods for calculating ED. Table 1 summarises the varying methods that have been used to calculate ED, which, for example, may be calculated based on: food alone; food and only milk beverages; and food and all beverages that provide energy. The EFS/LCF data were explored to investigate the possible ways that ED of the Scottish diet could be calculated. For the present study, all liquids that could be consumed as drinks (with the exception of meal replacements) were classified as a drink (e.g. milk = drink). Previously, other studies have classified milk incorporated into sauces and puddings (e.g. as a food rather than a drink) because they have had more information on how the food was consumed (Cox & Mela, 2000; Kant & Graubard, 2005; Ledikwe et al., 2005; Hartline-Grafton et al., 2009). This meant that it was not possible to use exactly the same method as previously published studies (e.g. classifying liquids consumed as foods or added to foods such as smoothies and milk on cereal) as part of food categories (Hartline-Grafton et al., 2009). Further to reviewing the literature and
Table 1 Examples of differing food intake methods used to calculate energy density (ED) and mean values (kJ 100 g–1) Calculation method
Comment*
Food only
Excludes all beverages
Food and milk
Includes liquids consumed as foods or added to foods Excludes tea, coffee, water and soft drinks
Food and milk Food and milk Food and liquid meal replacements Food and alcohol Food and juice Food, juice and milk Food and energy containing beverages
Includes food and dairy beverages with protein 3.1 g 100 g–1 Includes all milk Includes food and beverages that replace meals. Powdered meal replacements included as reconstituted Includes all beverages containing alcohol Includes 100% fruit and vegetable juices Includes 100% fruit and vegetable juices and milk Includes beverages with at least 21kJ (5 kcal 100 g–1)
Food and energy containing beverages Food and all beverages excluding water
Excludes water, tea and coffee infusions, diet drinks
Food and all beverages Dry matter Protein, Carbohydrate and fat only
Appears to include water Excludes all water and water in composition
Mean kJ 100 g–1 †
Reference
570–590 803 774 782 600–730 525‡ 703 590–650† 774
Cox & Mela (2000) Kant & Graubard (2005) Ledikwe et al. (2005) Hartline-Grafton et al. (2009) Prentice & Jebb (2003)
732 736 674 636
Ledikwe Ledikwe Ledikwe Ledikwe
545 480–490§
Kant & Graubard (2005) Cox & Mela (2000)
393
Ledikwe et al. (2005)
384¶ 320–350† 1900–1970§ 2080–2100§
Ledikwe et al. (2005) Cox & Mela (2000) Ledikwe et al. (2005) et et et et
al. al. al. al.
(2005) (2005) (2005) (2005)
Kant & Graubard (2005) Cox & Mela (2000) Cox & Mela (2000) Cox & Mela (2000)
*Refer to individual references for exact inclusion/exclusion criteria. The lower figure is average for lean subjects [body mass index (BMI) 20–25 kg m–2] and the higher figure is for obese subjects (BMI ≥ 30 kg m–2). ‡ ED for subset of women aged 16–64 years consuming no more than 35% energy from fat and at least 400 g of fruit and vegetables per day. § Lower figure is average for obese adults (BMI ≥ 30 kg m–2) and the higher figure is for lean adults (BMI 20–25 kg m–2). ¶ Excludes tap water but not bottled water. †
128
ª 2014 The British Dietetic Association Ltd.
W. L. Wrieden et al.
Estimating energy density from purchase data
examining the EFS/LCF data, it was considered feasible to use HH and eating out purchases of foods and drinks to calculate ED using five different methods (Table 2) after appropriate adjustment for waste and weight increase or loss as a result of cooking or dilution. Coding frames were designed for each of the ED methods described in Table 2. The coding frames (available on request) indicate which foods/drinks were included within each of the ED methods and list adjustment factors that were applied to foods/drinks to estimate the actual amount consumed (e.g. to only include edible portions, to adjust for cooking/dilution, to adjust purchases for waste, etc.). Dilution factors were applied to drinks that would be re-constituted with water before consumption (i.e. tea and coffee) but not to those that would be reconstituted with milk (i.e. hot chocolate) because milk used was already included in the calculations. Other foods that may not be consumed in their purchased state (e.g. flour, stock cubes, jelly cubes) were not given a factor because it was not possible to tell how these foods may be prepared and subsequently consumed. Information for these adjustment factors was taken from McCance and Widdowson’s The Composition of Foods and its supplements (Holland et al., 1992a,b, 1993; Chan et al., 1994, 1995, 1996; Food Standards Agency, 2002) or, where these data were not available, information was sought from manufacturers’ label data. Where no factor was necessary, a factor of 1.0 was applied. Estimates of edible waste published by the Waste & Resource Action Programme Survey (WRAP) (2008), which have been mapped by Defra to each of the food codes used in the EFS/LCF, were used to assign a waste factor to each food code. These waste figures are listed for single and multiple adult HHs and were linked to the appropriate type of HH before analysis. The EDs were calculated in three stages: Calculating Weight of Food/Drink: The total weight of food/drink for each HH (by ED method) was calculated by summing the weights of each food after making
•
adjustments for waste and multiplying by the adjustment factors described previously. Calculating Energy Content of Food/Drink: The total energy from food/drink for each HH (by ED method) was calculated by summing the energy content of each food after making adjustments for waste only because the nutrient values in the database are based on the foods in their purchased form and not in the form they are consumed. Calculating ED: The ED values per 100 g (by ED method) for each HH were calculated by dividing the total HH energy content for each ED method (2) by the total HH weight for each ED method (1) and multiplying by 100. Quintiles of ED for each ED method were calculated by year (to negate any difference in ED over time). Socioeconomic position was measured using the SIMD (Scottish Government, 2012), an area based index of deprivation based on indicators within seven individual domains of Current Income, Employment, Housing, Health, Education, Skills & Training, Geographic Access to Services & Telecommunications and Crime. The SIMD ranks small areas (called datazones) from most deprived (1) to least deprived (6505), presented here in quintiles, with ‘1’ being most deprived and ‘5’ being the least deprived. In view of policy needs to monitor inequalities in health (CSDH, 2008), the method chosen for measuring ED was required to detect differences as a result of socioeconomic position. The data were analysed in the complex samples component of SPSS, version 18 (SPSS Inc., Chicago, IL, USA), which allows for the data to be weighted according to the sampling methodology (ONS, 2010; original data collected by the ONS) to make the results representative of the Scottish population. General linear modelling was used to obtain mean, 95% confidence interval (CI) and an indication of statistical significance for differences and trends. Linear associations between ED and year or SIMD quintile were assessed by linear regression. The mean ED
•
•
Table 2 Methods used to define energy density categories Energy density method*
Comment
1 Food
Includes meal replacements (as reconstituted) Excludes all other liquids consumed as drinks As method 1 but also includes all milk (as reconstituted where required) As method 2 but also includes non-alcoholic beverages with at least 21kJ (5 kcal 100 g–1) (e.g. white coffee, sugar-containing soft drinks, diluted where required) Includes all food and beverages (as diluted where required and tea and coffee as consumed weight) with the exception of alcoholic beverages and tap water Includes all food and beverages (as diluted where required and tea and coffee as consumed weight plus alcoholic beverages) with the exception of tap water
2 Food and milk 3 Food, milk and energy containing (non-alcoholic) beverages 4 Food, milk and non-alcoholic beverages 5 All food and beverages
*Foods and drinks included/excluded in each of the methods are available on request.
ª 2014 The British Dietetic Association Ltd.
129
W. L. Wrieden et al.
Estimating energy density from purchase data
(overall and by quintile) with 95% CI was calculated for the period 2001 to 2009 overall for each ED method and differences in ED were examined over time and by quintile of SIMD. Results Food purchase data from 5020 HHs (11 374 people), over the period 2001–2009, were analysed for the present study. The annual sample size of this cross-sectional survey varied from 494 to 619 HHs, with a mean of 558 HHs per year. Table 3 provides an overall estimate of the average ED for the Scottish diet (combined for the years 2001 to 2009 because there was no difference in average ED between 2001 and 2009 for any of the methods) for the five different ED methods used. Marked differences are apparent in the estimates for ED depending on which method is used, particularly the effect of including and excluding drinks. ED calculated from food alone was over double that calculated from all food and drinks. Table 3 also provides estimates by quintiles of ED for each of the different ED categories, showing that the differences between the mean ED for the highest quintile (most dense) and lowest quintile (least dense) was 535 kJ 100 g–1 when calculated from food alone, 449 kJ 100 g–1 when using food and milk; and 455 kJ 100 g–1 when using all food and beverages.
Table 4 provides estimates of ED by SIMD quintile (for 5018 HHs; SIMD was not available for 2 HHs) for each of the different ED methods. These results show that, using ED calculated from food, as well as food and milk, there is a clear social gradient in that those who live in the most deprived quintile of SIMD consume diets statistically higher in ED compared to those who live in the least deprived quintile of SIMD (difference in means was 85 and 41 kJ 100 g–1, respectively, for the most versus the least deprived). This gradient was not apparent when ED was calculated using the three methods that included beverages. Discussion Food purchase data from the EFS/LCF provided the basis for exploring how to estimate the ED for the Scottish diet. This is the first study to have assessed the feasibility of using HH purchase data to estimate ED of populations within the UK. Five different methods to calculate ED ranging from that calculated from food only to that including all foods and beverages (excluding tap water) were considered. Food purchase data obtained from the EFS/LCF were used to calculate estimates of ED for the population using data from 2001 to 2009 for each of the five ED methods. As a result of the low ED of drinks, the estimate of ED for all food and beverages was approximately half that of food alone. The overall estimate of
Table 3 Mean energy density of the diet for each method from 2001 to 2009 Expenditure and Food Survey (EFS)/Living Costs and Food household and eating out data combined (kJ 100 g–1) Energy density quintile
Energy density method Food Food and milk Food, milk and energy containing (non-alcoholic) beverages Food, milk and non-alcoholic beverages All food and beverages
Overall
1 Least dense
3
4
5 Most dense
Mean 95% CI kJ 100 g–1
Mean 95% CI kJ 100 g–1
847 (839–854) 718 (713–724) 618 (614–623)
605 (600–610) 515 (510–519) 443 (439–446)
740 (738–742) 628 (626–629) 537 (536–5380)
825 (824–827) 698 (697–700) 599 (597–600)
920 (918–922) 779 (777–781) 670 (668–672)
1140 (1128–1152) 964 (953–975) 847 (838–857)
446 (441–451)
217 (214–221)
339 (338–341)
429 (428–431)
524 (522–526)
716 (704–727)
409 (405–414)
204 (201–207)
314 (312–316)
393 (391–394)
479 (477–481)
659 (649–670)
2
From 2006, the EFS moved from a financial year to a calendar year basis. As a consequence of this, the January to March 2006 data are duplicated in the 2005/2006 and the 2006 data. Because the sample size is different for each quintile for each method, the overall sample size is provided: 5020 households, 11 374 people and 45 091 people weighted (the results are weighted to the Scottish population; the number provided is approximately 1000th of the Scottish population multiplied by 9 because 9 years of data are used in the analysis).CI, confidence interval.
130
ª 2014 The British Dietetic Association Ltd.
W. L. Wrieden et al.
Estimating energy density from purchase data
Table 4 Mean energy density of the diet for each method by Scottish Index of Multiple Deprivation (SIMD) from 2001 to 2009 Expenditure and Food Survey (EFS)/Living Costs and Food household and eating out data combined (kJ 100 g–1) SIMD quintile 1 Most deprived Energy density method Food Food and milk Food, milk and energy containing NA) beverages Food, milk and NA beverages All food and beverages Number of households Number of people† Number of people weighted
2
3
4
5 Least deprived
Mean 95% CI kJ 100 g–1
P-value*
892 (878–905) 737 (726–748) 619 (609–629)
860 (848–873) 728 (717–738) 625 (614–636)
848 (834–862) 721 (709–732) 617 (608–626)
829 (815–842) 711 (699–724) 622 (611–632)
807 (796–817) 696 (686–706) 610 (600–619)
PUBLIC HEALTH NUTRITION AND EPIDEMIOLOGY Choosing the best method to estimate the energy density of a population using food purchase data W. L. Wrieden,1 J. Armstrong,2 A. S. Anderson,3 A. Sherriff4 & K. L. Barton3 1
School of Pharmacy and Life Sciences, Robert Gordon University, Aberdeen, UK School of Health and Life Sciences, Glasgow Caledonian University, Glasgow, UK 3 Centre for Public Health Nutrition Research, University of Dundee, Dundee, UK 4 University of Glasgow Dental School, Glasgow, UK 2
Keywords energy density, food consumption, household purchase data, nutrient intake, obesity. Correspondence W. L. Wrieden, School of Pharmacy & Life Sciences, Robert Gordon University, Riverside East (N548), Garthdee Road, Aberdeen AB10 7GJ, UK. Tel.: +44 (0) 1224 262856 E-mail: [email protected] How to cite this article Wrieden W.L., Armstrong J., Anderson A.S., Sherriff A. & Barton K.L. (2015) Choosing the best method to estimate the energy density of a population using food purchase data. J Hum Nutr Diet. 28, 126–134 doi:10.1111/jhn.12227
Abstract Background: Energy density (ED) is a measure of the energy content of a food component or diet relative to a standard unit of weight. Widespread variation in ED assessment methodologies exist. The present study aimed to explore the feasibility of calculating the ED of the Scottish diet using UK food purchase survey data and to identify the most appropriate method for calculating ED for use in the development of a Scottish Dietary Goal that captures any socioeconomic differences. Methods: Energy density was calculated using five different methods [food; food and milk; food, milk and energy containing (non-alcoholic) beverages; food, milk and all non-alcoholic beverages; and all food and beverages]. ED of the Scottish diet was estimated for each of the ED methods and data were examined by deprivation category. Results: Mean ED varied from 409 to 847 kJ 100 g–1 depending on the method used. ED values calculated from food (847 kJ 100 g–1) and food and milk (718 kJ 100 g–1) were most comparable to other published data, with the latter being a more accurate reflection of all food consumed. For these two methods, there was a significant gradient between the most and least deprived quintiles (892–807 and 737–696 kJ 100 g–1 for food and food and milk, respectively). Conclusions: Because the World Cancer Research Fund recommendations are based on ED from food and milk, it was considered prudent to use this method for policy purposes and for future monitoring work of the Scottish Diet to ensure consistency of reporting and comparability with other published studies.
Introduction Energy density (ED) is a measure of the energy content of a food component or diet relative to a standard unit of weight. As such, it is significantly affected by the composition of the food or diet, notably the water, fibre and macronutrient content. ED is a particularly valuable measure because it can provide a single composite measure for diet, which is indicative of the type of foods that make up a diet, and thus provide a measure of one aspect of the quality of the diet that is particularly useful in the 126
context of obesity. ED is defined by the World Cancer Research Fund (WCRF; World Cancer Research Fund/ American Institute for Cancer Research, 2007) as ‘the amount of energy per unit weight of foods or diets. The units of measure are kilojoules (kJ) or kilocalories (kcal) per 100 g (g–1)’. Although this definition for ED is widely accepted, the method by which ED is calculated and reported varies. The increasing interest in using ED as an indicator of the quality of the diet in relation to the risk of obesity (Johnson et al., 2009; Vergnaud et al., 2009; Perez-Escamilla et al., 2012) and chronic disease is ª 2014 The British Dietetic Association Ltd.
W. L. Wrieden et al.
hampered by the lack of clarity over the best method to use. In 2007, the WCRF (World Cancer Research Fund/ American Institute for Cancer Research, 2007) formulated a public health goal stating that the average ED of the overall diet be reduced towards 523 kJ 100 g–1 (125 kcal 100 g–1). This recommendation was based on ED calculated from food and milk as described by Prentice & Jebb (2003), although a number of other methods have been reported with a wide range of values for dietary ED (Cox & Mela, 2000; Kant & Graubard, 2005; Ledikwe et al., 2005; Hartline-Grafton et al., 2009). Clearly, there is a need for a priori decision to be made on the most appropriate method to use so that relevant values can be attained and compared between population groups (Johnson et al., 2009). One of the most widespread uses of dietary ED, and a factor in considering validity of the measure, is in relation to body mass index (BMI). Some studies have found positive associations between ED and BMI (Martı-Henneberg et al., 1999; Cox & Mela, 2000; Stookey, 2001; Kant & Graubard, 2005; Ledikwe et al., 2005; Greene et al., 2006; Iqbal et al., 2006; Bes-Rastrollo et al., 2008; Savage et al., 2008; Hartline-Grafton et al., 2009; Vernarelli et al., 2011); however, this may depend on the method used to calculate ED. Kant & Graubard (2005) compared three methods used to calculate ED. They found that ED based on food only and food and energy containing beverages were modest predictors of BMI. Cox & Mela (2000) compared six methods and found that obese subjects had diets with a significantly higher ED than lean subjects when ED was based on food and milk. Conversely, they did not find the same relationship in this sample when ED was based on food alone. Hartline-Grafton et al. (2009) reported that ED from food was related to BMI. However, their definition of food included liquids consumed as foods (e.g. smoothies and milk on cereal). They found no association between ED from beverages alone and BMI, thus illustrating the importance in clarifying the exact method used. Ledikwe et al. (2005) compared eight methods and reported that men had a higher dietary ED than women and that ED decreased with age. It was suggested this may be a result of an under-reporting of high ED foods and/or over-reporting of low ED foods (e.g. fruit and vegetables) by women and older individuals. Ledikwe et al. (2005) also found differences in ED of the diets by race/ethnicity for most calculation methods, which they suggest may be partly a result of different cultural norms regarding alcohol. Associations have also been found between income and the ED of the diet (Nichele et al., 2005; Kant & Graubard, 2007; Ricciuto & Tarasuk, 2007) and this is unsurprising considering the well established inverse relationship between ED and energy cost (Drewnowski et al., 2004; Drewnowski & ª 2014 The British Dietetic Association Ltd.
Estimating energy density from purchase data
Darmon, 2005; Maillot et al., 2007; Monsivais & Drewnowski, 2007; Waterlander et al., 2010). Many of the aforementioned ED calculations were performed on individual diet records collected by 24-h dietary recall (Kant & Graubard, 2005; Ledikwe et al., 2005; Hartline-Grafton et al., 2009) or weighed food diaries (Cox & Mela, 2000; Prentice & Jebb, 2003), which include detailed information on the type and weight of foods consumed and the mode of consumption (e.g. milk on cereal or milk as a drink). The primary purpose of the present study to identify the best method of estimating the ED of the Scottish Diet. There are no surveys based on individual diet records in the UK with a sufficiently sized Scottish sample. However, studies by Nichele et al. (2005) and Ricciuto & Tarasuk (2007) have successfully estimated ED using food purchase data in France and Canada, respectively. Wrieden et al. (2006) reviewed surveys that could be used to monitor the diets of the Scottish population on an annual basis for a Working Group on Monitoring Scottish Dietary Targets, concluding that the UK Expenditure and Food Survey (EFS) [renamed in 2008 the UK Living Costs and Food (LCF)] was the only survey with a sufficient sample size to monitor dietary guidelines over time in the Scottish population. It was recognised, by the Food Standards Agency Scotland and the Scottish Government, that it was important to have dietary ED as part of the monitoring process. Therefore, the present study explored the feasibility of estimating the ED of the Scottish diet using EFS/LCF food purchase data. The EFS/LCF food purchase data are currently the primary data source for monitoring progress towards the Scottish Dietary Targets (Wrieden et al., 2013). The EFS/ LCF is a continuous household (HH) purchase survey (of HHs selected randomly on an annual basis) in the UK commissioned jointly by the Office for National Statistics (ONS) and the Department for Environment and Rural Affairs (Defra). As part of the survey, individuals within each HH complete a detailed 14-day diary of all food and beverages purchased for consumption both in and out of the home from which mean food and nutrient consumption per person can be derived. The diaries provide valuable data on average population intakes, appropriate for population level goals, for specific food groups and nutrients. Using this dietary information, a robust methodology has been developed using the EFS/LCF data to monitor food and nutrient intakes in Scotland over the period 2001 to 2009 and this work continues (Wrieden et al., 2013). An important aspect of this work is determining trends in socioeconomic differences in the diet to inform research and policy in relation to health inequalities. The present study aimed to: (i) to explore the feasibility of calculating the ED of the Scottish diet using UK 127
W. L. Wrieden et al.
Estimating energy density from purchase data
food purchase survey data and (ii) to identify the most appropriate method for calculating ED for use in the development of a Scottish Dietary Goal that will capture any socioeconomic differences. Materials and methods The present study used food purchase data from the Scottish sample of the UK EFS (2001–2007) and the LCF [2008–2009; Department of the Environment & Rural Affairs (Defra), 2012] to estimate HH level dietary consumption data. This is a continuous but cross-sectional survey of HHs in the UK. Data for each year, in raw form, were obtained from the UK Data Archive (UK Data Archive, 2012). Variables on the sampling methodology of the EFS/LCF HHs and the variable for deprivation categories [Scottish Index of Multiple Deprivation (SIMD)] (Scottish Government, 2012) were obtained from the UK ONS (ONS, 2010) and Scottish Neighbourhood Statistics (Scottish Neighbourhood Statistics, 2012), respectively.
The published literature was reviewed to determine established methods for calculating ED. Table 1 summarises the varying methods that have been used to calculate ED, which, for example, may be calculated based on: food alone; food and only milk beverages; and food and all beverages that provide energy. The EFS/LCF data were explored to investigate the possible ways that ED of the Scottish diet could be calculated. For the present study, all liquids that could be consumed as drinks (with the exception of meal replacements) were classified as a drink (e.g. milk = drink). Previously, other studies have classified milk incorporated into sauces and puddings (e.g. as a food rather than a drink) because they have had more information on how the food was consumed (Cox & Mela, 2000; Kant & Graubard, 2005; Ledikwe et al., 2005; Hartline-Grafton et al., 2009). This meant that it was not possible to use exactly the same method as previously published studies (e.g. classifying liquids consumed as foods or added to foods such as smoothies and milk on cereal) as part of food categories (Hartline-Grafton et al., 2009). Further to reviewing the literature and
Table 1 Examples of differing food intake methods used to calculate energy density (ED) and mean values (kJ 100 g–1) Calculation method
Comment*
Food only
Excludes all beverages
Food and milk
Includes liquids consumed as foods or added to foods Excludes tea, coffee, water and soft drinks
Food and milk Food and milk Food and liquid meal replacements Food and alcohol Food and juice Food, juice and milk Food and energy containing beverages
Includes food and dairy beverages with protein 3.1 g 100 g–1 Includes all milk Includes food and beverages that replace meals. Powdered meal replacements included as reconstituted Includes all beverages containing alcohol Includes 100% fruit and vegetable juices Includes 100% fruit and vegetable juices and milk Includes beverages with at least 21kJ (5 kcal 100 g–1)
Food and energy containing beverages Food and all beverages excluding water
Excludes water, tea and coffee infusions, diet drinks
Food and all beverages Dry matter Protein, Carbohydrate and fat only
Appears to include water Excludes all water and water in composition
Mean kJ 100 g–1 †
Reference
570–590 803 774 782 600–730 525‡ 703 590–650† 774
Cox & Mela (2000) Kant & Graubard (2005) Ledikwe et al. (2005) Hartline-Grafton et al. (2009) Prentice & Jebb (2003)
732 736 674 636
Ledikwe Ledikwe Ledikwe Ledikwe
545 480–490§
Kant & Graubard (2005) Cox & Mela (2000)
393
Ledikwe et al. (2005)
384¶ 320–350† 1900–1970§ 2080–2100§
Ledikwe et al. (2005) Cox & Mela (2000) Ledikwe et al. (2005) et et et et
al. al. al. al.
(2005) (2005) (2005) (2005)
Kant & Graubard (2005) Cox & Mela (2000) Cox & Mela (2000) Cox & Mela (2000)
*Refer to individual references for exact inclusion/exclusion criteria. The lower figure is average for lean subjects [body mass index (BMI) 20–25 kg m–2] and the higher figure is for obese subjects (BMI ≥ 30 kg m–2). ‡ ED for subset of women aged 16–64 years consuming no more than 35% energy from fat and at least 400 g of fruit and vegetables per day. § Lower figure is average for obese adults (BMI ≥ 30 kg m–2) and the higher figure is for lean adults (BMI 20–25 kg m–2). ¶ Excludes tap water but not bottled water. †
128
ª 2014 The British Dietetic Association Ltd.
W. L. Wrieden et al.
Estimating energy density from purchase data
examining the EFS/LCF data, it was considered feasible to use HH and eating out purchases of foods and drinks to calculate ED using five different methods (Table 2) after appropriate adjustment for waste and weight increase or loss as a result of cooking or dilution. Coding frames were designed for each of the ED methods described in Table 2. The coding frames (available on request) indicate which foods/drinks were included within each of the ED methods and list adjustment factors that were applied to foods/drinks to estimate the actual amount consumed (e.g. to only include edible portions, to adjust for cooking/dilution, to adjust purchases for waste, etc.). Dilution factors were applied to drinks that would be re-constituted with water before consumption (i.e. tea and coffee) but not to those that would be reconstituted with milk (i.e. hot chocolate) because milk used was already included in the calculations. Other foods that may not be consumed in their purchased state (e.g. flour, stock cubes, jelly cubes) were not given a factor because it was not possible to tell how these foods may be prepared and subsequently consumed. Information for these adjustment factors was taken from McCance and Widdowson’s The Composition of Foods and its supplements (Holland et al., 1992a,b, 1993; Chan et al., 1994, 1995, 1996; Food Standards Agency, 2002) or, where these data were not available, information was sought from manufacturers’ label data. Where no factor was necessary, a factor of 1.0 was applied. Estimates of edible waste published by the Waste & Resource Action Programme Survey (WRAP) (2008), which have been mapped by Defra to each of the food codes used in the EFS/LCF, were used to assign a waste factor to each food code. These waste figures are listed for single and multiple adult HHs and were linked to the appropriate type of HH before analysis. The EDs were calculated in three stages: Calculating Weight of Food/Drink: The total weight of food/drink for each HH (by ED method) was calculated by summing the weights of each food after making
•
adjustments for waste and multiplying by the adjustment factors described previously. Calculating Energy Content of Food/Drink: The total energy from food/drink for each HH (by ED method) was calculated by summing the energy content of each food after making adjustments for waste only because the nutrient values in the database are based on the foods in their purchased form and not in the form they are consumed. Calculating ED: The ED values per 100 g (by ED method) for each HH were calculated by dividing the total HH energy content for each ED method (2) by the total HH weight for each ED method (1) and multiplying by 100. Quintiles of ED for each ED method were calculated by year (to negate any difference in ED over time). Socioeconomic position was measured using the SIMD (Scottish Government, 2012), an area based index of deprivation based on indicators within seven individual domains of Current Income, Employment, Housing, Health, Education, Skills & Training, Geographic Access to Services & Telecommunications and Crime. The SIMD ranks small areas (called datazones) from most deprived (1) to least deprived (6505), presented here in quintiles, with ‘1’ being most deprived and ‘5’ being the least deprived. In view of policy needs to monitor inequalities in health (CSDH, 2008), the method chosen for measuring ED was required to detect differences as a result of socioeconomic position. The data were analysed in the complex samples component of SPSS, version 18 (SPSS Inc., Chicago, IL, USA), which allows for the data to be weighted according to the sampling methodology (ONS, 2010; original data collected by the ONS) to make the results representative of the Scottish population. General linear modelling was used to obtain mean, 95% confidence interval (CI) and an indication of statistical significance for differences and trends. Linear associations between ED and year or SIMD quintile were assessed by linear regression. The mean ED
•
•
Table 2 Methods used to define energy density categories Energy density method*
Comment
1 Food
Includes meal replacements (as reconstituted) Excludes all other liquids consumed as drinks As method 1 but also includes all milk (as reconstituted where required) As method 2 but also includes non-alcoholic beverages with at least 21kJ (5 kcal 100 g–1) (e.g. white coffee, sugar-containing soft drinks, diluted where required) Includes all food and beverages (as diluted where required and tea and coffee as consumed weight) with the exception of alcoholic beverages and tap water Includes all food and beverages (as diluted where required and tea and coffee as consumed weight plus alcoholic beverages) with the exception of tap water
2 Food and milk 3 Food, milk and energy containing (non-alcoholic) beverages 4 Food, milk and non-alcoholic beverages 5 All food and beverages
*Foods and drinks included/excluded in each of the methods are available on request.
ª 2014 The British Dietetic Association Ltd.
129
W. L. Wrieden et al.
Estimating energy density from purchase data
(overall and by quintile) with 95% CI was calculated for the period 2001 to 2009 overall for each ED method and differences in ED were examined over time and by quintile of SIMD. Results Food purchase data from 5020 HHs (11 374 people), over the period 2001–2009, were analysed for the present study. The annual sample size of this cross-sectional survey varied from 494 to 619 HHs, with a mean of 558 HHs per year. Table 3 provides an overall estimate of the average ED for the Scottish diet (combined for the years 2001 to 2009 because there was no difference in average ED between 2001 and 2009 for any of the methods) for the five different ED methods used. Marked differences are apparent in the estimates for ED depending on which method is used, particularly the effect of including and excluding drinks. ED calculated from food alone was over double that calculated from all food and drinks. Table 3 also provides estimates by quintiles of ED for each of the different ED categories, showing that the differences between the mean ED for the highest quintile (most dense) and lowest quintile (least dense) was 535 kJ 100 g–1 when calculated from food alone, 449 kJ 100 g–1 when using food and milk; and 455 kJ 100 g–1 when using all food and beverages.
Table 4 provides estimates of ED by SIMD quintile (for 5018 HHs; SIMD was not available for 2 HHs) for each of the different ED methods. These results show that, using ED calculated from food, as well as food and milk, there is a clear social gradient in that those who live in the most deprived quintile of SIMD consume diets statistically higher in ED compared to those who live in the least deprived quintile of SIMD (difference in means was 85 and 41 kJ 100 g–1, respectively, for the most versus the least deprived). This gradient was not apparent when ED was calculated using the three methods that included beverages. Discussion Food purchase data from the EFS/LCF provided the basis for exploring how to estimate the ED for the Scottish diet. This is the first study to have assessed the feasibility of using HH purchase data to estimate ED of populations within the UK. Five different methods to calculate ED ranging from that calculated from food only to that including all foods and beverages (excluding tap water) were considered. Food purchase data obtained from the EFS/LCF were used to calculate estimates of ED for the population using data from 2001 to 2009 for each of the five ED methods. As a result of the low ED of drinks, the estimate of ED for all food and beverages was approximately half that of food alone. The overall estimate of
Table 3 Mean energy density of the diet for each method from 2001 to 2009 Expenditure and Food Survey (EFS)/Living Costs and Food household and eating out data combined (kJ 100 g–1) Energy density quintile
Energy density method Food Food and milk Food, milk and energy containing (non-alcoholic) beverages Food, milk and non-alcoholic beverages All food and beverages
Overall
1 Least dense
3
4
5 Most dense
Mean 95% CI kJ 100 g–1
Mean 95% CI kJ 100 g–1
847 (839–854) 718 (713–724) 618 (614–623)
605 (600–610) 515 (510–519) 443 (439–446)
740 (738–742) 628 (626–629) 537 (536–5380)
825 (824–827) 698 (697–700) 599 (597–600)
920 (918–922) 779 (777–781) 670 (668–672)
1140 (1128–1152) 964 (953–975) 847 (838–857)
446 (441–451)
217 (214–221)
339 (338–341)
429 (428–431)
524 (522–526)
716 (704–727)
409 (405–414)
204 (201–207)
314 (312–316)
393 (391–394)
479 (477–481)
659 (649–670)
2
From 2006, the EFS moved from a financial year to a calendar year basis. As a consequence of this, the January to March 2006 data are duplicated in the 2005/2006 and the 2006 data. Because the sample size is different for each quintile for each method, the overall sample size is provided: 5020 households, 11 374 people and 45 091 people weighted (the results are weighted to the Scottish population; the number provided is approximately 1000th of the Scottish population multiplied by 9 because 9 years of data are used in the analysis).CI, confidence interval.
130
ª 2014 The British Dietetic Association Ltd.
W. L. Wrieden et al.
Estimating energy density from purchase data
Table 4 Mean energy density of the diet for each method by Scottish Index of Multiple Deprivation (SIMD) from 2001 to 2009 Expenditure and Food Survey (EFS)/Living Costs and Food household and eating out data combined (kJ 100 g–1) SIMD quintile 1 Most deprived Energy density method Food Food and milk Food, milk and energy containing NA) beverages Food, milk and NA beverages All food and beverages Number of households Number of people† Number of people weighted
2
3
4
5 Least deprived
Mean 95% CI kJ 100 g–1
P-value*
892 (878–905) 737 (726–748) 619 (609–629)
860 (848–873) 728 (717–738) 625 (614–636)
848 (834–862) 721 (709–732) 617 (608–626)
829 (815–842) 711 (699–724) 622 (611–632)
807 (796–817) 696 (686–706) 610 (600–619)
