Empirical dietary surveys frequently examine the food intake of a population in terms of nutrient intake. This approach provides valuable information on nutrient adequacy and excess, and allows testing of a priori hypotheses on the association between nutrients and morbidity. However, the approach does not allow the complexities of dietary intake of individuals, or groups of individuals, to be considered in terms of their overall dietary pattern.
In recent years there has been increasing interest in the identification of dietary patterns as consumed by populations. It has been suggested that such analyses could shed light on the complex relationship between diet and chronic disease(Reference Schwerin, Stanton, Smith, Riley and Brett1–Reference Farchi, Mariotti, Menotti, Seccareccia, Torsello and Fidanza3). From a public health perspective, identification of groups within a population which have similar patterns of diet would be of value to policy makers for translating national dietary goals into practical dietary recommendations for the public, for monitoring population trends towards nutritionally ‘healthier’ diets, for identification and surveillance of those at nutritional risk, and for tailoring and targeting public health nutritional interventions. It is also this aspect that is most amenable to change by intervention(Reference Van Horn, Obarzanek, Friedman, Gernhofer and Barton4–Reference Ammerman, Lindquist, Lohr and Hersey9).
Previous approaches to analysis of dietary patterns have utilised frequency of food use to develop food variety scores or a qualitative food use profile of a population(Reference Randall, Marshall, Grahman and Brasure2, Reference Campbell, Roe and Eickwork10, Reference Krebs-Smith, Smiciklas-Wright, Guthrie and Krebs-Smith11). More recently, multivariate statistical techniques have been used to examine the combination of foods consumed by populations, relating these to either population characteristics or morbidity. Several studies have used factor analysis to classify dietary patterns in adults according to the frequency of reported food consumption(Reference Randall, Marshall, Grahman and Brasure2, Reference Nicklas, Webber, Thompson and Berenson12) or reported food intake(Reference Barker, McClean, Thompson and Reid13–Reference Gregory, Foster, Tyler and Wiseman15). However, the factors identified by this technique do not refer to identifiable groups of individuals within a population, and hence do not give an indication of the prevalence of a particular type of diet. On the other hand, cluster analysis aims to identify relatively homogeneous groups within the population based upon selected attributes (dietary variables). The technique has not been widely used to analyse dietary data.
The National Diet and Nutrition Survey (NDNS) of children aged 1½; – 4½; years provided detailed information on the food intake of a national sample of children living in Britain, together with important socio-economic and lifestyle characteristics. The present paper reports an analysis of the NDNS of children aged 1½; – 4½; years in which we used the multivariate statistical technique of cluster analysis to identify groups within this population who reported similar patterns of diet. The food types that characterise the groups, together with the nutrient intakes, sociodemographic and lifestyle characteristics of the groups, are presented herein. To our knowledge cluster analysis has previously not been used to characterise children’s diets in the UK.
Methods
The database used was the Dietary and Nutritional Survey of British children aged 1½; – 4½; years. Briefly, fieldwork was carried out between July 1992 and June 1993, on behalf of the UK Government (Department of Health and the Ministry of Agriculture, Fisheries and Food), by the Social Survey Division of the Office of Population Censuses and Surveys and the Micronutrient Status Group of the Medical Research Council (MRC).
The sample was recruited using a multistage random probability design, with postal sectors as the first stage. The Postage Address File (PAF) was used as a sampling frame. All postal sectors in Wales, England and mainland Scotland were stratified according to region and 1991 census data on social class. One hundred postal sectors were selected as first-stage units, with the chance of selection being proportional to size. One eligible child was randomly selected from each household, making the sample of 2101 children.
The survey design included face-to-face interviews with the child’s mother/carer, to provide information on sociodemographic characteristics of the child’s household including age, sex, social class, income, benefits, geographical area of residence, cigarette smoking and household composition.
A structured interview carried out by trained fieldworkers was completed for 1859 children (88 % of the identified sample) and was considered to be representative of the population in terms of sociodemographic characteristics according to the 1992 General Household Survey(Reference Thomas, Goddard, Hickman and Hunter16).
Ethical approval was obtained from the National Health Service Local Research Ethics Committee for each location and from the MRC Nutrition Unit’s Ethical Committee.
Carers/parents of the child were issued with calibrated food weighing scales and asked to keep a weighed record of all food consumed by the child during a period of four consecutive days (including Saturday and Sundays) for 1675 children. The information was checked during subsequent visits by the interviewer. The response rate for the 4 d intake was 80 %. The present analysis uses only the sample that kept the 4 d weighed intake. One thousand six hundred and seventy-five individuals completed the food intake.
The weighed intake method has the advantage of being the most precise one available for accurately recording amounts of food consumed. The method does not have the disadvantages of inaccuracies attributable to errors in recall and errors involved in estimating portion size are also minimised. However, validation studies of the weighed intake method in adults that have used either doubly labelled water to assess to energy expenditure or urinary N as a marker for dietary N have shown that under-reporting bias is present(Reference Livingstone, Prentice, Strain, Coward, Black, Barker, McKenna and Whitehead17, Reference Bingham and Cummings18).
Analyses were conducted for boys and girls separately. Nineteen food/drink groups were used in the analysis (food groups are detailed in the Appendix) Continuous food and beverage group values (all estimated in g/week or ml/week) were standardised by converting to the standard normal deviate. The clustering technique used was a hierarchical agglomerative (or stepwise) technique available in the SPSS for Windows software. Ward’s method was used, based on squared Euclidean distances(Reference Everitt19). In Monte Carlo studies, Ward’s method has been found to be the most robust clustering method using a similarity matrix based upon squared Euclidean distances(Reference Blashfield20, Reference Edelbrock21). The matrix of distances based upon squared Euclidean distance was computed followed by stepwise fusion of cases. The clustering coefficient was then used to indicate the stage on the agglomeration schedule where large changes between fusions were evident as compared with immediately preceding stages(Reference Everitt19).
As possible instability of the results could be one of the limitations of a cluster analysis, we tested the stability of the cluster solution. Two methods were used: (i) discriminant analysis to test the degree of association between group membership assigned by cluster analysis using nineteen food/beverage groups; and (ii) by randomly splitting the data into two, clustering separately in each subset and comparing cluster membership in each split sample.
Statistical comparisons were made across the clusters in terms of reported food group consumption, intakes of macronutrients and micronutrients, selected socio-economic, demographic and behavioural variables. Parametric one-way ANOVA was used to test for between-group differences in frequency distribution of food groups and nutrients. All food groups were either logged or squared to make the distribution normal. The χ 2 test was used for categorical variables. Statistical analyses were performed using SPSS version 12 (SPSS Inc., Chicago, IL, USA) and STATA version 7 (Stata Corporation, College Station, TX, USA) statistical software packages.
Tables present median food/beverage intakes for boys and girls separately. Cluster median food/beverage intakes below 50 % of the male and female median intake were considered ‘low’. Cluster median intakes between 50 and 99 % and 100 and 149 % were considered respectively ‘moderately low’ and ‘moderately high’, and intakes above 150 % of the male and female median intake were considered ‘high’.
Results
Identification of clusters
In order to test the stability of the solution obtained, a discriminant analysis was undertaken to test the degree of association between group membership assigned by cluster analysis using nineteen food/beverage group variables. The level of agreement between group membership identified by cluster analysis and predicted group membership using discriminant analysis was 87 % among boys and 86 % among girls. We also examined a split sample to test whether the results were different in each split sample. The results show that the split samples were the same identifying three large clusters. These results indicate relatively good overall agreement between actual and predicted group membership using cluster and discriminant analysis, and using the split samples.
Boys’ dietary clusters
Table 1 shows the food group intake by the three clusters for boys.
Median dietary intake (g/week or ml/week) for the three clusters, and the relative levelFootnote *, for boys and girls aged 1½; – 4½; years, Great Britain, 1992–1993
* L, low (<50 % of median intake); ML, moderately low (50–100 % of median intake); MH, moderately high (100–150 % of median intake); H, high (>150 % of median intake).
The ‘Traditional Diet’ (BC1; n 79; 9·3 %) had moderately high (100–150 % of the median intake) intake of white refined cereals, cakes/puddings, egg dishes, bacon/ham, beef/veal/lamb/pork, poultry, meat products, sugar/confectionery and soft drinks. This diet was also moderately low (50–100 % of the median intake) in pasta/rice, wholegrain cereals, high-fat and low-fat dairy products, fat spreads, fish/shellfish, vegetables/salad, chips/potatoes, fruit/nuts and fruit juices.
The ‘Healthy Diet’ (BC2; n 444; 52·3 %) was moderately high in wholegrain cereals, low-fat dairy products, egg dishes, fat spreads, poultry, vegetables/salad, fruit/nuts and fruit juices. This diet was moderately low in pasta/rice, white refined cereals, cakes/puddings, high-fat dairy products, bacon/ham, beef/veal/lamb/pork, prepared meat products, fish/shellfish, chips/potatoes and sugar/confectionery, and low (<50 % of the median) in soft drinks.
The ‘Convenience Diet’ (BC3; n 325; 38·3 %) was moderately high in white refined cereals, cakes/puddings, high-fat dairy products, egg dishes, fat spreads, bacon/ham, beef/veal/lamb/pork and sugar/confectionery, and high (>150 % of the median intake) for prepared meat products, chips/potatoes and soft drinks. This diet was moderately low in pasta/rice, wholegrain cereals, low-fat dairy products, poultry, fish/shellfish, vegetables/salad, fruit/nuts and fruit juices.
Girls’ dietary clusters
Table 1 shows intake of food groups by the three clusters for girls.
The ‘Traditional Diet’ (GC1; n 36; 4.3 %) was moderately high in pasta/rice, wholegrain cereals, cakes/puddings, egg dishes, fat spreads, bacon/ham, beef/veal/lamb/pork, poultry, prepared meat products, fish/shellfish, chips/potatoes and sugar/confectionery, and high in soft drinks. This diet was moderately low in white refined cereals, high-fat dairy products, low-fat dairy products, vegetables/salad and fruit/nuts, and low in fat spreads and fruit juices.
The ‘Healthy Diet’ (GC2; n 487; 58·7 %) was moderately high in wholegrain cereals, low-fat dairy products, egg dishes, poultry, fish/shellfish, vegetables/salad, fruit/nuts and fruit juices and was moderately low in pasta/rice, white refined cereals, cakes/puddings, high-fat dairy products, fat spreads, bacon/ham, beef/veal/lamb/pork, prepared meat products, chips/potatoes, sugar/confectionery and soft drinks.
The ‘Convenience Diet’ (GC3; n 304; 36·6 %) was moderately high in pasta/rice, white refined cereals, cakes/puddings, fat spreads, bacon/ham and sugar/confectionery, and high in prepared meat products, chips/potatoes and soft drinks. This diet was moderately low in wholegrain cereals, high-fat dairy and low-fat diary products, egg dishes, beef/veal/lamb/pork, poultry, fish/shellfish, vegetables/salad, fruit/nuts and fruit juices.
Macronutrient density
Table 2 shows energy and macronutrients as a percentage of energy for boys and girls.
Mean energy intake and macronutrient density (percentage of energy or g/1000 kcal) in the sample population by sex and dietary cluster: boys and girls aged 1½; – 4½; years, Great Britain, 1992–1993
Among boys the ‘Traditional Diet’ had the highest carbohydrate, sugar, total fat, SFA and MUFA as a percentage of energy compared with the ‘Healthy Diet’, which had the lowest percentage of energy as carbohydrate, sugar, total fat, SFA and MUFA, and energy, but the highest percentage of energy as protein. The ‘Convenience Diet’ was mid way between the ‘Traditional Diet’ and the ‘Healthy Diet’, but with high energy values. There were no significant differences for n-3, n-6 and total PUFA.
Among girls, the ‘Healthy Diet’ had the highest percentage of energy as protein, and the lowest percentage of energy as carbohydrate, sugar, total fat and SFA, whereas the ‘Traditional Diet’ had the highest percentage of energy as carbohydrate, sugar, total fat, SFA and MUFA, and the lowest energy. The ‘Convenience Diet’ was between the ‘Traditional Diet’ and the ‘Healthy Diet’ according to the values of macronutrient densities, whereas its energy was the highest among the diet clusters. There were no significant differences between the clusters for n-3, n-6 and total PUFA.
Micronutrient density
Table 3 shows the micronutrient density of the three clusters for boys and girls.
Mean micronutrient density (percentage of energy or mg/100 kcal, μg/1000 kcal) of the sample population by sex and dietary cluster: boys and girls aged 1½; – 4½; years, Great Britain, 1992–1993
Among boys, the ‘Traditional Diet’ had the lowest density for all vitamins and minerals apart from vitamin B6 and vitamin B12, which were the same as in the ‘Convenience Diet’, and carotene, which was lower in the ‘Healthy Diet’.
Among girls, the ‘Traditional Diet’ had the lowest density for all vitamins and minerals apart from Cu and vitamin D, which were the same as in the ‘Convenience Diet’, whereas pantothenic acid and biotin were lower in the ‘Convenience Diet’.
Social and economic profile
Table 4 shows social and economic characteristics for boys’ and girls’ families and by cluster.
Socio-economic and behavioural variables by sex and by dietary cluster: boys and girls aged 1½; – 4½; years, Great Britain, 1992–1993
Among boys, those consuming the ‘Traditional Diet’ included the highest proportion of manual workers and the highest proportion of families receiving benefits compared with the ‘Convenience Diet’ and the ‘Healthy Diet’. There was no significant difference between household income, region or mother’s education by the diet clusters. Among those consuming the ‘Traditional Diet’ there was a high proportion of light smokers and heavy smokers and the lowest proportion of non-smokers compared with the other diet groups. Among those consuming the ‘Traditional Diet’, there was a lower proportion of parents who lived in a house/bungalow and the highest proportion of parents living in a flat compared with other diet groups. In the ‘Healthy Diet’ group, a greater proportion of children were not feeling well during the diet survey, compared with the ‘Traditional Diet’ and the ‘Convenience Diet’.
Among girls there was no significant difference in social class, household income, region, mother’s smoking, mother’s education or type of accommodation by diet group. The percentage of parents receiving benefits was lower in the ‘Traditional Diet’ compared with the ‘Healthy Diet’ and ‘Convenience Diet’. The proportion of children not feeling well during the diet survey was lower in the ‘Healthy Diet’ than in the ‘Traditional Diet’ and ‘Convenience Diet’ groups.
Discussion
In the present paper we report the results of an investigation that used cluster analysis to identify groups of children aged 1½; – 4½; years with similar patterns of diet within the UK population. Compared with a more traditional approach of a priori classification of individuals, e.g. by social class, followed by ANOVA, cluster analysis adopts a more dynamic approach to exploring patterns of food intake by grouping participants with comparable combinations of food types.
We characterised the sample using nineteen food groups and then used cluster analysis to identify similar eating patterns. Three clusters were identified in the sample of boys and three clusters in the sample of girls characterising dietary patterns we labelled the ‘Healthy Diet’, the ‘Convenience Diet’ and the ‘Traditional Diet’. Some 52·3 % of boys and 58·7 % of girls consumed the ‘Healthy Diet’; 38·3 % of boys and 36·6 % of girls consumed the ‘Convenience Diet’; and 9·3 % of boys and 4·3 % of girls consumed the ‘Traditional Diet’.
Among boys, those consuming the ‘Traditional Diet’ included the highest percentage of manual workers and parents receiving benefits, and the highest percentage of mothers who smoked compared with the ‘Convenience Diet’ and ‘Healthy Diet’ groups. There was a lower proportion of parents living in a house or bungalow compared with the other diet groups. There was no significant difference in income or region by diet group. Among girls, a higher proportion of parents received benefits in the ‘Traditional Diet’ group compared with the other diet groups, but no other differences in social/economic variables. The ‘Traditional Diet’ had the lowest micronutrient density for most of the vitamins and minerals, for both boys and girls, compared with the ‘Healthy Diet’ and the ‘Convenience Diet’.
In previous research, a priori classification by social class and gender showed that among girls there was no significance difference for food groups, but for boys vegetables/salad, fruit/nuts, egg dishes and bacon/ham were significantly different by social class(Reference Gregory, Collins, Davies, Hughes and Clarke22). In the present study, differences in mean levels by clusters were larger and combinations of undesirable/desirable dietary and other factors were observed. For example, the ‘Traditional Diet’ included high fat and low micronutrient densities, and the group had a high prevalence of maternal smokers. On the other hand, the ‘Healthy Diet’ group had a low mean fat intake, higher micronutrient densities and fewer maternal smokers.
Other studies have shown that childhood diet may influence the development of chronic diseases in adult life. The Bogalusa Heart Study has found that coronary atherosclerosis and essential hypertension can begin in childhood, and both are risk factors for CVD(Reference Berenson, Wattigney, Bao, Nicklas, Jiang and Rush23, Reference Berenson, Srinivasan and Nicklas24). Other studies have shown that childhood fruit and vegetable consumption may have a long-term protective effect on cancer risk in adulthood(25). In the present study children in the ‘Convenience Diet’ and ‘Traditional Diet’ groups were not eating enough fruit and vegetables according to recommendations. Findings like this underline the importance of the UK Government’s national school fruit scheme which entitles 4–6-year-old children a piece of fruit each school day(Reference Wells and Nelson26, Reference Fogarty, Antoniak, Venn, Davies, Goodwin, Salfield, Stocks, Britton and Lewis27).
It is well established from studies in adults that smoking status is associated with dietary quality(Reference Fulton, Thomson, Elton, Brown, Wood and Oliver28–Reference Jeffrey, Hampl and Betts33). Smokers have been reported to have diets lower in fibre and antioxidants and higher in fats. They consume less fruit and vegetables and wholemeal bread and more meat products, chips and soft drinks. Passive smoking has also has also been shown to be associated with dietary quality(Reference Thornton, Lee and Fry34, Reference Koo, Kabat, Rylander, Tominaga, Kato and Ho35). Passive smokers who live with smokers have food and nutrient intakes that are intermediate between those of smokers and non-smokers(Reference Johnson, Wang, Smith and Connoly36, Reference Crawley and White37). As many children eat family foods, it is likely that their diet will be affected to some extent by the smoking habits of their parents.
Four studies have investigated the effects of parental smoking on the diet quality of their children. Johnson et al.(Reference Johnson, Wang, Smith and Connoly36) described dietary intakes among children in low-income families in the USA according to the smoking status of their parents; Crawley and White(Reference Crawley and White37) analysed the diets of British teenagers by smoking status; and Burke et al.(Reference Burke, Gracey, Milligan, Thompson, Taggart and Bellin38) investigated the health behaviours in 10–12-year-olds in Australia. All of these studies suggested that the children of smokers consumed a diet that conformed less closely to current recommendations on healthy eating than did the children of non-smokers. Rogers et al.(Reference Rogers and Emmett39) have shown that dietary differences between children of smokers and non-smokers are in line with those observed between the diets of adult smokers and non-smokers. In the present study, boys eating a ‘Traditional Diet’ were more likely to have mothers who smoked and to be receiving benefits, although the patterns was not reproduced in girls.
Other studies among adults have shown that more health-conscious behaviour is associated with higher social class(Reference Pryer, Nichols, Elliott, Thakrar, Brunner and Marmot40). In the present study, children from higher social classes had the ‘Healthy Diet’ with the benefits of higher micronutrient density and healthy macronutrient density. Their parents were less likely to be on benefits and there were fewer maternal smokers.
In conclusion, our research has identified three diet groups in both boys and girls aged 1½; – 4½; years in the UK population that differed not only in reported dietary intake, but also with respect to their micronutrient and macronutrient density, social status and economic profile. Cluster analysis has been used to group foods or nutrients for the adult population(Reference Farchi, Mariotti, Menotti, Seccareccia, Torsello and Fidanza3, Reference Pryer, Cook and Shetty41–Reference Haines, Hungerford and Popkin45), but we know of no other studies where cluster analysis has been used to identify food patterns in young children in Great Britain. Our results should be relevant to the development, monitoring and targeting of public health nutrition policy in the UK. In particular, they could be used to develop a tailored health promotion programme based on the diet clusters, by positively reinforcing the healthy diet and targeting the convenience and the traditional diet for significant improvements. Further research is needed on the consequences for chronic disease in the future for these children.
Acknowledgements
Both the authors have read and approved the paper for publication, and both were involved in the conception, writing and revising of the paper. J.A.P. did the analysis. There are no financial or other relationships that lead to competing interests. This paper has not been published elsewhere in part or in full or been submitted elsewhere.
Appendix Food groups
1. Pasta/rice and other miscellaneous cereals.
2. White bread and refined breakfast cereals such as cornflakes, Rice Krispies, Sugar Puffs, Honey Smacks, etc.
3. Brown and granary bread and wholegrain cereals such as All Bran, Shredded Wheat, Weetabix, Cheerios, etc.
4. Biscuits, cakes, pastries and puddings such as buns, cakes, pastries, fruit pies, doughnuts, jam tarts, scones, ice cream, instant whip, fruit crumble, custard puddings, cream desserts, jelly, fools, sponge puddings, milk puddings.
5. High-fat dairy such as whole milk, cream, cheese, whole-milk yoghurts, fromage frais.
6. Low-fat dairy such as semi-skimmed milk, skimmed milk, low-fat yoghurts, low-fat yoghurt drinks, low-fat fromage frais.
7. Eggs and egg dishes such as boiled, scrambled, omelettes, soufflé, quiche, scotch eggs, etc.
8. Fat spreads such as butter, polyunsaturated margarines, other margarines and spreads.
9. Bacon and ham such as bacon joints, rashers, gammon joints/steaks, ham.
10. Beef, veal, lamb and pork: beef and veal dishes such as beef and veal joints, steaks, minced beef, stewing steak, beef stews and casseroles, meat balls, lasagne, chilli con carne, curry, bolognaise sauce; lamb and lamb dishes such as lamb joints, chops, curries, Irish stews, lamb casseroles and stews; pork and pork dishes such as joints, chops, belly rashers, pork stews/casseroles, sweet and sour pork, etc.
11. Poultry such as coated chicken, chicken and turkey drumsticks, chicken pieces, nuggets, fingers, burgers, roast chicken or turkey, barbecued/curries/stews/casseroles, chicken/turkey roll.
12. Prepared meats including burgers, kebabs, sausages, meat pies and pastries, pork pies, veal and ham pies, sausage rolls, other meat products such as game pies, faggots, black pudding, meat paste, canned meats, salami, meat loaf, etc.
13. Fish/shellfish including white fish in batter or breadcrumbs, fish fingers, fish cakes, fish paste, fish in sauces, fish pie, kedgeree, oily fish including herrings, kippers, mackerel, sprats, salmon, tuna and sardines, and shellfish.
14. Vegetables and salad including salad vegetables such as lettuce, tomatoes, cucumber, coleslaw and prepared salads; vegetables including beans/pulses, cooked vegetables, vegetable casseroles/stews, curries, cauliflower cheese.
15. Chips and potato products including fried potatoes and chips potato waffles, hash browns, roast, sautéed and croquettes, other potatoes including boiled, mashed, jacket and instant potato, crisps, puffs, rings, Twiglets.
16. Fruit and nuts including fruit raw, cooked, canned, fruit pie fillings; nuts including almonds, hazelnuts, mixed nuts, peanuts, peanut butter, Bombay mix and seeds.
17. Sugar and confectionery including sugar and preserves including white and brown sugar, black molasses, treacle, syrup, jams, marmalade, glace cherries, mixed peel, marzipan; confectionery includes boiled sweets, gums, pastilles, fudge, chews, mint, rock, liquorice, toffee, popcorn, chocolate bars, filled bars, assortments.
18. Fruit juices including single fruit juice, mixed fruit juice, canned, bottles, cartons, still, carbonated, freshly squeezed.
19. Soft drinks including carbonated and low-calorie versions, fruit squashes, cordials, fruit drinks, syrups.
