Food preferences, likes and dislikes for food items, are developed in early childhood(Reference Alderson and Ogden1). Parents and carers determine the food behaviour(Reference Alderson and Ogden1, Reference Hoerr, Lee and Schiffman2) of infants and toddlers by the availability of food. Maternal attitudes and health behaviour may influence children's food patterns but the implications for the child's diet may still differ from the mother's own diet(Reference Alderson and Ogden1–Reference Francis, Hofer and Birch3). Previous evidence suggests similarity of food habits in a family, and especially between diets of mothers and their children(Reference Francis, Hofer and Birch3–Reference Brekke, van Odijk and Ludvigsson8). Traditional and health-conscious food choices established in childhood were observed to remain even in adulthood(Reference Mikkilä, Räsänen and Raitakari9).
Complementary feeding of the child has started with infant formulas at the average age of 2 months in Finland(Reference Hasunen and Ryynänen10, Reference Erkkola, Pigg and Virta-Autio11). Mashed vegetables, fruits and other solids follow after the age of 4–5 months(Reference Robinson, Marriott and Poole7, Reference Erkkola, Pigg and Virta-Autio11–Reference Koehler, Sichert-Hellert and Kersting14). Beside home-made food, manufactured baby foods(Reference Nasirpour, Scher and Desobry15), convenience foods(Reference Alexy, Sichert-Hellert and Rode16) and other manufactured food products(Reference Buckley, Cowan and McCarthy17) are part of the childhood diet. Convenience-related quality of food is determined(Reference Buckley, Cowan and McCarthy17) to be associated with reducing the time or other input required in food shopping, preparation or cooking the meal. The nutritional composition of manufactured food products has been discussed due to their dried, isolated or condensed ingredients(Reference Nasirpour, Scher and Desobry15).
Child health-care clinics give advice about feeding practices to the mothers of almost all infants in Finland(Reference Hasunen and Ryynänen10, Reference Huurre, Laitinen and Hoppu18). Nutrition counselling aims at promoting breast-feeding and encourages a balanced diet. Healthy food choices such as consumption of skimmed milk, whole-grain bread, and fruits and vegetables are emphasised in child health-care clinics in contrast to undesirable food behaviour, for example the consumption of high-energy snacks(Reference Hoerr, Lee and Schiffman2, Reference Talvia, Räsänen and Lagström6, Reference Ziegler, Hanson and Ponza19–Reference Maffeis, Grezzani and Perrone22). The consumption of sugar-sweetened beverages has increased in the past decade and is under consideration due to replacing more nutritious beverages(Reference Ziegler, Hanson and Ponza19, Reference Ludwig, Peterson and Gortmaker23). The consumption of high-sugar food products has been reported to be associated with mothers and their children at least in infants(Reference Brekke, van Odijk and Ludvigsson8), in 2-year-olds(Reference Hoerr, Lee and Schiffman2) and in 5-year-old daughters(Reference Fisher, Mitchell and Smiciklas-Wright5).
Cluster analysis has been used in three dietary surveys of infants or toddlers, which resulted in six(Reference Knol, Haughton and Fitzhugh24) and seven clusters(Reference Räsänen, Lehtinen and Niinikoski25) or analysed beverages only(Reference LaRowe, Moeller and Adams26). Dietary patterns have been analysed more frequently by either principal component analysis(Reference Robinson, Marriott and Poole7, Reference Northstone and Emmett27–Reference North and Emmett29) or factor analysis(Reference North and Emmett29, Reference Nicklaus, Boggio and Issanchou30) in early childhood. The present study has the benefit of having dietary data for the child–mother pairs, which enables to study the similarity of dietary clusters within these dyads. The aim of the present study is to define the dietary clusters in the mothers and in three age groups of children in the Diabetes Prediction and Prevention (DIPP) Nutrition Study. The present study also elucidates similarities of dietary clusters within the mother–child dyads. The associations of social background factors with the clusters predicting selections of convenient foods (including ready-to-eat baby foods and fast foods) within the children were determined.
Subjects and methods
In the DIPP Study(Reference Kupila, Muona and Simell31), newborn infants from the areas of three university hospitals in Finland were screened for HLA-DQB1 conferred susceptibility to the type 1 diabetes using cord blood samples. Infants carrying increased genetic susceptibility (HLA-DQB1*02/0302 heterozygous and DQB1*0302/x-positive subjects; x stands for homozygosity or a neutral allele) belong to cohorts being monitored for diabetes-associated auto-antibodies at 3–12-month intervals. The procedures of the study were approved by the local Ethics Committees. The mother or other representative of the family signed informed written consents.
Children and their mothers with dietary data
The DIPP Nutrition Study falls within the framework of the DIPP Study(Reference Virtanen, Kenward and Erkkola32) and was carried out in two university hospitals (Oulu and Tampere). The present study comprises the at-risk children (n 2268) belonging to time-restricted, cross-sectional age groups on annual visits in 2003–2004. The food consumption was collected by 3-d food records (FR) about the first (n 719), third (n 708) or sixth (n 841) birthday. The FR covered both weekdays and weekend and separate forms were provided for the day care. The parents and day carers were instructed to record all the foods and drinks by estimated portions as volume or pieces. The FR was checked by a trained nurse in the study centre at annual visit. The FR were entered into a dietary database using standard volume-to-weight conversions by in-house software. Food consumption was aggregated to average daily food use (g/d) by the food classification of the food composition database(33). Children's sex and day care (home/outside home) during recording days were entered as background factors.
The mothers participating in the DIPP Nutrition Study (n 4939) gave postnatal estimation of their diet during the 8th month of pregnancy(Reference Erkkola, Karppinen and Javanainen34). The dietary data of the mothers were enquired by a 181-item FFQ, which was specifically developed and validated for the present study design(Reference Erkkola, Karppinen and Javanainen34). Detailed information of the procedures of the FFQ and social background of the mothers has been described earlier(Reference Arkkola, Uusitalo and Kronberg-Kippila35). If frequency data were inconsistent or there were ten or more food items with missing frequency data, the questionnaire was rejected (1·6 %) from the analyses. Other missing frequencies were imputed as zero. Daily energy intake and the consumption (g/d) of food items were calculated by food classification described in the earlier report(Reference Arkkola, Uusitalo and Kronberg-Kippila35). The cluster analysis was carried out in all the mothers of the DIPP Nutrition Study. The effect of omitted frequencies of the FFQ was studied by sensitivity analysis of dietary clusters, and no association between omitting frequencies and clusters was observed.
The social factors of the mother and the family were enquired and entered into the study data covering age, basic and professional education, and number of siblings. The matching pairs of a child and the mother were picked to study the familial association of dietary clusters within the dyads (Fig. 1). Some mothers were included twice (n 134) in the pair analyses of the present study because they had twins (n 42) or their children were represented in two (or three) age groups.
Data handling procedures
Forty-nine and fifty-two selected food groups were used to form the clusters of children and mothers, respectively. Foods that were consumed by less than 10 % of the individuals in each age group were omitted from the analysis. Such variables play little role in the description of typical behaviour, but often have undesirable effects on the group structure in terms of very small clusters of outliers. Items with a very skew distribution were coded as binary 0–1 variables, indicating whether the item was consumed at all or not. Continuous variables were winsorised at the 95th quantile (to avoid distortion by outliers), and then standardised to zero mean and unit variance before the analysis.
Hierarchical cluster analysis
After pre-processing of the data, the individuals were grouped by Ward's minimum variance method(Reference Ward36) implemented in SAS/PROC CLUSTER (SAS Institute, Cary, NC, USA). Boys and girls were not separated in the analysis. Different numbers of clusters were considered, and the one with the most realistic grouping was selected for reporting. Spearman's rank correlations between cluster indicator variables and the food items were estimated in order to describe cluster characteristics relative to the other groups. The rank correlation was selected because it measures here the general tendency of values to be greater (or smaller) in one group than in the others, but omits the scale of using only the rankings. Summary statistics (n, mean, interquartile range) were used to quantify absolute differences between the groups of children and mothers.
Analysis of associations
The associations between the background characteristics, mother's and children's dietary clusters were assessed by frequency tables and by χ2-tests for the null hypothesis of independence of rows (the mother's cluster) and columns (the child's cluster). Statistical significance was taken as less than 5 % (two-sided). The associations between cluster membership and background variables (sex, day care, maternal age, number of siblings, basic and professional education of the mother) were investigated by logistic regression models using one and multiple explanatory variables. Obsolete and overlapping effects and second-order interactions were eliminated from the multiple logistic regression model using model selection procedures. Chosen models were not sensitive to the choice of selection procedure.
Hierarchical cluster analysis resulted in three main dietary clusters in the age groups of 1, 3 or 6 years (Table 1). The ‘healthy’ or ‘healthy, low-fat’ dietary cluster was detected for each age group with high correlations in skimmed milk, whole-grain bread and vegetables. The ‘traditional’ cluster with a relatively high intake of dairy spread and high-fat milk was also determined for all ages. The Spearman correlation coefficients between clusters and food consumption varied, however, by age. The cluster ‘fast food, sweet’ had the highest positive coefficients with the intakes of sugar-sweetened beverages, fried potatoes, chips, nuts and dried fruit in 3- and 6-year-old children. In infants, the cluster ‘ready-to-eat, baby foods’ had high positive coefficients with intakes of infant formulas and manufactured baby foods and negative coefficient with intakes of potatoes, vegetables, bread and spread.
* Coefficients ≤ − 0·20 or ≥ 0·20.
The differences in food consumption profiles were remarkable for the clusters of 6-year-old children, e.g. consumption of milks, soft drinks, potato or berry dishes, but slightly smaller among younger children (Tables 2–4). Similarly, for some food groups, the consumer proportion was less than 25 % as shown by the lower quartile value of 0. Even 6-year-old children consumed few food items in 3 d and the lower quartile was 0 for many food groups (Table 4). The consumption of sweetened beverages was 3-fold for 6-year-old members of the cluster ‘fast food, sweets’ compared with the members of the cluster ‘modern, healthy’. Even the lower quartile consumed sweets daily in all clusters of 3- and 6-year-old children. For 3-year-old children, the mean consumption of potatoes, meat dishes and bread was on the same level by all the clusters but differences existed in the consumption of porridge, cooked vegetables and milk (Table 3).
In the baby cluster ‘ready-to-eat baby foods’, the consumption of the lower quartile was 0 for most of the food groups except manufactured baby food products (Table 2). The consumption profile of fruit-based baby foods was as common as in all the three clusters. Instead, manufactured baby foods containing meat or cereals were common for the infants in the cluster ‘ready-to-eat baby foods’. Otherwise, skimmed or low-fat milks and cooked vegetables were common in the cluster ‘healthy’, while high-fat milk, home-made meat dishes and porridge were common in the cluster ‘traditional’ among the infants.
The hierarchical cluster analysis of all the mothers enrolled in the DIPP Nutrition Study and giving dietary data (n 4862) resulted in six main clusters (Table 5). Most mothers belonged to the health-oriented clusters: ‘fat conscious’; ‘modern, healthy’; or ‘small amounts’. Unfavourable dietary elements belonged to the clusters ‘fast food, plenty’, ‘refined, sugar and butter’ and ‘fast food, sweet’ of mothers.
Some familial dependence on dietary clusters was observed in mother–child pairs of 6-year-old children (P = 0·035) but not in younger children (Table 6). In 6-year-old children, the cluster ‘healthy, low-fat’ had a high proportion in the clusters ‘fat conscious’ and ‘modern, healthy’ of mothers. A higher frequency than expected was observed for children being members of the cluster ‘fast food, sweet’ when mother belonged to the cluster ‘fast food, plenty’, ‘refined, sugar and butter’ or ‘sweet, fast food’. Marginal significance of familiality (P = 0·054) at the age of 3 years was observed within child–mother dyads. At the age of 1 year, no indication of familiality was observed.
The cell values are proportions (%) of the children in clusters of their mothers.
* In 1-year-old infants, the differences are not marked.
† Arrows indicate the differences between the observed and expected cell frequencies that contribute a value of ≥ 1 to the χ2-test statistic. The arrow ↑ shows a larger observed frequency and ↓ shows a smaller observed frequency than expected in 3- and 6-year-old children.
Associations of social background factors with the dietary patterns were studied for the clusters representing convenience foods, i.e. ‘ready-to-eat baby-food’ and ‘fast foods, sweets’ in the age groups of children (Table 7). At the age of 1 year, the cluster ‘ready-to-eat baby foods’ was prominent for boys and children cared at home, but the cluster was not associated with maternal characteristics. Maternal characteristics, young age and low basic or professional education of the mother, were associated with the cluster ‘fast food, sweet’ of the child at the age of 3 years. None of the background factors was associated with the cluster ‘fast food, sweet’ at the age of 6 years.
* P values indicate that the variable was significant at the 10 % level in a multiple logistic regression model after the variable selection procedure.
The consistency between the dietary clusters of the children and their mothers was obvious in the age group of 6-year-olds. Both the clusters ‘healthy, low-fat’ and the cluster ‘fast food, sweet’ in children had a higher frequency than expected in the corresponding clusters of mothers. This is consistent with the earlier results. Similarities in the child–mother pairs have been observed for a prudent diet(Reference Robinson, Marriott and Poole7), fruits and vegetables(Reference Hannon, Bowen and Moinpour4, Reference Talvia, Räsänen and Lagström6), regulation of less-healthy foods(Reference Alderson and Ogden1) but also for high-sugar beverages(Reference Hoerr, Lee and Schiffman2, Reference Fisher, Mitchell and Smiciklas-Wright5). However, the maternal background was not associated with the convenient dietary cluster of that age in the present study, which may also imply the increasing independency of the child in food selections.
The young age and low education of the mother was associated with the convenient dietary cluster only in 3-year-olds. The weak association between convenient dietary clusters in pre-school children and maternal background may result from great differences in dietary habits by age during years of transition from breast-feeding to family food. In previous studies(Reference Hoerr, Lee and Schiffman2, Reference Hannon, Bowen and Moinpour4–Reference Talvia, Räsänen and Lagström6, Reference Brekke, van Odijk and Ludvigsson8), some dependence between food consumption of the children and maternal characteristics has been reported. High maternal education associated by earlier studies with healthy food pattern(Reference Robinson, Marriott and Poole7, Reference Mikkilä, Räsänen and Raitakari9, Reference Knol, Haughton and Fitzhugh24, Reference Northstone and Emmett27, Reference North and Emmett29) and fibre intake(Reference Garemo, Arvidsson Lenner and Nilsson37), and young age with consumption of high-sugar foods and drinks(Reference Brekke, van Odijk and Ludvigsson8).
For infants and small children, mothers might serve meals or special servings, which may better follow dietary guidelines or assumed guidelines(Reference Alderson and Ogden1). In 1-year-old children of the present study, this may be associated with the cluster ‘ready-to-eat baby foods’. In a European comparison, parents in many countries preferred home-made food for their infants(Reference Robinson, Marriott and Poole7, Reference Synnott, Bogue and Edwards13, Reference Koehler, Sichert-Hellert and Kersting14) but in Finland, the manufactured baby food products (fruit purée, baby meat products and baby porridge) were served to half of the infants according to the present results as well as in previous Finnish studies(Reference Hasunen and Ryynänen10–Reference Freeman, van't Hof and Haschke12). On the contrary, the infant guideline dietary pattern had negative coefficients for manufactured baby food in the UK(Reference Robinson, Marriott and Poole7).
Three dietary clusters were sufficient for the interpretation of the present results among the children studied. The earlier Finnish study, applying the cluster analysis in 7-year-old children(Reference Räsänen, Lehtinen and Niinikoski25), identified four clusters ‘cereals’, ‘sugar and sweets’, ‘bread and skimmed milk’ and ‘dairy’, which corresponded rather well with the present results in 6-year-old children. The health-conscious pattern as well as fast-food or junk-food pattern have been identified in children by the cluster analysis(Reference Knol, Haughton and Fitzhugh24, Reference Räsänen, Lehtinen and Niinikoski25) and principal component analyses(Reference Robinson, Marriott and Poole7, Reference Northstone and Emmett27–Reference North and Emmett29) in other countries. The basic methodology of the cluster analysis is to group individuals, whereas the factor analysis is to group the variables (foods)(Reference Newby and Tucker38). The cluster analysis was preferred because it may be easier to interpret as each individual belongs to one cluster only and because it is more applicable to the risk analysis later(Reference Newby and Tucker38). The present results from the hierarchical cluster analysis in the mothers were partly similar to our previous factor analysis findings from a subpopulation of the same mothers(Reference Arkkola, Uusitalo and Kronberg-Kippila35). At least, the factors ‘healthy’, ‘low-fat foods’ and ‘fast food’ overlapped the clusters of the present study.
The dietary clusters of the children were based on the FR of 3 d and, obviously, they do not reflect the whole variety or habits of food selection. Weekend days may differ in the children's diet from that of weekdays(Reference Garemo, Lenner and Strandvik21), but this was controlled for in the present study by guiding the diary recording to cover one weekend day and two weekdays. There are limitations in the present study design to define associations in dietary clusters within mother–child dyads. The study protocol did not give the best opportunity to find the association between the diet of 6-year-old children and their mothers. Our sample may be selected towards a more healthy food behaviour and due to drop-outs the selectivity may be the strongest at the age of 6 years. Furthermore, the background and the diet of the mothers were collected at the time of birth which may hinder the association with the diets of 6-year-olds. The children studied in the present paper are carrying increased genetic susceptibility for type 1 diabetes but the dietary clusters correspond to the earlier Finnish results(Reference Räsänen, Lehtinen and Niinikoski25). The duration of breast-feeding(Reference Erkkola, Pigg and Virta-Autio11) in the DIPP children also corresponded to the general impression(Reference Hasunen and Ryynänen10). We know little about how parents react to information about increased genetic risk, but it has been shown that information caused no differences at anxiety levels in the parents of high-risk infants compared with the parents of low-risk infants(Reference Simonen, Korhonen and Simell39).
The type of day care, at home or in kindergarten, seemed to influence the food behaviour, since the proportion of subjects in the cluster ‘fast food, sweets’ was the lowest among children cared in kindergarten. The present results give evidence for a better dietary quality in children cared outside home(Reference Ziegler, Hanson and Ponza19), but the results are not consistent and vary by nutrient(Reference Ziegler, Hanson and Ponza19, Reference Garemo, Lenner and Strandvik21). Day care outside home covers a considerable proportion of daily food consumption but may be more common for older and more educated mothers. Mothers may have difficulties in putting dietary guidelines into practice at home(Reference Alderson and Ogden1, Reference Koehler, Sichert-Hellert and Kersting14, Reference Huurre, Laitinen and Hoppu18, Reference Krebs20). Dietary counselling has been reported more frequently by nurses than by mothers in visits to child health-care clinics(Reference Huurre, Laitinen and Hoppu18). It can be speculated that dietary counselling may not face the practical needs of the family or mothers are too stressed or preoccupied to notice all guidelines offered to them.
None of the infant clusters was associated with the dietary clusters of the mother. Neither the ‘ready-to-eat baby foods’ cluster in 1-year-old children in the present study was associated with maternal characteristics. Thus, the present results do not confirm previous results(Reference Robinson, Marriott and Poole7). Complementary feeding of infants has been started too early compared with the guidelines in many countries(Reference Freeman, van't Hof and Haschke12, Reference Koehler, Sichert-Hellert and Kersting14). The high frequency and early introduction of baby foods may give a reason for studying the effects of convenient food habits on later food habits in longitudinal studies. The cross-sectional study design of the present study did not allow for studying longitudinal trends in childhood dietary patterns.
As a conclusion, the present results confirm the concern of consumption of fast foods and high-sucrose sweets and bakery in pre-school children(Reference Garemo, Lenner and Strandvik21, Reference Ludwig, Peterson and Gortmaker23, Reference LaRowe, Moeller and Adams26), and therefore the parents must be aware of the effect of food available at home. Parents should be encouraged to provide their children healthy food and ensure availability of fruits and vegetables in forms that support the increasing self-regulation of the child.
M. L. O., J. N. and S. M. V. planned the present study and M. L. O. was responsible for writing the article; J. N. was responsible for statistical analysis; L. U., J. J. T. and T. A. prepared the data for the analyses; C. K.-K. coordinated the management of nutrition data; M. K. and R. V. are responsible for the clinical study; S. M. V. has planned the nutrition study within the DIPP project and is responsible for the DIPP Nutrition Study. All the authors read the manuscript and contributed to the writing. None of the authors had any conflicts of interest. Financially supported by: the Academy of Finland (grants 63 672, 79 685, 79 686, 80 846, 201988 and 210632); the European Foundation for the Study of Diabetes, the Finnish Diabetes Association; the Finnish Diabetes Research Foundation; the Finnish Pediatric Research Foundation; the Häme Foundation of the Finnish Culture Fund; the Juho Vainio Foundation; the Yrjö Jahnsson Foundation; the European Foundation for the Study of Diabetes; Medical Research Funds, Turku, Oulu and Tampere University Hospitals; the Juvenile Diabetes Research Foundation (grants 197032, 4-1998-274, 4-1999-731 and 4-2001-435); Novo Nordisk Foundation; EU Biomed 2 Program (BMH4-CT98-3314).