Diet patterns are associated with demographic factors and nutritional status in South Indian children

Abstract The burden of non‐communicable chronic disease (NCD) in India is increasing. Diet and body composition ‘track’ from childhood into adult life and contribute to the development of risk factors for NCD. Little is known about the diet patterns of Indian children. We aimed to identify diet patterns and study associations with body composition and socio‐demographic factors in the Mysore Parthenon Study cohort. We collected anthropometric and demographic data from children aged 9.5 years (n = 538). We also administered a food frequency questionnaire and measured fasting blood concentrations of folate and vitamin B12. Using principal component analysis, we identified two diet patterns. The ‘snack and fruit’ pattern was characterised by frequent intakes of snacks, fruit, sweetened drinks, rice and meat dishes and leavened breads. The ‘lacto‐vegetarian’ pattern was characterised by frequent intakes of finger millet, vegetarian rice dishes, yoghurt, vegetable dishes and infrequent meat consumption. Adherence to the ‘snack and fruit’ pattern was associated with season, being Muslim and urban dwelling. Adherence to the lacto‐vegetarian pattern was associated with being Hindu, rural dwelling and a lower maternal body mass index. The ‘snack and fruit’ pattern was negatively associated with the child's adiposity. The lacto‐vegetarian pattern was positively associated with blood folate concentration and negatively with vitamin B12 concentration. This study provides new information on correlates of diet patterns in Indian children and how diet relates to nutritional status. Follow‐up of these children will be important to determine the role of these differences in diet in the development of risk factors for NCD including body composition.


Introduction
The World Health Organisation (WHO) has estimated that 53% of all deaths in India in 2008 were attributable to non-communicable conditions such as cardiovascular disease and type 2 diabetes (World Health Organisation 2011). It is predicted that by 2030, such conditions will account for 75% of deaths (Patel et al. 2011). Since the 1970s, there has been a reduction in intakes of wholegrain cereals, pulses, fruits and vegetables in India while intakes of meat products, refined grains and salt have increased (Popkin 2002;Misra et al. 2011;Popkin et al. 2012). In parallel with these dietary changes, there is evidence of increased prevalence of child and adolescent overweight. It has been estimated that in 2011, the prevalence of childhood overweight was 15 million and that of abdominal obesity was 4 million (Gupta et al. 2012). Data from large cross-sectional surveys in Delhi have shown that there was an increase in adolescent obesity from 9.8% in 2006 to 11.7% in 2009 (Pandey et al. 2009). In Kerala, two cross-sectional surveys of more than 20 000 children found that the percentage of overweight children increased from 4.94% to 6.57% between 2003 and 2005 (Raj et al. 2007). Recent data suggest that prevalence of adolescent overweight is almost 20% in urban areas (Goyal et al. 2010;Jain et al. 2010).
Diet and body composition are important contributory factors to chronic disease risk in adult life (Shetty 2002;Misra et al. 2011). Since evidence suggests that dietary patterns 'track' from childhood to adulthood (Dunn et al. 2000;Mikkila et al. 2005), it is important to study diet patterns during childhood as this may provide an opportunity to intervene and prevent chronic disease. However, little is currently known about the dietary patterns of children in India. Studies in European children have looked at the associations between diet patterns and body composition and found that 'snacking' and 'energy dense' patterns are associated with overweight (Lioret et al. 2008) and higher fat mass (Johnson et al. 2008). South Asian children tend to have a higher body fat percentage compared with Western children (Yajnik et al. 2003;Krishnaveni et al. 2005), but to our knowledge diet patterns and their relationships with body composition have not been studied among South Asian children. Despite the rapid urbanisation currently occurring in South Asia (United Nations Department of Economic and Social Affairs 2011), diet patterns are likely to differ from those of Western children.
The three aims of this study were: (1) to describe diet patterns of children living in and around the city of Mysore, located in central South India, using a principal component analysis (PCA) of food frequency data; (2) to examine demographic variables as correlates of diet patterns; and (3) to examine diet patterns as correlates of body composition and micronutrient status (folate and vitamin B12).

Method Participants and Setting
Children were recruited from the Mysore Parthenon study, a birth cohort set up to investigate the long-term cardiovascular risk outcomes associated with maternal gestational diabetes and body composition of the infant at birth. Details of the cohort have been published elsewhere . In brief, between June 1997 and August 1998, pregnant women attending the antenatal clinic of Holdsworth Memorial Hospital, living in the city of Mysore and surrounding rural areas were recruited to the study if they fulfilled the following criteria: non-diabetic prior to pregnancy; <32 weeks gestation at time of recruitment; planning to deliver at Holdsworth Memorial Hospital. A total of 1233 women were eligible for the study, 830 (67%) agreed to participate. Babies were included in the study if they were singletons and had no major congenital anomalies; 663 of 674 babies met the inclusion criteria. Of the 663, 41 were born to women with gestational diabetes. In 2007, 539 (81%) children attended for follow-up at 9.5 years (56 refused, 8 were not traced, 26 moved away, 25 died and nine were withdrawn from the study for medical reasons) Dietary data were available for 538 children. This study was conducted according to the guidelines laid down in the Declaration of Helsinki and all procedures involving human subjects were approved by the Holdsworth Memorial Hospital Ethics Committee.

Key messages
• There are few data on the dietary patterns of Indian children. Given that diet in childhood may affect chronic disease risk, we investigated the diet patterns of 9.5 year old South Indian children. • Two diet patterns were identified: 'snack and fruit' and 'lacto-vegetarian'. Both patterns were associated with markers of nutritional status. • Knowledge of these patterns may be useful for designing targeted interventions to improve nutritional status and for developing food based dietary guidelines in this population.

Procedure
Data on maternal parity and religion were collected during pregnancy. Parental education, occupation, socio-economic status and family type were recorded at 9.5 years. In terms of education, parents were categorised as having completed: fewer than 10 years education (below secondary level); 10 years education (secondary level); or more than 10 years education (above secondary level). Occupations were classified as follows: professional, e.g. university teacher, chemist, lawyer; skilled, e.g. goldsmith, carpenter, police constable; unskilled, e.g. labourer, vegetable seller. Socio-economic status data were collected using the Standard of Living questionnaire from the Indian National Family Health Survey (International Institute for Population Sciences (IIPS) 2000).The respondent was the child's parent or guardian. Children were classified as living in an urban or rural area based on their address at 9.5 years. We defined towns with a population greater than 100 000 as urban areas based on Indian Government census data (Government of India 2012). We used data collected at the time of pregnancy to determine the proportion of the cohort that had migrated from rural to urban areas or vice versa. As close as possible to the age of 9.5 years, children were asked to visit the research centre at the Holdsworth Memorial Hospital, accompanied by their parents. A food frequency questionnaire (FFQ) was administered by one of three trained nutritionists. Child and maternal anthropometry, and collection of blood samples, were carried out on the same day.

Measurement of Blood Nutrient Concentrations
We chose to measure plasma folate and vitamin B12 concentrations in the children as there has recently been significant interest in the status of these nutrients and later risk of chronic disease (Yajnik et al. 2006(Yajnik et al. , 2008Christian & Stewart 2010). In addition, a large proportion of the Indian population are vegetarian and there is some evidence of vitamin B12 deficiency in certain settings (Yajnik et al. 2006;Pathak et al. 2007;Krishnaveni et al. 2009). Parents and children were asked in a neutral manner whether the child had consumed any food or drink other than water in the 12 h prior to the appointment, if the child was not fasted an alternative appointment was made for another day. Fasting venous blood samples (15 mL) were collected. Plasma was separated and stored at -80°C prior to analysis. Plasma concentrations of folate and vitamin B12 were determined by microbiological assay (Kelleher et al. 1987;Horne & Patterson 1988) at the Diabetes Unit, KEM Hospital Research Centre, Pune, India. Intra-and inter-assay coefficients of variation (CV) were <8% for both assays.

Dietary Assessment
A trained nutritionist interviewed the child and a parent or guardian (usually the mother) in order to collect dietary data; both child and adult provided information. A 136-item FFQ (see Appendix S1) was developed based on responses to 24 h recalls administered to children in the cohort at 8 years. All food and drinks reported were listed and local nutritionists added any foods that had not been reported during the recalls but were thought to be consumed by the study population. In order to help respondents to conceptualise the child's diet (Cade et al. 2002), the FFQ foods were then divided into 15 categories (beverages; fruit; dried fruit and nuts; rice foods; wheat foods; finger millet (ragi) foods; cooked vegetable dishes; salad; meat/poultry and eggs; jam/chutney; sugar added to foods; savoury snacks; sweet snacks; fast food; milk/ milk products). The reference time period for the FFQ was a typical month and the response categories were daily, weekly or monthly with participants stating on how many occasions per day, week or month they consumed the food item. To differentiate between children who were consuming six varieties of seasonal fruit all year round and those who were only consuming them when in season, there were two items on the FFQ for each fruit corresponding to each scenario. When it was reported that fruit was consumed seasonally, a correction factor was applied to the frequency to take account of the number of months of the year that the fruit was available, e.g. mangos are generally available for a quarter (3 months) of the year so the frequency was multiplied by 0.25. The season in which the FFQ was administered was recorded. The Indian Government Meteorological Department season classifications were used: Winter, January and February; Premonsoon, March, April, May; Monsoon, June, July, August and September; Post-monsoon, October, November and December (available at: http://www. imd.gov.in/doc/termglossary.pdf).

Anthropometry and Bio-impedance Measurements
Height was measured to the nearest 0.1 cm using a microtoise wall-mounted stadiometer (CMS Instruments, London, UK). Weight was measured to the nearest 100 g using an electronic digital weighing scale (Salter, Kent, UK). Body mass index (BMI) was calculated as weight divided by the square of height (kg m -2 ). Weight and BMI were compared with WHO international child growth standards (World Health Organisation 2007). Head, chest, midupper arm and abdominal circumferences were all measured to the nearest mm using anthropometric tape. Subscapular and triceps skinfold measurements were made in triplicate using Harpenden callipers (CMS Instruments) and the mean was used in the analyses. All measurements were made on the left side of the body. There were five measurers and mean intra-observer CVs were 1.9% for triceps and 4.0% for subscapular skinfolds; mean inter-observer CVs were 6.8% for triceps and 7.7% for subscapular skinfolds.
Whole body impedance at 50 kHz was measured using a Quadscan 4000 analyser according to the manufacturer's instructions (Bodystat, Isle of Man, British Isles). Impedance (W) at 50 kHz and percentage body fat, based on the pre-programmed Bodystat equation for children, were recorded. The CV for impedance measurements was <1%.

Diet patterns analysis
The 136 foods on the FFQ were condensed to 52 food groups based on nutrient content and typical use, for example puffed rice and rice flakes were grouped as 'processed rice foods' and cabbage palya (cooked vegetable preparation served dry), green leafy vegetable palya and green leafy vegetable curry were grouped as 'green leafy vegetable dishes'. PCA without rotation was used to identify the children's diet patterns. PCA is a statistical technique that has been widely used in diet patterns research (Hu 2002;Newby & Tucker 2004). It identifies foods that are consumed together and produces new variables (components) that are independent linear combinations of the dietary variables accounting for maximum variance (Joliffe & Morgan 1992). Each component can represent a particular diet pattern. Pattern scores were calculated for each child based on the weekly frequency of consumption of items from the food group and the coefficient for that food group. Weekly frequencies were calculated by multiplying daily intakes by 7 or by dividing monthly intakes by 4. These values were summed for all 52 food groups to provide scores that represent the child's adherence to a particular pattern. Pattern z-scores were then calculated with a mean (SD) of 0 (1). The number of diet patterns described in the results section was based on the identification of a break in the scree plot, the component eigenvalue being >2 and interpretability of the pattern for this population. Food groups with coefficients >|0.2| were considered to be discriminatory.

Correlate and outcome variables
Variables that were not normally distributed were transformed by taking natural logarithms. Differences in mean pattern scores between boys and girls were examined using t-tests. Analyses of variance (ANOVAs) and correlation coefficients were used to assess univariate associations between predictor variables and pattern scores. Correlate variables were season of the year that the FFQ was administered, socio-demographic characteristics and maternal BMI. All variables significantly associated with pattern score in the univariate analyses were entered into multivariate regression models to identify independent correlates of pattern scores. Based on the findings, post hoc two-factor ANOVAs were used to test for interaction. The associations between pattern scores and nutritional status of the child including height, BMI, folate and vitamin B12 concentrations, and body composition were assessed using the Wilcoxon rank-sum test for trend across all four quarters of pattern score.
All data were analysed using Stata version 11 (Stata Corporation, College Station, TX, USA). Table 1 shows descriptive and anthropometric data relating to the 538 children (254 male) studied, together with parental and family characteristics. Children were on average 2 SD lighter and had a median BMI approximately 1 SD lower than the WHO international reference population (World Health Organisation 2007). Almost two-thirds of mothers and fathers had at least 10 years education, half of the mothers had a BMI >25 kg m -2 . The proportion of families that were Muslim was considerably greater than the Indian national survey figure of 12.5% (International Institute for Population Sciences (IIPS) and Macro Interntional 2007). The proportion of nuclear families was almost identical to the survey figure of 60.5%.Three quarters of the families lived in urban areas. The majority (91%) of children were living in the same setting as they had been since birth. Of the 50 children whose status had changed, 32 had moved from rural to urban areas and 18 had moved from urban to rural.

Diet Patterns
We carried out separate PCA for girls and boys and found minimal differences in the patterns (data not shown) therefore, the data presented relate to the sexes combined. Based on the scree plot and eigenvalues, we examined the first three components to assess interpretability. The first component, which explained 9.1% of the variance, was named 'snack and fruit' and was characterised by high intakes of snacks, fresh fruit, sweetened drinks, rice and meat dishes (biryani), noodles and leavened breads. The second component explained 7.5% of the variance and was named the 'lacto-vegetarian' diet as it was characterised by high intakes of finger millet (ragi), traditional rice dishes, yoghurt (curd), vegetable dishes, sugar added to foods and a low frequency of meat consumption. The third component explained less variance (4.1%). Its key characteristics were high intakes of malt-based hot drinks and low intakes of tea and coffee but there was no clear pattern in the remaining food groups (data not shown). It was not considered to provide a meaningful or interpretable pattern of foods in the context of chronic disease risk and was not included in any further analyses. Table 2 shows the coefficients for each of the 52 food groups relating to the first and second components of the PCA. Consumption frequencies of the discriminating foods for each pattern (highlighted in bold in Table 2) for children with scores in the top and bottom quarters of the distributions are shown in Table 3. For the majority of these foods, there was a two-to ninefold difference in median frequency of intake between those in the lowest and highest quarters. The foods with the greatest differences for the snack and fruit pattern were sweetened drinks and fruit, while for the lactovegetarian pattern they were finger millet (ragi) and yoghurt (curd).

Snack and Fruit Pattern
Univariate analyses showed that the snack and fruit pattern was predicted by season of interview with the pre-monsoon and monsoon seasons being associated with higher scores (Table 4). Muslim religion, living in an urban area, the head of the household having an unskilled occupation, and being part of a nuclear family were all significantly associated with scores on this pattern. The multivariate analysis suggested that season, religion and dwelling were all significant independent correlates of scores (Table 5). Season was the strongest independent correlate and explained half (13%) of the variance in pattern scores. The association between family type and pattern score was of borderline significance.

Lacto-vegetarian Pattern
Season of FFQ administration was associated with the lacto-vegetarian pattern, with those interviewed during the pre-monsoon season having higher scores (Table 4). Maternal BMI was negatively associated with pattern scores while higher scores were predicted by being of Hindu religion, rural dwelling, greater duration of paternal education and skilled occupation of the head of household. In the multivariate model (Table 5), religion was the strongest correlate of pattern scores with maternal BMI, dwelling and season all being significant independent correlates. However, for this pattern season accounted for only 1% of the variance explained by the model. Religion alone accounted for 16% of the variance. There was an apparent paradox whereby the children with higher scores on the lacto-vegetarian pattern were more likely to live in rural areas but the occupation of the head of the household was more likely to be professional. We therefore explored the associations between standard of living and pattern scores in the urban and rural children separately and found that there was a positive association between standard of living and pattern score within the urban group of children although the effect size was small (correlation coefficient = 0.165, P = 0.002) and no effect in the rural group (correlation coefficient = -0.01, P = 0.881). Results of a post hoc test for interaction showed that there was a significant interaction between standard of living and urban/rural dwelling (P = 0.033).

Associations between diet patterns and the child's nutritional status
Snack and fruit pattern scores were not related to the children's height but were negatively related to their adiposity as measured by BMI, subscapular skinfolds and body fat percentage (Table 6). There were no significant associations with plasma folate or vitamin B12 concentrations. In contrast, the lacto-vegetarian pattern was not associated with adiposity, but pattern scores were positively associated with plasma folate concentrations, and negatively with vitamin B12 concentrations.

Discussion
To our knowledge, this is the first study in India to examine associations between diet patterns and demographic and body composition variables in a large number of children. We identified two main diet patterns. Variability in adherence to the diet patterns reflected biologically important differences in food intakes (Table 3). Pattern scores were predicted by the children's socio-demographic and background characteristics, and were associated with measures of adiposity and nutritional status.

Snack and Fruit Pattern
Many of the foods that characterised this pattern, including sweetened drinks, branded noodles and leavened breads, were shop-bought items and few were home-cooked foods. The demographic factors that were associated with this diet pattern were urban dwelling and being part of a nuclear family. This suggests that the pattern may have been related to a more modern style of family life. The snack and fruit pattern was also associated with being Muslim which was unexpected. It possibly reflects the fact that the Muslim religion prescribes fewer dietary restrictions than the Hindu religion. Scores for the snack and fruit pattern were inversely related to BMI and fat mass. This finding is not consistent with the increase in childhood adiposity occurring in India in conjunction with urbanisation (Misra et al. 2007). The snack and fruit pattern comprised healthy and unhealthy aspects, with more micronutrients from fruit but high levels of fat and sugar. Although sugar-sweetened drinks were a discriminatory item for this pattern, intakes were low compared with typical Western populations (Harrington 2008;Hafekost et al. 2011). It is conceivable that if the nutrition transition Table 3. Monthly frequency of consumption of discriminating foods for each pattern by children whose pattern scores fall in the lowest and upper most quarters of the distribution* continues, consumption of such drinks will increase. A recent study assessing the effect of rural to urban migration in India found that urban dwelling adults consumed up to 80% more fruit than their rural counterparts (Bowen et al. 2011). It is thought this is due to wider availability of produce and greater purchasing power among urban dwellers. Urban dwellers also tended to have a higher energy and fat intake.

Lacto-vegetarian Pattern
This pattern represents a traditional South Indian lacto-vegetarian diet. Children adhering to this pattern were more likely to be Hindu and rural dwelling and have mothers with a low BMI. Our findings suggest that there were two groups of children with high scores on this pattern: rural children who consumed this diet because it was readily available and relatively inexpensive; higher caste Hindus who lived in the city, had a professional head of household, and consumed this diet because of tradition and religious observance. The finding that there was a weak but statistically significant association between standard of living index and lacto-vegetarian pattern scores among children who lived in urban areas may support this suggestion. The lacto-vegetarian pattern was also strongly positively related to plasma folate concentrations and inversely related to plasma vitamin B12. This is consistent with high vegetable consumption and low animal food intake (Joint FAO/WHO Expert Consultation on Human Vitamin and Mineral Requirements 2004). Season of FFQ administration was a predictor of pattern scores but explained a small proportion of the variance. It is possible that some of the vegetables that are characteristic of this diet are less widely available in winter months or that religious festivals are associated with changes in diet at certain times of the year.

Associations between diet patterns and demographic characteristics
We are not aware of any data from South Asia on children's diet patterns and demographic variables. A study in urban Pakistani men of low socio-economic status found that education and income were positively associated with 'high meat diet' scores and that a 'prudent' diet (characterised by a high intake of eggs, fish, raw vegetables and fruit) was associated with being better educated (Yakub et al. 2010). In addition, a large cross-sectional study in rural parts of India, showed that low fruit and vegetable intake was associated with low socio-economic status (Kinra et al. 2010). Maternal education is positively associated with adherence to a healthy diet pattern in Australia and the UK (Ambrosini et al. 2010;Cribb et al. 2011). Our data do not show such a relationship. This may be due to less variability in maternal education in this population or the fact that we did not identify a specifically 'prudent' or 'healthy' diet pattern in this cohort.

Associations between diet patterns and child body composition
Several studies in Western populations have shown inverse associations between healthy diet patterns and measures of childhood obesity including waist circumference and overweight (Ritchie et al. 2007;Lioret et al. 2008). A study in Bengalee women aged 35 years and above found an independent positive association between adherence to a diet characterised by high intakes of hydrogenated and saturated fat, and BMI and waist circumference (Ganguli et al. 2011). In Pakistani men, there was a positive association between a 'high meat' diet and waist hip ratio (Yakub et al. 2010). The negative association between adiposity and the snack and fruit pattern requires explanation. It is possible that the diets of urban dwelling children in this study have only recently started to contain foods with higher fat and sugar and that the effects on body composition of these foods will become evident as the children grow older. We will be able to address this question in the continued follow-up of these children.

Associations between diet patterns and child folate and vitamin B12 status
Among Pakistani adults, 'high meat' and 'Western' diet patterns were associated with raised homocysteine levels (Eilat-Adar et al. 2009;Yakub et al. 2010). We found that following a lacto-vegetarian diet was positively associated with folate status, this is similar to a 'high vegetable' diet in the Pakistan study. The negative association with vitamin B12 levels may be a particular concern for the rural poor who may have limited access to animal foods. The lack of association between these nutrient biomarkers and the snack and fruit pattern might be expected as variations in consumption of many of the foods that characterised this pattern would not contribute to significant differences in intakes of these nutrients.

Strengths and limitations
We used an FFQ that was developed for the study. The retained patterns together explained 16.6% of the variation in the 52 food groups. The proportion of variance explained is dependent on the number of variables entered into a PCA and the number of components retained (Newby & Tucker 2004;Crozier et al. 2006). This makes it difficult to compare diet pattern studies directly. However, in other diet patterns analyses of European data, where similar numbers of input variables have been used a comparable proportion of variance was explained by two patterns (Crozier et al. 2006;Robinson et al. 2007;Fisk et al. 2011). Although FFQs may be subject to measurement error, patterns defined using FFQ data have been shown to be comparable to patterns defined using other assessment methods, and pattern scores from different methods are highly correlated (Hu 2002). Furthermore, a number of studies have shown associations between diet patterns and chronic disease outcomes (Fung et al. 2001;Osler et al. 2001;Miller et al. 2010), highlighting patterns analysis as a potentially useful technique to gain a greater understanding of the importance of diet behaviour. The multivariate models (Table 5) explained less of the variance in snack and fruit pattern scores compared with the lacto-vegetarian pattern. This may be due to variability in adherence to the snack and fruit pattern being associated with unmeasured factors such as family income, taste preferences, marketing and social desirability.
A strength of our data is that we assessed folate and vitamin B12 status. The association between folate and vitamin B12 concentrations and the lactovegetarian pattern scores was as expected, and adds strength to the validity of our dietary data. It is conceivable that activity could have been associated with both adiposity and the children's dietary patterns. We were not able to assess such relationships at this age. However, data collected when the children were younger (mean, SD age 7.8, 1.1 years) indicated that the children were relatively inactive and there was little variability in activity between children (Kehoe et al. 2012).
The absence of an obviously 'healthy' or 'prudent' pattern in our analysis is of interest and is in contrast to many analyses conducted in the United States and Europe (Newby & Tucker 2004). As with many countries, it has been observed that over recent years, there is increasing prevalence of advertising for high sugar and high energy foods aimed at children in India (de Cruz 2004;Kaur & Singh 2006;Consumers International 2008). It will be important to determine how diet patterns are affected by such influences as a result of the nutrition transition.
In conclusion, we have shown that meaningful diet patterns can be identified in this population and these patterns are associated with demographic and body composition variables in Indian children. These findings may be helpful in developing food-based dietary guidelines in this population and also in designing interventions aimed at prevention of overweight and other risk factors for chronic disease. As we follow this cohort, we will use these patterns to study associations between diet and risk factors for chronic disease in childhood and later life.