Nearly one quarter of children from resource-rich countries are already overweight or obese by age 5 years,1^–3 suggesting that preventing obesity needs to begin in early life. However, there is little evidence on how to prevent obesity in preschool children,4^–6 so identification of modifiable risk factors is essential to inform the development of effective interventions.7

Researchers and policy-makers have advocated an ecological systems approach to addressing obesity rather than focusing on risk factors in isolation.8^–10 In this approach, obesity is conceptualised as being influenced by factors across multiple levels: individual and family risk factors as well as characteristics of the community and region. The UK Government Office for Science published the Foresight Tackling Obesities report with a systems map illustrating the determinants of obesity across these levels and their interrelations.9 However, few studies have the breadth and depth of information to investigate the relationships between obesity and individual-, family-, community- and area-level factors. Furthermore, to our knowledge, this approach has not been used to examine obesity in young children.

The Millennium Cohort Study (MCS) is a nationally representative, contemporary cohort of British children followed since infancy and provides an opportunity to simultaneously assess risk factors for early childhood overweight using an ecological systems approach. Data from this ethnically and socioeconomically diverse cohort were used to examine individual-, family-, community- and area-level risk factors for overweight (including obesity) in children age 3 years.

METHODS

Participants

The MCS is a prospective study of UK children born in the new millennium. Families were invited to participate if they were eligible for Child Benefit (a universal benefit for families with children) and resident in England, Wales, Scotland and Northern Ireland when the child was aged 9 months.11 The MCS employed a stratified clustered sampling design to over-represent children living in disadvantaged areas and from ethnic minority groups. Additional information on the sampling framework has been reported previously.12 The original sample consisted of 18 819 children (18 553 families), born between September 2000 and January 2002, with a mean age of 9.2 months (standard deviation (SD) 0.5) at the first contact (response 72%). Eighty per cent (14 630/18 296) of the singleton infants participated in the second contact, which took place between September 2003 and January 2005, when the children were mean age 37.7 months (SD 2.5). Attrition was highest among families from Northern Ireland, in electoral wards in England defined as “ethnic” (based on the 1991 Census, if at least 30% of residents were from an ethnic minority group) and “disadvantaged” wards from all UK countries (above the upper quartile of the Child Poverty Index).13 At both contacts, main respondents (over 99% were natural mothers) and their partners (of those interviewed, over 99% were natural fathers) were interviewed in the home. Data were obtained from the UK Data Archive, University of Essex. The MCS received ethics approval from the South West and London Multi-Centre Research Ethics Committees for the first and second contacts respectively.14

Among the 14 630 singleton children, 13 188 had data available for analysis. Families were excluded if the main respondent was not female (185), the partner respondent was not male (132), there were two singleton cohort children from the same family (10) or the child had a missing height or weight (802) or a height-for-age, weight-for-age or body mass index (BMI; weight/height2)-for-age z-score ⩽–5 or ⩾5 (467). Some participants satisfied more than one exclusion criterion. Children were more likely to have missing or implausible BMI data if they were from an ethnic minority group, lower income family or if their mother was a lone parent, had a lower academic qualification or lower socioeconomic circumstances (p<0.01).

Outcome measure

At the second contact, trained interviewers measured the children’s weight and height without shoes or outdoor clothing. The children were weighed using Tanita HD-305 scales (Tanita UK Ltd, Middlesex, UK), recorded to the nearest 0.1 kg, and height was measured using Leicester Height Measure Stadiometers (Seca Ltd, Birmingham, UK), recorded to the nearest 0.1 cm. Childhood overweight (including obesity) was defined by the International Obesity TaskForce cut-offs for BMI, which are age and sex specific.15

Potential risk factors

Potential risk factors were chosen based on prior evidence,10 16^–19 including previous research on early childhood obesity in the MCS. Overall, 26 risk factors were examined across the following levels (table 1): individual (6), family (13), community (5), area (2). All individual-, family- and community-level risk factors were reported by parents, usually the mother. Area-level risk factors were derived from the child’s address at the first contact. Definitions of risk factors are presented in table 1, while more detailed descriptions are published elsewhere.17^–19 Only those factors not previously defined are described here. Birthweight z-scores were calculated using the British 1990 growth reference,20 and adjusted for gestational age and gender. Parental height and weight at the first contact was collected by self-report. Children were grouped as having neither parent overweight (BMI <25 kg/m2) or their father, mother or both parents overweight (BMI ⩾25 kg/m2). At the first contact, mothers were asked about their smoking habits immediately before pregnancy, whether they changed during pregnancy and, if so, the month they changed. Maternal smoking during pregnancy was defined as the number of cigarettes smoked daily at 3 months’ gestation. Mothers also reported whether the cohort child was their first live born or not (parity). Ward type was based on the grouping of electoral wards (as established in 1998) as defined for the MCS sampling framework. Wards that were not “ethnic” or “disadvantaged” (previously defined) were considered “advantaged”. In Wales, Scotland and Northern Ireland, there was no “ethnic” stratum.12

Statistical methods

All analyses were conducted using STATA statistical software, version 9.2 SE (Stata Corporation, Texas, USA), with survey commands to account for the clustered sampling design and obtain robust standard errors. Weighted percentages were derived and regression analyses were conducted using survey and non-response weights to account for the clustered sampling and attrition between contacts. p-values were calculated by an adjusted Wald test.

Unadjusted logistic regression analyses were conducted to examine the relationships between individual-, family-, community- and area-level factors and early childhood overweight (table 1). Factors significant at the p⩽0.05 level were included in the adjusted models in a stepped approach. Model 1 included individual-level risk factors only. In Models 2–4, family-, community- and area-level risk factors were added sequentially. Only one interaction was examined, specified a priori, between “child’s gender” and “which parent overweight”. The interaction was not significant in the unadjusted analyses and not included in the adjusted models.

RESULTS

At age 3 years, 18.0% (2410) of children were overweight and an additional 5.0% (691) were obese. As illustrated in table 2 (first column), the MCS represents a diverse sample of preschool children in modern Britain. Thirteen per cent of children were from an ethnic minority group, 14% of mothers were lone parents, and 50% of families had an annual household income of £22 000 or less.

Unadjusted analyses

Individual-level

Children were more likely to be overweight if they had a greater birthweight z-score, were black (compared with white children), introduced to solid foods before 4 months (compared with after 4 months) or watched at least 1 hour of television daily (compared with less than 1 hour) (table 2). Children were less likely to be overweight if they were Indian or Pakistani or had ever been breastfed (compared with never breastfed). There was no association between early childhood overweight and the child’s gender.

Family-level

Children were more likely to be overweight if their mother was a lone parent (compared with non-lone mothers), smoked 1–19 cigarettes daily during pregnancy (compared with none), both parents were overweight or obese, their mother only, or their father only (compared with neither), their mother was overweight prepregnancy (compared with normal weight) or worked at least 21 hours each week (compared to never worked) (table 2). There was a tendency for children from more advantaged groups (measured by maternal socioeconomic circumstances, education and household income) to be at lower risk for overweight (p = 0.1 level). There were no associations between early childhood overweight and maternal age at first live birth or MCS birth, parity, number of children in the household or whether the child has meals at regular times.

Community-level

Children were more likely to be overweight if their mother reported that there was no access to a garden (compared with those who did have access) (table 2). There were no consistent patterns between early childhood overweight and access to food shops, neighbourhood conditions, satisfaction with the area or access to places where children can play safely.

Area-level

Children were more likely to be overweight if their family lived in Wales or Northern Ireland (compared with England) or in a disadvantaged ward (compared with advantaged); however, children were less likely to be overweight if their family lived in an “ethnic” ward (table 2).

Adjusted analyses

Results from the stepped analysis are presented in table 3. After mutual adjustment for individual-level factors (Model 1), early childhood overweight was directly associated with birthweight, black ethnicity, early introduction of solid foods and television use for at least 3 hours daily. Protective factors for early childhood overweight were Indian ethnicity and breastfeeding. All relationships, except television viewing, were maintained after adjustment for family- (Model 2), community- (Model 3) and area-level (Model 4) factors.

After mutual adjustment for individual- and family-level factors (Model 2), early childhood overweight was directly associated with lone motherhood, smoking during pregnancy (1–19 cigarettes daily), parental overweight, prepregnancy overweight and maternal employment (21+ hours/week). These relationships were maintained after adjustment for community- (Model 3) and area-level (Model 4) factors.

The relationship between early childhood overweight and access to a garden was not maintained after it was added in Model 3 or after adjustment for area-level factors (Model 4). After area-level factors were added in Model 4, early childhood overweight was only associated with residence in Wales.

Forward and backward stepwise logistic regression analyses were conducted to check the validity of the model from the stepped analysis. All significant factors from the fully adjusted stepped model were retained in the final forward and backward models, demonstrating agreement between the three approaches (data not shown).

DISCUSSION

Among a nationally representative cohort of contemporary British preschool children, 23% were overweight or obese at age 3 years. We found that primarily individual- and family-level risk factors were associated with early childhood overweight, including birthweight, black ethnicity, early introduction of solid foods, lone motherhood, smoking during pregnancy, parental overweight, prepregnancy overweight and maternal employment of 21+ hours/week. Protective factors for early childhood overweight were Indian ethnicity and breastfeeding for at least 4 months. Although there was limited evidence for relationships between community-level factors and early childhood overweight, children from Wales were more likely to be overweight than children from England.

The MCS provided the breadth and depth of social and health information on children, their families and their environment to simultaneously examine risk factors for early childhood overweight across multiple levels. The MCS is longitudinal so we have been able to examine the cumulative influence of risk factors, such as infant feeding practices or maternal employment patterns, on later overweight. As some factors are likely to be on the causal pathway in the development of overweight, Buchan and colleagues suggest that unadjusted relationships should not be disregarded even if adjusted relationships are no longer significant.7 A stepped analysis has allowed us to examine the impact of sequential adjustment on risk factors at each level, reflecting an ecological systems approach to addressing early childhood overweight. Few studies have used this approach. Although the forward and backward stepwise regression models identified the same risk factors as the ecological systems approach, they would not have revealed factors that lost significance after adjustment—potentially missing modifiable risk factors on the causal pathway.7

There are limitations to the MCS data. No measures of body composition were collected in the MCS, and BMI cannot discriminate between lean and fat mass. Although BMI is a reasonable proxy for body fatness at a population level, particularly at the higher end of the distribution,21 it is less reliable when comparing children from different ethnic groups.21 22 Although information was collected on the children’s television viewing habits at age 3 years, little else is known about their physical activity and dietary patterns. Furthermore, some observations, such as television viewing, have been only collected at one time point, so the temporality of the associations cannot be determined. These factors can be re-examined at subsequent sweeps when the MCS children are of school age. The questions about community-level factors were not developed for research on childhood obesity. There may be limitations with the questions themselves or mothers’ interpretation of the questions. At the first contact, mothers were asked about health behaviours during pregnancy and early postpartum. Recall bias may have influenced the information collected. Although tobacco consumption during pregnancy is often under-reported,23 recall of breastfeeding practices has been found to be reliable and valid.24 As we found that smoking during pregnancy increased the risk of early childhood overweight, under-reporting of smoking is likely to underestimate the strength of the association. At both contacts, parents reported their height and weight. Although research suggests that misreporting of height and weight could lead to underestimations of obesity prevalence,25 estimates of parental overweight (including obesity) in the MCS are similar to those reported in a national survey of adults in England.26 Any misreporting of height and weight is likely to underestimate the strength of the association between parental body size and early childhood overweight.

A review of policy-relevant risk factors for early childhood overweight identified individual-, family- and community-level factors, but found that few studies examined risk factors across multiple levels.16 The individual- and family-level risk factors we identified are consistent with the published literature on early childhood overweight16 as well as on risk factors for overweight in older children.10 Previous studies have also reported that an increasing birthweight is associated with overweight in young children and across the life course.10 27 28 There has been limited research on ethnic differences in overweight among British children, particularly young children. At age 3 years, we found that black children were more likely to be overweight than white children, while South Asian children were less likely to be overweight. However, research has shown that South Asian children exhibit greater insulin resistance, a precursor to cardiovascular disease, at similar adiposity levels in white children.29 Deurenberg and colleagues have found that, for the same BMI, children from different ethnic groups have dissimilar percentages of body fat.22 This suggests that ethnic variation in overweight (including obesity) may not represent true differences in body composition.

There is evidence that breastfeeding protects against overweight across the life course,30 which is consistent with the relationship in the MCS. However, a study by Toschke and colleagues found limited support for an association between breastfeeding and fat mass, except that children breastfed for at least 6 months had lower total and trunk fat masses in the top decile compared with children never breastfed.31 While there are less consistent relationships with the introduction of solid foods,16 28 we found an increased risk of overweight with early weaning. We also found that high levels of television viewing were associated with overweight, consistent with other studies in young children.16

There is evidence for a socioeconomic gradient in childhood overweight, with children from lower socioeconomic circumstances at higher risk.32 We found some differences by socioeconomic factors in unadjusted analyses, but only lone parenthood was significant in the final model. Our results support previous research indicating that parental overweight increases the risk of overweight in their offspring across the life course.27 28 We also found that prepregnancy overweight was associated with early childhood overweight, consistent with other studies,16 and independent of parental overweight postpartum. There is an increasing body of research indicating that smoking during pregnancy is associated with an increased risk of early childhood overweight.16 33 We also found support for a dose–response relationship with cigarette consumption (test for trend in the final model, p<0.001), which has been reported previously.16 33 The relationships with prepregnancy overweight and smoking during pregnancy suggest a possible influence of programming in utero.33 34 Maternal hours worked per week was also found to predict early childhood overweight. Support for this relationship has been reported primarily in older children.35

Few other studies have examined community-level factors and early childhood overweight.16 The lack of relationships with community-level factors could be due to limitations of the data collected; however, other research in preschool children has also found no evidence for relationships between different community-level factors and overweight.16 In contrast, studies in school-age children have reported associations between community-level factors and overweight,36 suggesting that the environment may influence weight gain when children are older. Although little is known about country and regional differences in early childhood overweight, children in the MCS from Wales and Northern Ireland were found to be at higher risk for overweight than children from England. One other study has reported variation in early childhood overweight by state of residence in Australia.3

Implications for policy and practice

Our results provide additional evidence that obesity prevention needs to begin in early life. When an ecological systems approach was applied to examining overweight in young children, individual- and family-level risk factors were found to be more influential than community- and area-level factors. Our findings suggest that obesity prevention should focus on supporting parents, particularly in promoting maternal health behaviours during pregnancy and postpartum. There are a limited number of interventions on preventing early childhood obesity, and most have targeted dietary, physical activity and sedentary behaviours during the preschool years.4^–6 Wen and colleagues are conducting a randomised controlled trial of home visits over the first 2 years of life with the aim of preventing overweight in preschool children.37 While this novel study addresses gaps in the evidence base, additional interventions focusing on early factors are needed.

As parental overweight is one of the strongest predictors of early childhood overweight as well as overweight across the life course,28 supporting parents to maintain a healthy weight is an important component of obesity prevention. The US has a target to reduce the number of overweight adults,38 and addressing adult obesity is a component of the UK Government’s obesity strategy.9 39 Furthermore, maternal overweight prepregnancy is increasing in the US40 and the UK.41 The recently published cross-government obesity strategy for England proposes that obesity prevention will be integrated into routine health visits to identify at-risk families.39 During the first antenatal visit, health professionals will give advice to overweight mothers on healthy weight gain during pregnancy. Furthermore, health visits during infancy will focus on promoting a healthy weight in early life.

An ecological systems approach to tackling obesity requires action across government departments. This is well illustrated by action on infant feeding. It has been suggested that improving breastfeeding rates could significantly reduce the number of cases of childhood obesity.42 There is international support to increase breastfeeding rates,38 43 with the aim for mothers to exclusively breastfeed for 6 months.44 The WHO Child Growth Standard promotes breastfeeding by monitoring infant growth with a chart developed from exclusively breastfed infants.45 The cross-government obesity strategy for England endorses the adoption of the WHO Child Growth Standard as well as an information campaign to promote breastfeeding.39 Legislation has been proposed in England46 and enacted elsewhere47 48 to protect breastfeeding in public places and/or at work. In the UK, government efforts to increase paid parental leave49 and promote work/life balance50 may help to support infant feeding recommendations.44

Most risk factors for early childhood overweight, including parental overweight and maternal smoking during pregnancy, are modifiable or would allow at-risk groups, such as lone mothers or overweight women early in pregnancy, to be identified. Improving modifiable risk factors could reduce childhood obesity at the population level.42 51 As the short- and long-term consequences of obesity are a significant burden to society,9 even a small decrease in the prevalence of obesity at the population level is worth achieving. However, modifying health behaviours requires action at many different levels. Policies at the community and regional levels can help to create an environment that supports healthy behaviours for parents and their children.