Methods

Study population

The Health Survey for England (HSE)13 14 draws annually a nationally representative random sample of the free-living general population using multistage stratified probability sampling. Up to two individuals aged 2–15 were selected randomly from each household. In this study, data from the 1995–2007 surveys were used. Response rates ranged between 85% in 1997 and 66% in 2004 and 2007. Children were defined as aged 2–10 and adolescents as 11–18 years.

Measurements

The weight and height measuring methodology has been described in detail elsewhere.10 13 14 Body Mass Index (BMI) is calculated as weight (kg) divided by squared height (m²). Obesity rates were computed using the age- and sex-specific international classification BMI standards.15 The occupation of the head of the household was used to determine socioeconomic status. The Registrar-General's classification was used to classify the sample into manual and non-manual social classes.16

Statistical analysis

The prevalence of obesity for 1995–2007 was calculated for each year separately, by sex, age group (2–5, 6–10, 11–15, 16–18 years) and social class (manual/non-manual); moving averages of prevalence were plotted for each group. Moving averages were calculated as the average of each year and its two neighbours, for example the moving average for 2003 was the average of 2002, 2003, and 2004. The only exceptions were 1995 and 2007, the moving averages of which were calculated as the average of each year with its neighbour, that is 1996 and 2006 respectively. A linear trend was fitted to the prevalence data (x₁,x₂,…,x₁₁), for the time period 1995–2007. The value of the series was then projected p years ahead, where p = 8 (corresponding to the year 2015) by extrapolating a linear trend. For all projected rates, the 95% prediction intervals were also calculated as follows: y ± (t_n-2 * standard error). The standard error is given by:

The assumption underlying the linear projections is that the time trends over the examined period will continue unchanged. An alternative set of projections was also made to allow for acceleration or slowing down in the rate of change in prevalence of obesity. For this set of projections, power (of formwhere t  =  year-1994) and exponential curves (of form ) were fitted to the data. For presentation in the graphs, the best fitting curve between power and exponential was chosen based on the R². In summary, the two scenarios are: a) Scenario 1: projections are based on the linear trend observed in 1995–2007 and assume that the rate of increase will remain constant till 2015; b) Scenario 2: projections are based on the best fitting curve (exponential or power) to allow for acceleration or deceleration in obesity prevalence trends.

All analyses were run using SPSS statistical package, version13.

Each year's survey was approved by the appropriate Research Ethics Committee.

Results

Tables 1 and 2 present the obesity prevalence rates from 1995 to 2007 by sex and age-group. Figure 1 presents moving averages of prevalence.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Moving averages of obesity prevalence in children and adolescents living in England between 1995 and 2007.

Prevalence trends

Children aged 2–10

Overall, among boys the prevalence of obesity increased from 3.1% in 1995 to 6.9% in 2007 (table 1). The overall prevalence of obesity among girls aged 2–10 increased from 5.2% in 1995 to 7.4% in 2007 (table 2). Considering the moving averages of prevalence (figure 1), there was an upward trend up to approximately 2004/05 in all age groups. Between 2004/05 and 2007 the trend tended to stabilise (boys) or decrease (girls 6–10 years).

In relation to social class, the annual obesity prevalence was on average 0.6% (boys aged 2–10) and 1.5% (girls aged 2–10 years) higher in participants from manual than from non-manual households (figure 2).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Extrapolation of percentage obese children (aged 2–10 years) in England, by sex and social class.

Adolescents aged 11–18

Among male adolescents the prevalence increased from 2.7% in 1995 to 4.8% in 2007. Obesity rates for female adolescents aged 11–18 increased from 4.7% to 6.1%. There was an upward trend between 1995 and 2004/05 in all age groups of male and female adolescents aged 11–18. Similarly, the prevalence was higher in female than in male adolescents throughout the examined time period (tables 1 and 2).

In relation to social class, the 1995–2007 annual obesity prevalence was on average 1.2% (boys 11–18 years) and 2.1% (girls 11–18 years) higher in participants from manual than from non-manual households (figure 3). The prevalence of obesity in 1995 was 2.7% among males from non-manual classes compared with 2.9% for manual groups, increasing to 4.8% and 5.3% in 2007 respectively. For female adolescents in the non-manual social class, the prevalence of obesity was 2.8% in 1995 and increased to 5.3% in 2007, with a tendency to decline in the more recent years. For females from manual social classes the prevalence of obesity was 6.5% in 1995, increasing slightly to 6.7% in 2007, with a tendency to decline in the more recent years.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Extrapolation of percentage obese adolescents (aged 11 to 18 years) in England, by sex and social class.

Projections to 2015

Table 3 shows the projections to 2015 and figure 4 shows the linear and non-linear projection curves to 2015 for children and adolescents.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Extrapolation of percentage obese children and adolescents in England, by broad age group and sex.

Children aged 2–10

With a linear trend obesity rates in 2015 will be similar among pre-school boys aged 2–5 and boys 6–10, but the projected rate of obesity for all boys is 13.5% based on the exponential scenario compared with 10.1% based on the linear one. For girls the exponential and linear trend-projected rates of obesity are comparable.

Overall, the estimated 2015 prevalence was higher than the 2007 levels for all boys and for girls in manual social classes, regardless of forecasting method. For children in the manual social class, the projected obesity prevalence in 2015 is 10.7% (95% CI 6.6 to 14.9) for boys and 11.2% (7.0 to 15.3) for girls. For non-manual groups the projected obesity prevalence in 2015 is 7.9% (3.7 to 12.1) for boys and 5.4% (1.3 to 9.4) for girls. Assuming an exponential trend, obesity prevalence in 2015 for boys from manual social classes households will increase to 14.2%. In the non-manual groups, prevalence will be considerably lower than in manual (boys: 9.5%; girls: 5.3%). For girls from non-manual households the projected prevalence of obesity will follow a downward trend, but prevalence is projected to increase in girls from manual households.

Adolescents aged 11–18

The linear-projected rate of obesity for male adolescents is 8.0% (R² = 0.41), 9.5% in the exponential trend (R² = 0.42) and 6.9% in the power (R² = 0.54) scenarios. For females the three projected rates are comparable (R² was 0.37, 0.42 and 0.59 for the linear, exponential and power models respectively).

Overall, the estimated 2015 prevalence was higher than the 2007 levels for all boys and girls in manual and non-manual social classes, regardless of forecasting method. For adolescents in the manual social class, the linear model projected rate is 10.0% (5.2; 14.8) for males and 10.4% (5.0; 15.8) for females. For non-manual social classes the projected obesity prevalence is 6.7% (3.4; 10.0) for boys and 8.3% (4.3; 12.4) for girls. If the trend is assumed to be power, obesity prevalence in 2015 for male and female adolescents from manual social class households will increase to 8.2% and 9.1% respectively. In the non-manual groups projected prevalence will be considerably lower, at 5.6% for males and 7.9% for females.

Discussion

Prevalence trend

The present results show that there has been a marked increase in the prevalence of obesity between 1995 and 2004/2005, followed by a tendency to level off or decrease to 2007. Stabilising or reversing recent trends in overweight or obesity prevalence in European countries have been reported for children in France (3–1417 and 7–918 years), Switzerland, (6–13 years),19 and Sweden (girls 10–11 years).20 In the USA, Ogden et al showed that there was no significant change in the prevalence of high BMI for age among children and adolescents between 2003/2004 and 2005/2006 or between 1999 and 2006.21 Thus, at least in the developed countries, the obesity epidemic may be slowing down. At the individual and family level the stabilisation of the trends may be related to the considerable media attention that issues around obesity have received in the recent years,22 and the subsequent body weight awareness this may have generated. At the policy level, it could be partly the result of anti-obesity policies and strategic targets, such as the 2004 Public Service Agreement by the English government that set a target to halt the year-on-year increase of obesity in children under 1123 and the French National Nutrition and Health Programme.24

Projections to 2015

When interpreting the present results it has to be taken into account that the method used to project future prevalence of obesity makes assumptions about future changes in obesity based only on past patterns (1995–2007). In boys and girls aged 2–10, the exponential curve provided a better fit to the data than the power curve, suggesting that rates of change in obesity prevalence have been accelerating. In contrast, the power curve provided a better fit in boys and girls aged 11–18, suggesting that the overall rate of change has been decelerating. This approach may be limited because it does not attempt to forecast and take into account changes in obesity determinants (eg ethnic composition, income distribution, or behaviour, such as physical activity or content and amount of food and drink consumed) over the forecasting period. Acknowledging that factoring such parameters into the present projections would have been extremely difficult, a relatively short forecasting period was chosen.

The Foresight report,11 which used HSE 1995–2004 data, estimated the projected 2050 obesity rate to be 26% for both males and females under 20 years. Intermediate projections were 10% by 2015 and 14% by 2025, but no age- and sex-specific projections were specified. It is likely that these projections over-estimate forecast rates of obesity as they do not take into account the 2005–7 prevalence when there was a consistent tendency for stabilisation of the trend. Another obvious limitation of attempting to make long-term predictions of a largely multifactor condition, such as obesity, is the breadth of societal, economic, demographic, technological, and lifestyle changes that are likely to occur over the forecasting period. For these reasons the Foresight forecasting exercise may be less useful for intermediate-term policy and health services planning.

Social class differences

The socioeconomic gap in young obesity in favour of higher socioeconomic strata has been documented previously.10 25 26 In the present study, observed increases in obesity prevalence were more pronounced in boys and girls from manual social class households. This trend differentiation is reflected by the projections to 2015. The 2015 projected rate for girls from non-manual households is at 1995 levels (around 5%), in contrast to girls from manual households where the projected rate is over twice as high (around 11%). The social class gap is evident in boys too: the prevalence does not level off in those in manual classes, whose prevalence continued to rise year-on-year to 2007 (figure 2). The present projections indicate that by 2015 there will be a considerable prevalence gap between boys from manual and non-manual classes (rates will be higher in manual classes by approximately 35% (linear) to 50% (exponential)) and girls from manual and from non-manual classes (higher rates in manual by approximately 35% (linear) or 25% (exponential) scenario)). Among adolescents, the socioeconomic gap is pronounced among males but more subtle among females. The projected obesity prevalence for male adolescents from manual social classes is approximately 45% (power) to 50% (linear) higher than males from non-manual classes.

The widening socioeconomic gap in children's and adolescents' obesity may be partly due to difficulties to reach and communicate health messages to families from lower socioeconomic groups. Previous research shows higher socioeconomic status groups tend to follow recommendations for health behaviours27 and respond more actively to health-related media messages28 than do those of lower socioeconomic status. It is possible that the recent exponential increase in obesity and obesity-related (diet and physical activity) media messages22 has been received more positively by non-manual than by manual families. As lower socioeconomic groups tend to be wary of measures and messages aimed at changing their lifestyle because they see these as “nanny-statism” that erodes their autonomy,29 it is possible that policies targeting children's eating and physical activity habits have not have been perceived favourably by manual classes. Another possibility is that a stabilisation in obesity prevalence among children from manual households is taking longer to occur than among children from non-manual households, but it will eventually occur. However, the present data can neither support nor reject such a prediction.

Study strengths and limitations

Study strengths include the large, nationally representative samples covering 13 years, the objective measurement of weight and height, and that the sampling, recruitment, and data collection methods (including measurements) remained unchanged throughout the 13 years.

A limitation of the present study is that no adjustments were made for the trend for changes in sociodemographic factors or lifestyle behaviours that affect obesity. Although HSE is designed to be nationally representative, the authors acknowledge that the decline in response rates in recent years may have introduced respondent bias in the more recent years, for example if households refusing to take part in the survey (non-respondents) are more likely to be obese than respondents. This is possible as people from lower social classes are more likely to be both non-responders in survey research30 31 and obese.32 The decline in response rates may have resulted in under-estimating obesity prevalence in recent years and subsequently in the present projected rates. It could be argued that it is a difficult task to provide an accurate forecast of future levels of obesity, given that levels can depend so much on short-term influences. However, the purpose of the present projections is not so much to provide an accurate forecast of future levels, but to predict what might happen if trends observed in recent years were to continue.

Conclusions

The prevalence of obesity in children and adolescents has been rising over the past decade and the inequalities in children's obesity documented before10 are widening. If trends continue as they have been between 1995 and 2007, in 2015 the number and prevalence of obese young people is projected to increase dramatically and these increases will affect lower social classes to a larger extent. Thus, it is essential to implement effective strategies for the management and prevention of young obesity and the reduction of social class inequalities in health.