Results

THE DATASET

The original dataset contained the heights of 25 264 and weights of 25 222 children. Of these 25 047 were matched records (where system numbers matched). Some 24 583 (98%) records from the above, also had valid postcodes. Some of these were multiple records on the same children; first observations were kept and the subsequent ones were dropped. There were finally 20 980 complete single records. Of these, seven observations for BMI z scores were considered to be implausible (more or less than seven standard deviations (SD)). Analysis was carried out with and without these seven observations, without any significant difference in the results; however the normality of the distribution of the BMI z scores was significantly affected by the inclusion of these values. The results presented below are therefore, for 20 973 records.

DISTRIBUTION OF THE VARIABLES

Age

The median age of the children was 8.8 years (range = 5.1 to 14.6). The age quarters had means (ranges) in years as follows: 6.4 (5.1 to 7.1); 8.1 (7.2 to 8.9); 9.7 (9.0 to 11.7); 12.2 (11.8 to 14.6).

Sex

There were 10 693 boys (51%) and 10 280 girls (49%).

BMI

The BMI z scores were normally distributed with mean = 0.17 (SD=1.08).

Townsend scores

The mean Townsend score by enumeration district was 1.23 (SD=3.94). The Townsend score quarters had means (ranges) as follows: −3.3 (−5.4 to −2.1); −0.7 (−2.0 to +0.6); 2.3 (0.7 to 4.2); 6.7 (4.3 to 12.4).

DISTRIBUTION OF OBESE CHILDREN

Overall, there were 1046 obese children (5.0%). There was no significant difference in the overall prevalence of obesity between the two sexes (boys 5.2%, girls 4.8%; χ² test p=0.2). In both sexes, there was a significant trend for higher levels of obesity with age (p<0.001), the effect becoming almost double in the oldest age group, as compared with the youngest (table 1). The prevalence of obesity also increased with increasing Townsend scores, in both boys and girls (table2).

UNIVARIATE ANALYSIS

On logistic regression, the crude odds ratio for obesity in the highest quarter of Townsend scores when compared with the lowest quarter was 1.29 (95% CI 1.01 to 1.67, p=0.041) for boys and 1.39 (95% CI 1.08 to 1.79, p=0.011) for girls. Age was also found to be an independent risk factor for obesity with the odds ratio for oldest to youngest age quarter being 1.64 for boys (95% CI 1.28 to 2.11, p<0.001) and 2.17 for girls (95% CI 1.67 to 2.83, p<0.001).

MULTIVARIATE ANALYSIS

On using multiple logistic regression, the relation between obesity and Townsend scores was not confounded by age. The age adjusted odds ratio for obesity between the highest and lowest quarter of Townsend score for both boys (OR=1.29, 95% CI 1.00 to 1.65, p=0.049) and girls (OR=1.39, 95% CI 1.08 to 1.80, p=0.011) was similar to the corresponding crude odds ratio (table 3). In both boys (p=0.017) and girls (p=0.018), there was a significant trend test for increasing obesity with increasing deprivation scores. On introducing the interaction terms in the model, there was a significant interaction between Townsend score and age in the girls (p=0.029) but not the boys (p=0.138). On examining this further, the association between obesity and deprivation was found to be significant only in the oldest age group of girls (11.7 to 14.6 years) with the odds ratio of the highest quarter as compared with the youngest being 1.95 (95% CI 1.23 to 3.08, p=0.005), see table 4.

Discussion

The results demonstrate a strong positive association between childhood obesity and socioeconomic deprivation. Plymouth, a relatively deprived part of UK had prevalent rates of obesity two and half times that expected nationally (5% v 2%). Within Plymouth, rates of childhood obesity were 29% (boys) to 39% (girls) higher in the poorest fourth of the city as compared with the richest fourth. Although some studies have previously reported this association,12-14 others have either failed to confirm it,2 15 shown very modest effects2 16 17or even found a negative association.18 A comprehensive literature review was similarly inconclusive.11 One possible reason why we have been able to demonstrate this association so convincingly is the setting of the study. Plymouth is not only deprived but also has huge socioeconomic variations. The size and nature of the dataset was also an advantage as it included more than 95% of the primary school age population of the city (limiting the effect of chance) with limited ethnic variability (therefore limiting bias), being an almost entirely white population (98.8%).19 Although only state schools were included in the study, the potential bias associated with omitting private schools would be in the direction of under-estimating the influence of material deprivation on the outcome.

MEASUREMENT OF OBESITY

The other possible explanations for differences in the results of various studies lie around methodological issues. The biggest variations are in the measurement and classification of obesity. Although skinfold thickness is useful in clinical settings, BMI (weight/height²) is now becoming the measure of choice for public health purposes because of its reasonable reliability and validity and also its ease and comfort.22 25 26 Although there seems to be a consensus developing around the use of BMI as the measurement tool of choice, there is less consensus about the cut off value to define obesity. Previous studies have used cut offs between 75th to 98th centile without any convincing scientific basis. We have opted for the 98th centile primarily for reasons of consistency—this is the suggested cut off for obesity in the UK growth charts, which we have used to calculate our centiles.21 We argue a sound theoretical basis for this choice. There is significant evidence to suggest that most obese children do not become obese adults. However, a larger proportion of obese adults were obese as children and there is an exponential relation between the severity of obesity in childhood and chances of obesity in adulthood.27-29 Hence using a cut off like the 98th centile increases the chances of selecting a population of children most likely to end up as obese adults and suffer the adverse consequences of obesity.

MEASUREMENT OF DEPRIVATION

The other issue is around the measurement and classification of socioeconomic deprivation. Previous studies have used indicators such as maternal education, single parent families or father's social class. We have used the Townsend material deprivation score that has been used extensively by researchers in the UK. The main disadvantage of using this score is the problem of “ecological fallacy”, as the data were not collected at individual level and may not accurately reflect the socioeconomic status of the child. However, we have tried to limit the effect of this fallacy by using Townsend scores at enumeration district level, which represents approximately 150 households. There is, however, an argument in favour of using area-based measures in studies looking at health effects of socioeconomic position by virtue of the inclusion of the contextual effects of residing in poor neighbourhoods30 such as access to cheap and healthy food options and sporting facilities.31

WIDENING SOCIAL DIFFERENCES IN OBESITY DURING ADOLESCENCE

An important finding of the study is the strongest association in the girls in the oldest age quarter (over 11.7 years). It was not possible to control the results of this association for pubertal status in girls of this age, as the data were not available. Absence of information on pubertal status makes it very difficult to interpret growth changes during adolescence. However, prima facie, controlling for pubertal status would have strengthened the association, based on studies that have shown a positive association between early pubertal onset and increased body mass index28 32 as well as an early onset of puberty in children from higher social classes.33 However, more recent studies have not found a social class differential in pubertal onset in developed countries,32 34 possibly because of better nutrition. These results reflect the results of some of the previous studies that have shown that social differences for obesity heighten during adolescence.14 34 35 This would also explain the gender difference in the results; a larger proportion of boys in this cohort being pre-pubertal as compared with the girls.

WHY ARE POORER CHILDREN FATTER?

Some studies, largely in adults, have tried to understand the reasons for secular trend in obesity, especially in the poorer classes. A population shift towards positive energy balance because of concurrent consumption of calorie dense food and decline in physical activity is implicated.2 This is likely to be more prominent in lower socioeconomic groups where limited finances force the poor to buy more efficiently. These efficiencies often lead to the purchase of foods richer in energy (high in fat and sugar) to satisfy hunger; which are much cheaper per unit of energy than less fattening foods (like fruits and vegetables).31 There is some evidence for decreased physical exercise in poorer people.10 Behavioural studies also suggest that adolescent girls in higher socioeconomic groups tend to be more image conscious and are therefore more likely to put in greater efforts in terms of dietary restraint and physical exercise.11 27 Gender differences in dieting27 and greater amount of physical exercise10 could contribute to the observed weaker association in boys.

IMPLICATIONS OF THE FINDINGS

This study highlights the growing problem of childhood obesity, especially in deprived populations. Evidence from other studies suggests that children from poorer households are also more likely to grow into severely obese adults.35 There is evidence that children from poorer households are more likely to have other risk factors adverse to health leading to greater overall morbidity and mortality in adulthood.6 Suitable interventions promoting healthy eating and physical activity from the early school years onward, need to be considered. Such interventions must be directed at lower socioeconomic groups and be relevant to them. Most importantly, these will have to be accompanied by appropriate policy measures that enable the underprivileged to access cheap and nutritious foods.