Is utility-based quality of life associated with overweight in children? Evidence from the UK WAVES randomised controlled study

For all participants, height and weight measures were taken at school by trained researchers using standardised instruments and procedures. Height was measured to the nearest 0.1 cm using a Leicester height measure. Weight was measured in light clothing without shoes to the nearest 0.1 kg using a Tanita SC-331 S body composition analyser. BMI was calculated by dividing weight (in kilograms) by height (in metres) squared (kg/m2) and used to categorise the children into underweight, healthy weight, overweight and obese groups. The 2^nd, 85^th and 95^th centiles of the UK 1990 Growth reference charts for BMI [28] were used to define the four weight categories, in line with standard UK definitions [29].

As the focus of this study was to explore the association between weight status and HRQOL when measured in utility terms, two instruments were selected for the measurement of HRQOL. Both are generic instruments and thus are designed to measure a wider notion of HRQOL and are not specific to any one disease or condition. The CHU9D is a preference-based utility instrument designed exclusively for use in children and previous research has shown this instrument is the most appropriate choice in this age group [30]. As a utility-based instrument, it is designed to produce a HRQOL score that is preference-based and set between the values of 0 (death) and 1 (full health), however like many preference-based utility instruments, it does produce scores that are deemed to be ‘worse than death’ and therefore have values of less than 0. The PedsQL was chosen as a ‘gold standard’ comparator as this is a widely used HRQOL instrument validated for use in this age group and was the instrument of choice for the WAVES trial from which the data was generated. Although this instrument is non-utility based would be expected to generate HRQOL values which move in the same direction as the CHU9D utility values.

Using the known-groups method, the CHU9D (but not the PedsQL) differentiated HRQOL in children of different ethnic origin (p =0.028) with White British children having the highest mean utility score (Table 2). There was a statistically significant trend of decreasing HRQOL by increasing level of deprivation which was identified by both instruments (P < 0.05). When children were categorised into two groups according to their weight status, neither instrument differentiated between the two groups.

Table 3 shows the distribution of the CHU9D dimensions by weight status category. Overall, the majority of children had no or few problems for all dimensions, irrespective of weight status. There were no underlying differences in the distribution of response to any of the CHU9D dimensions between children in the different weight categories.

Table 4 shows how the mean PedsQL scores corresponded with the options for each of the CHU9D dimensions. The mean PedsQL total scores decrease linearly with increasing severity on each of the CHU9D dimensions.

Figure 1 shows the relationship between the CHU9D utility values and the PedsQL total scores. Although there is a moderate association between the instruments with higher CHU9D utility values corresponding with higher PedsQL total scores, there are some anomalies. For example, one child reported a CHU9D utility of 0.32, yet had a PedsQL total score of 76.09, and another child reported a CHU9D utility score of 0.9, yet had a PedsQL total score of 13.04.

Overall, the correlation between the CHU9D utility values and PedsQL total scores showed a statistically significant moderate, positive correlation (rs = .4696, p = <0.001). The content and coverage of the two instruments were further assessed by examining the correlation between individual CHU9D dimensions and the theoretically similar PedsQL domains (Table 5).

Using conventional cut-off values for Spearman’s ρ, we found that each CHU9D dimension was either weakly, or very weakly correlated with each of the pre-determined PedsQL domain functioning scores. As the CHU9D dimensions are coded with 1 as highest level and 5 as lowest level, the signs on the coefficients were consistently negative.

Table 6 shows the results of the linear mixed regression model (with random effect for school) which compared the CHU9D utility score between the two weight status groups, adjusted for potential confounders (age, gender, ethnicity and deprivation quintile). Children who are overweight or obese have a lower CHU9D utility value (i.e. poorer HRQOL) but this association is not statistically significant. Children from a non-White British background have lower mean CHU9D utility values and this association approaches significance (p = 0.07) for the South Asian population. Also, children from the least deprived areas have significantly higher CHU9D utility values relative to children from the most deprived areas.

The results indicate that there is no statistically significant relationship between the CHU9D utility values and weight status in children aged 5–6 years. Adjusted for potential confounding factors, compared to the healthy/underweight group, children who were overweight/obese reported lower CHU9D utility values, but this effect was not statistically significant. A similar result was found using the PedsQL. When focusing on the CHU9D dimensions, there were no statistically significant differences in scores by child weight status group for any of the dimensions.

Four previous studies that have measured utility-based HRQOL in children [18, 37‐39] have shown similar findings. In a US-based study, Belfort et al. (2011) used the Health Utilities Index-2 (HUI-2) instrument to measure utility-based HRQOL in children and adolescents aged 5–18 years, and found that utility scores were, on average, 0.04 lower in overweight/obese participants compared with healthy weight [37]. Boyle et al. (2010) used the EQ-5D-Y to investigate the effect of weight on the HRQOL in a UK-based population aged 11–15 years and found children who were overweight or obese had a significantly lower HRQOL than children of healthy weight [38]. A recently published paper explored the relationship between BMI and HRQOL using CHU9D in two cohorts of Australian children, aged 9–12 years and 14–16 years. They found mean CHU9D utility values to be lower in children who were overweight or obese (compared to ‘healthy’ weight children), but this effect was only significant in the younger age group [39]. Despite these reports of a negative relationship between HRQOL and being overweight in children, the evidence is mixed in terms of whether this effect reaches statistical significance. Within a UK-based pilot study that was linked to this study, the same direction of effect was found, but there was no statistical difference between utility values and weight status groups in children aged 5–6 years [18]. Three reasons were offered to help explain this result. The first related to the small pilot sample (n = 160), that may not have been large enough to assess subgroup differences. The sample size within this study population is substantially higher, and a similar result was found. The second reason suggested that the CHU9D is not sensitive enough to detect a difference in utility-based HRQOL between overweight and non-overweight children. In this study, the PedsQL total scores are available for comparison, and although the PedsQL shows a negative relationship between weight and HRQOL, again this does not reach statistical significance. Thirdly it was suggested that within this age group, the co-morbidities attached to obesity do not substantially affect HRQOL when measured on a 0–1 utility scale, and it is only once these children approach adolescence that the effects of being overweight have a negative impact on utility values. This might help explain the results within this study.

This study has also contributed evidence on the construct validity of the CHU9D and the results support the convergent and the discriminant validity of the instrument. The most significant, consistent finding within the study population was that HRQOL when measured using both the CHU9D and the PedsQL, was lower within children from the most deprived areas, compared to children from the least deprived areas. This demonstrates that both instruments are discriminating between these groups of children with known differences. Also with respect to discriminant validity, the results showed that the mean CHU9D values were significantly higher for all children with a PedsQL total score greater than or equal to the sample median total PedsQL score, compared to children with a PedsQL total score less than the sample median. Furthermore, PedsQL total scores corresponded well with the individual CHU9D dimensions, with a lower mean PedsQL total score with increasing severity on each CHU9D dimension. Regarding the convergent validity, overall, there was a moderate, statistically significant positive correlation between the PedsQL total scores and the CHU9D utility values. However, despite this correlation between the overall scores of both instruments, we found only a weak, or very weak correlation between the dimensions of each instrument that were pre-determined as being theoretically similar. One possible explanation is that although the PedsQL total scores and the CHU9D utility values tap into a similar underlying construct (HRQOL), the individual dimensions of each instrument, while appearing quite similar, might actually be describing something that is quite specific and different. So at the dimension level the correlations are weak but when combined, the overall instrument scores become moderately correlated.

The data within this study was collected from the WAVES trial which was designed to include a diverse socioeconomic and multi-ethnic population. Parental consent for participation in the WAVES trial was obtained for 57 % of eligible pupils which could lead to sample selection bias. However when the proportion consented out of those eligible was considered by several socio-demographic characteristics, although there was some variation, the differences were generally modest (sex (boys = 65 %, girls = 67 %), ethnicity (white = 75 %, South Asian = 61 %, Black African Caribbean = 64 %; deprivation (most deprived quintile = 65 %, least deprived quintile = 72 %).

As it is rare to have utility information available for children as young as 5 years and for this to be reported for different weight groups, this study contributes this much needed evidence. There are some limitations to note however. First, this paper reports data from a trial and the available data therefore were restricted to what was collected as part of the trial. Ideally, it would have been interesting to assess the convergent validity of the CHU9D utility data with HRQOL data collected using an obesity-specific HRQOL instrument. This would have allowed us to determine if the weak association between weight and utility-based HRQOL in this age group was due to there being no underlying relationship there at all or a lack of sensitivity with detecting the negative effects of being overweight through use of a generic instrument. However, the PedsQL is widely viewed as a ‘gold standard’ generic measure of HRQOL, and has been validated and used in diverse populations. We suggest this as an area for future research. Second, all questions within the PedsQL and the CHU9D were read out to children by an interviewer and this might have had an influence on how children responded. This was a pragmatic decision as children in this age group have very different reading abilities making self-completion problematic but it could have influenced children’s responses to the questions. Third, because of the very small number of children who were measuring ‘underweight’ in our sample (3 %) a decision was made to pull the ‘underweight’ and ‘healthy’ weight children into one weight category. There is no a priori reason to assume that the HRQOL of underweight and healthy weight children are similar but we could not explore this in a statistically robust fashion and the focus of this paper was on the effects of being overweight on HRQOL, not underweight. To enable a comprehensive analysis of the effects of being underweight would have required a purposive sampling approach to ensure adequate numbers of children in this category.