The primary aim of this study was to explore the association between BMI and the utilisation of pharmaceuticals and costs in German children and adolescents. There is no significant correlation between BMI group and the probability of drug utilisation. However, the number of pharmaceuticals used is 14% higher for obese children compared with normal weight children. Furthermore, there is a positive trend to physician-prescribed medication in obese children and adolescents. Regarding those children with drug use, costs are 24% higher for obese children compared with the normal weight group. Thus, a positive association of childhood obesity and drug utilisation and costs is already visible in children and adolescents. A comparison with physician, therapist and hospital costs shows that pharmaceutical costs make an important component of total healthcare costs in children and adolescents.
This is the first study estimating the excess drug costs resulting from obesity based on a representative cross-sectional sample of the German child and adolescent population using a bottom-up approach. One of the main advantages of this approach is the possibility of comparing utilisation and costs in population subgroups, for example with respect to sociodemographic variables and BMI. Although analyses based on comprehensive administrative statistics might give better estimates of the actual level of expenditure for the respective institutions, they are mostly not a representative sample of the population and do not include patients' out-of-pocket expenditures. However, in the German healthcare system, out-of-pocket expenditures are especially relevant for pharmaceuticals. Furthermore, these studies often do not include clinical data, such as measured weight for height, and are therefore limited to cases of diagnosed obesity.
However, several limitations of this study must be pointed out. Most importantly, analyses were based on a cross-sectional survey. Therefore, the results allow for conclusions concerning correlations, but not causal relationships. While obesity was shown to increase the risk of numerous health problems, some illnesses might also induce weight gain. This sample included underweight children and adolescents, although these were not the focus of our analyses. The results show that very underweight children and adolescents cause the highest mean costs, but also had the highest standard deviation. Though underweight is not associated with the probability of incurring costs, there is a significantly positive association with total costs. However, on account of the relatively small percentage of cases, the results for this group should be interpreted with caution. High pharmaceutical costs resulting from low weight seem plausible as extreme underweight might lead to an impaired immune system . However, causality is again unclear, as extreme underweight might also be a consequence of severe or enduring illness, which itself implies increased healthcare utilisation. To definitely answer the question of causality, longitudinal data are required.
Although the problem of recall error should be small considering the short time period, it cannot be excluded as participants are asked to provide information retrospectively - in this case, to state the utilisation of pharmaceuticals for the previous 7 days. Moreover, the results may not be fully representative of the total population with regard to sociodemographic characteristics. These cross-regional differences were reduced by post-stratification weighting for age, sex, region and nationality . Furthermore, as about 34% of those contacted did not respond to the survey, non-response bias cannot be excluded. Costs may be underestimated, because very sick children might not have taken part in the study. However, extensive non-response analyses have been conducted that show only moderate differences in sociodemographic as well as health-related characteristics . As far as non-response is explained by age, sex, region or nationality, it is accounted for by using the respective weights. Statistical analyses did not include adjustments for comorbidities. Correlation between health problems and overweight is not trivial; thus, the excess cost approach tries to capture all the differences between the analysed BMI groups.
Regarding the estimation of drug costs, several assumptions were necessary that may have caused over- or underestimation of costs. The estimation of pharmaceutical costs was based on the DDD, as suggested by the WHO. However, this measure tends to overestimate drug consumption: first, if pharmaceuticals are not specifically for children, the DDD refers to the daily dose for adults; second, it presumes full compliance. Furthermore, the frequency of drug intake was estimated based on the four response categories 'daily' and 'several times a day', 'often, but not daily' and 'less often than once per week'. It is not clear how this affects cost results. Yet, data for a more precise population-based assessment of pharmaceutical costs are not available so far.
Sensitivity analyses were performed regarding package sizes, seasonal effects and legal price discounts. Discount contracts between the pharmaceutical industry and healthcare insurers could not be taken into account, because they are not publicly available. Although all these changes affected the extent of costs in total, none affected the differences between BMI groups.
In this study, we estimated the costs for drug consumption, not actual expenditures, which might be even higher, if packages are only partly used and leftovers are thrown away. As utilisation of pharmaceuticals was requested for the last 7 days, the extrapolated yearly cost estimates should be interpreted with caution.
SES (based on parental income and education) was included as a confounder in statistical analyses because it may influence health care utilisation as an 'enabling factor' . Yet, it has to be noted that SES may also be associated with overweight and obesity, but the direction of the causal relationship is not clear [44, 45]: Although low income might have a negative impact on health behaviour resulting in weight gain, overweight and obesity in adults could also impede labour market outcomes and cause lower wages  - in our case this is only relevant if we assume a high correlation of the weight status of parents and their children. However, a recalculation of the regression model without SES as a confounding variable did not change our results.
The medical literature often questions BMI as a valid and accurate measure of overweight and obesity . Especially for younger children, alternative approaches have been proposed with a slightly higher sensitivity . However, the information required to compute the BMI is easy to collect and common in a number of social science data sets. A recently published study suggests that BMI serves as a good surrogate marker for obesity in population studies .