The impact of childhood overweight and obesity on healthcare utilisation

Economics and Human Biology 27 (2017) 84–92

Contents lists available at ScienceDirect

Economics and Human Biology journal homepage: www.elsevier.com/locate/ehb

The impact of childhood overweight and obesity on healthcare utilisation Edel Doherty* , Michelle Queally, John Cullinan, Paddy Gillespie Health Economics and Policy Analysis Centre, J.E. Cairnes School of Business and Economics, NUI Galway, Ireland

A R T I C L E I N F O

Article history: Received 9 December 2016 Received in revised form 23 March 2017 Accepted 10 May 2017 Available online 12 May 2017 Keywords: Overweight and obesity BMI Healthcare utilisation Children Instrumental variables Ireland

A B S T R A C T

Rising levels of childhood overweight and obesity represent a major global public health challenge. A number of studies have explored the association between childhood overweight and obesity and healthcare utilisation and costs. This paper adds to the literature by estimating the causal effect of child overweight and obesity status on use of general practitioner (GP) and hospital inpatient stays at two time points using instrumental variable (IV) methods The paper uses data from two waves of the Growing Up in Ireland survey of children when they are 9 and 13 years respectively and uses the biological mother’s body mass index (BMI) as an instrument for the child’s BMI. Our results demonstrate that child overweight and obesity status do not have a significant effect on healthcare utilisation for children when they are 9 years, but do have a large and significant effect at 13 years. Across all our models, the effects on both GP and hospital inpatient stays are found to be larger when endogeneity in childhood BMI status is addressed. Previous studies that did not address endogeneity concerns are likely to have significantly underestimated the impact of child overweight and obesity status on healthcare utilisation. © 2017 Elsevier B.V. All rights reserved.

1. Introduction The rising trends in overweight and obesity have been described as a global epidemic that represents a major challenge for health care systems (World Health Organization, 2016), both in terms of their current and projected future demands on already resource-constrained healthcare budgets. The prevalence of childhood overweight and obesity worldwide increased over 47% between 1980 and 2013 (Ng et al., 2014). In 2014, an estimated 41 million children under 5 years of age were affected by overweight or obesity (World Health Organization, 2016). While there is some evidence that prevalence rates are stabilising in developed countries, they are doing so at a very high rate of well over 20% and there is no evidence that the prevalence in children is decreasing (Rokholm et al., 2010; Skinner et al., 2016). Against this backdrop of high prevalence rates and the subsequent implications for health outcomes, economic outcomes and healthcare budgets, there have been calls to urgently reverse the trends in childhood overweight and obesity (World Health Organization, 2016). Childhood overweight and obesity is accompanied by a number of adverse consequences for physical and mental health in both the

* Corresponding author. E-mail address: [email protected] (E. Doherty). http://dx.doi.org/10.1016/j.ehb.2017.05.002 1570-677X/© 2017 Elsevier B.V. All rights reserved.

short- and long-term (Llewellyn et al., 2016). Much of the focus in the literature has been on the long-term effects of childhood overweight and obesity as it manifests into adulthood, as there was a perception that many of the adverse consequences occur later in adulthood (see Llewellyn et al. (2016) for a review). However, overweight and obesity in childhood leads to many acute health problems and much suffering during childhood. In some cases children are now displaying what were once thought to be “adult diseases related to obesity”, such as non-alcoholic fatty liver disease (Uppal et al., 2016) and hypertension (Rosner et al., 2013), as a result of childhood obesity. Furthermore, from a policy perspective, the burden to the health care system is of particular interest considering the existent projections of increasing health care expenditure in western countries. Relative to overweight and obesity in adulthood, there is limited published evidence on the economic burden associated with child or adolescent overweight and obesity. The earliest studies seeking to assess this burden date back only to 2006 and they do not provide a clear picture, with much ambiguity surrounding the impact of childhood obesity on healthcare costs for children (John et al., 2010, 2012). For example, some of those studies reported a positive relationship between childhood overweight and obesity and healthcare usage and/or costs (Hampl et al., 2007; Finkelstein and Trogdon, 2008; Trasande and Chatterjee, 2009; Au, 2012; Lynch et al., 2015; Bianchi-Hayes

E. Doherty et al. / Economics and Human Biology 27 (2017) 84–92

et al., 2015; Hayes et al., 2016; Carey et al., 2015), while other studies reported to the contrary (Skinner et al., 2008). Some studies did not focus specifically on the burden arising during childhood, but the overall lifetime excess costs associated with childhood obesity (Finkelstein et al., 2014; Sonntag et al., 2016). A crucial limitation of these studies, however, is that they provide estimates of the association between BMI status and healthcare usage or costs rather than estimates of the causal effect.1 This is a subtle but critical point for the analysis of obesity and its impacts. For example, if a child becomes obese as a result of an accident that impairs their mobility and the accident also leads to greater healthcare utilisation, then failing to adequately control for this will lead to an overestimate of the causal effect. On the other hand, if children from certain categories of households are more likely to be obese and also have poorer access to health care services, then this could lead to an underestimate of the true effect. In both examples (and there are likely many more), failure to adequately control for factors that impact healthcare utilisation and that are also correlated with child BMI status leads to omitted variable bias. A second endogeneity concern relates to potential reverse causality in the relationship between utilisation and BMI. For example, if children who have more contact with healthcare professionals receive more advice about their weight status, diet, exercise, etc. and react accordingly, then analysing the correlation between utilisation and BMI may also pick up this effect. Thus, in order to correctly estimate the causal effect of BMI on utilisation, an identification strategy is required that appropriately addresses these concerns. In fact, Pelone et al. (2012) note that most of the research designs cannot state with confidence that overweight and obesity are the underlying causes of higher healthcare usage and costs for children. This paper directly addresses this gap in the literature by employing an instrumental variable (IV) approach to identify the causal effect of child overweight and obesity on healthcare utilisation in Ireland. More specifically, it analyses data from two waves of the Growing Up in Ireland (GUI) study, a nationally representative survey collecting information from families of children when they are 9 years and 13 years respectively and uses the BMI of the biological mother as an instrument for child BMI. This approach is based on previous studies that have exploited the use of a biological relative’s BMI as an instrument for adult BMI status, albeit in different contexts (Cullinan and Gillespie, 2016; Cawley and Meyerhoefer, 2012; Cawley, 2004; Kline and Tobias, 2008; Smith et al., 2009). An accurate assessment of the causal pathways between child BMI status and healthcare usage is important given the rise in childhood overweight and obesity worldwide. Within Europe, Ireland along with the United Kingdom are projected to have the highest rates of overweight and obesity by 2025 (Ng et al., 2014). Moreover, with some exceptions (Breitfelder et al., 2011; Wenig, 2012; Sonntag et al., 2016, 2015; Hayes et al., 2016; Batscheider et al., 2014), many of the previous studies examining childhood obesity and healthcare utilisation or costs are US-based and questions arise as to their generalisability to countries with different prevalence rates of overweight and obesity, as well as different healthcare systems and treatment structures. Within this context, Ireland is a noteworthy case study given its mix of public and private healthcare finance and provision. While predominantly a taxation-financed public healthcare system requiring the majority of the population to pay out of pocket for GP care, voluntary private health insurance is typically used to pay for acute care in both public and private hospitals (Nolan and Smith, 2012).

1 See Cawley and Meyerhoefer (2012) for a discussion of this issue in the context of adult obesity.

85

In addition to being one of the first studies to causally infer the relationship between child BMI status and healthcare usage, this paper adds to the literature in a number of other ways. For example, we use independently measured BMI for mothers and children rather than measures based on self-reported weight and height. We also examine healthcare utilisation at two points in time for children (at 9 and 13 years) to explore how the patterns of usage may differ by BMI status and age. This allows us to examine whether patterns of utilisation differ by overweight or obese status at both ages and to compare patterns of utilisation for those who were overweight or obese at both ages to those who were overweight or obese at one age only. Furthermore, we employ an extremely rich and nationally representative sample allowing us to control for many covariates and factors that may explain healthcare utilisation. The paper is structured as follows: Section 2 presents a detailed description of the data and describes our empirical approach. Section 3 presents the main empirical results, Section 4 presents some extensions to the analysis, while Section 5 discusses the implications of our results and concludes. 2. Data and methods The data analysed is from two waves of the GUI survey, a nationally representative face-to-face survey of children living in Ireland that also surveys their parents and the principals and teachers in the child’s school. The first wave of the GUI survey collected information on 8568 nine year old children between 2007 and 2008, representing approximately 14% of all nine year olds in the Republic of Ireland in 2008 (Murray et al., 2009; McCrory and Layte, 2012). Overall 7525 of these children were interviewed again between 2011 and 2012 for the second wave of data collection. Further details of the wave one and wave two surveys, including the sampling procedures, are discussed, respectively, in Murray et al. (2010) and Thornton et al. (2016).2 Our primary analysis focuses on presenting results from separate binary models where the dependent variables are (1) whether the child has had contact with their GP over the previous 12 months and (2) whether the child has stayed as an inpatient in hospital over their lifetime. A key feature of the GUI data is that it contains measured weight and height (i.e. not self-reported) for both children and parents, allowing for more accurate measurements of BMI status. According to the GUI survey instrumentation, the height and weight measurements were taken at the time of interview. Weight measurements of parents and children were recorded to the nearest 0.5 kg using a SECA 761 medically approved flat mechanical scales which graduated in one-kilogram increments and had an upper capacity of 150 kg. Parents and children were asked to wear light clothing for weight measurement. Height was recorded to the nearest millimetre using a Leicester portable height stick (Layte and McCrory, 2011: p.10–11). We present models with child BMI included as a continuous variable and also undertake additional analyses with children categorised as nonoverweight, overweight or obese, based on the BMI cut-offs in Cole et al. (2000).3 Specifically, for wave 1, since all children are aged 9 years, we use the 9 years and 6 months cut-offs for girls and boys, which are 19.45 and 19.46 for overweight and 23.46 and 23.39 for obese, respectively. For wave 2 we use the 13 years and 6 months cut-offs for girls and boys, which are 22.98 and 22.27 for overweight and 28.20 and 27.25 for obese.

2 For a recent application of this data examining socioeconomic gradients in childhood obesity, see Walsh and Cullinan (2015). The data was also recently used to analyse the impact of obesity on self-rated health in Cullinan and Gillespie (2016). 3 These cut-offs were also used in Walsh and Cullinan (2015) using the same data.

86

E. Doherty et al. / Economics and Human Biology 27 (2017) 84–92

Table 1 Variable Definitions and Sample Descriptive Statistics. Variable Name

Dependent variables GP Usage Inpatient Stays Independent variables Child Characteristics Child BMI Child BMI Categories

Female Birth Weight When Born

Irish Citizen Health Status

Variable Description

GUI: Wave 1 % or Mean (Std. Deviation)

GUI: Wave 2 % or Mean (Std. Deviation)

=1 if child has contacted their GP (by phone or in person) over the previous 12 months; = 0 42.28% otherwise =1 if child has had an inpatient stay in hospital over the course of their lifetime (excluding time of 42.13% birth); = 0 otherwise

46.19%

Child’s independently recorded BMI Child’s BMI Classification 1. Non-Overweight (Reference category) 2. Overweight 3. Obese =1 if child is a girl; =0 for boys Child’s birth weight in kilograms Child born late, on time or early Late: (42 weeks or later) (Reference category) On time: (37–41 weeks) Somewhat early: (33–36 weeks) Very early: (32 weeks or less) =1 if child Irish citizen; =0 otherwise Child’s health in the past year 1. Very healthy, no problems (Reference Category) 2. Healthy, but a few minor problems 3. Sometimes quite ill/Almost always unwell =1 if child has a chronic illness or disability; =0 otherwise

17.80 (2.92)

20.58 (3.62)

75.92% 18.40% 5.68% 51% 3.51 kgs (0.604)

75.66% 19.29% 5.05% 51% 3.52 kg (0.601)

24.28% 63.62% 10.72% 1.39% 95%

24.48% 63.44% 10.72% 1.37% 95%

74.69% 24.09% 1.22% 9.95%

77.72% 21.05% 1.23% 9.79%

25.98 (4.74)

26.51 (4.94)

0.88%

0.71%

49.12% 32.30% 17.70% 39.76 (5.27) 12.92%

41.73% 36.29% 21.27% 41.21 (3.57) 17.98%

0.79% 5.06% 18.93% 40.98% 34.24% 2.05 (3.22)

1.01% 5.28% 20.97% 38.40% 34.33% 2.40 (3.36)

15.66% 57.33% 27.02%

11.07% 57.28% 31.65%

20.30% 66% s21,694. (s14,056) 45.50% N = 6566

26.54% 60.53% s17,675 (s9344) 45.02% N = 5924

Child Chronic Illness Mother’s Characteristics Mother's BMI Mother’s independently recorded BMI (Instrument) 1. Underweight Mother’s BMI Categories 2. Normal Weight 3. Overweight 4. Obese Age Mother’s age in years Chronic Illness =1 if has a chronic illness; =0 otherwise Health Status In general, how would you say your current health is? 1. Poor (Reference Category) 2. Fair 3. Good 4. Very good 5. Excellent Depression Score Total depression score (based on the Centre for Epidemiological Studies Depression Scale, CESD8) =1 if primary education Education Level =2 if secondary education =3 if higher education Household Characteristics Medical Card =1 if household has a medical card; =0 otherwise PHI =1 if household has private health insurance; =0 otherwise Income Equivalised household income in Euros Urban =1 if family live in an urban location; =0 otherwise Observations

46.00%

Source: Analysis of GUI data.

A range of other potential determinants of health care use are also considered, the selection of which were based broadly on Andersen and Newman's (2005) theoretical framework for modelling healthcare utilisation and more specifically on previous research on healthcare utilisation conducted for the child cohort in Ireland (Layte and Nolan, 2015). These controls include child characteristics such as gender, birthweight, gestation age and citizenship, mother’s characteristics such as age, health status, education status, marital status and depression score and household characteristics such as income, location, medical card and health insurance status. In models of healthcare utilisation for Ireland, it is important to control for the medical card status (that is, eligibility for free GP care) and private health insurance status of the child’s family (Nolan and Smith, 2012). For all models presented in this paper we apply sampling weights in line with

the sample design to ensure the data is representative of the wider population of children at these ages. Table 1 presents details and descriptive statistics for our variables for wave 1 and 2. It is worth noting that missing data on several of the variables reduced the sample sizes somewhat for analysis.4 While direct comparisons between waves should be treated with some caution due to attrition between waves, Table 1 shows that there was approximately a four percentage point increase in both the use of GP services and inpatient stays at 13 years

4 The main variables for which there was missing data in wave 1 include: child BMI (5% missing), mother’s BMI (9% missing), mother’s depression score (8% missing) and household income (7%). In wave 2, missing data was mainly observed for child BMI (3%), mother’s BMI (6%) and household income (8%).

E. Doherty et al. / Economics and Human Biology 27 (2017) 84–92

compared to 9 years. With respect to child characteristics, we can see that there are fewer non-overweight children and more overweight children at 13 years compared to 9 years. We also note that there is a slight improvement in the health status of children at 13 years compared to when they were younger. In terms of maternal characteristics, a higher proportion of mothers fall into the overweight and obese categories when their children are 13 years. We observe that there is approximately a five percentage point increase in mothers reporting having a chronic illness when their children are 13 years compared to 9 years. We also see some differences in the attrition rates based on socio-economic variables. Fewer mothers with a primary level education only completed the second wave and a higher proportion of mothers with a third level education completed it. There are also some evident effects of the recession between the waves with the proportion of households having a medical card increasing and the proportion with private health insurance declining between the two waves. We also observe a decrease in equivalised household income between the two waves. In terms of our empirical strategy, we first consider separate binary probit models (PROBIT) for the two dependent variables, each estimated without and with additional controls. Next, following Cawley and Meyerhoefer (2012), we use mother’s BMI, mother’s BMI squared and mother’s BMI cubed as instruments in IV probit models (IVPROBIT) estimated with controls. We also examine the relationship between healthcare utilisation and BMI at two time points; when children are 9 years and 13 years respectively. Defining healthcare utilisation for child i (HUi) as a binary variable indicating use (1) or no use (0) of the service in question (i.e. GP contact or inpatient stay), the general specification of our IV probit model is: HU
i ¼ b1 BMIi þ b2 Xi þ ui

ð1Þ

BMIi ¼ g 1 Zi þ g 2 Xi þ vi

ð2Þ

*

where HUi is latent unobserved healthcare utilisation, BMIi is a (continuous) endogenous variable, Xi is a vector of exogenous variables (controls) and Zi a vector of instruments. The model assumes that the error terms ui and vi are independent and identically distributed multivariate normal for all i, while the coefficients to be estimated are represented by b and g. For our purposes b1 is the primary coefficient of interest. In this recursive model we do not observe HUi* but instead define:  0if HU
i < 0 ð3Þ HU i ¼ 1if HU
i  0 Estimation is via maximum likelihood (using the ivprobit command in Stata), though a two stage estimator can also be used allowing test statistics to be calculated for the so-called ‘first stage’ equation (i.e. Eq. (2)). In particular, the ‘power’ of the instrument is typically assessed using the F-statistic from this first stage regression. Where BMI categories are considered instead of continuous BMI in our IV probit model the ivprobit command is no longer appropriate and instead we use the user-written CMP command in Stata (see Roodman (2009) for a detailed description of the command). For our approach to be considered valid, two requirements must be established. The first, power, is that our instrument (biological mother’s BMI) is a powerful predictor of child BMI. As described later, our F-statistics estimates easily exceed the conventional F-statistic benchmark of 10 in a series of first stage regressions (Stock et al., 2012). The second requirement is that the instrument is ‘valid’, meaning that the IV is not correlated with the error term

87

in the second stage regression. In the context of this study this means that the BMI of the child’s mother is not correlated with the child’s residual healthcare usage. In other words, we assume that the instrument does not suffer from the same potential endogeneity problem as child BMI, though we acknowledge there is no empirical/statistical way to check whether this requirement holds. Cawley and Meyerhoefer (2012) discuss, in detail, the evidence showing the strong genetic links in BMI rather than BMI being determined by common household factors. As a falsification test in their study, they find that the weight of a stepchild is not a significant predictor of parental weight, providing evidence of the validity of the instrument. We also find that the weight of a step parent/guardian is not a significant predictor of child weight in this study (F = 1.20). 3. Results Table 2 presents the average marginal effects for children aged 9 years on GP usage and inpatient stays for our range of models. For comparability, we present probit and IV probit models for both dependent variables. Our F-statistic from the first stage IV regression is 76.16 suggesting that our instrument is a powerful predictor of child BMI status. The results show little or no significant effect of child BMI status on healthcare utilisation for either GP or inpatient stays for children when they are 9 years. These results are robust to the inclusion of various covariates and whether BMI is coded as a continuous or categorical variable. Table 3 presents the results for healthcare utilisation for children when they are 13 years. The F-statistic from the first stage IV regression is 64.51. The effect of BMI status on healthcare utilisation at 13 years is positive and significant across all models. A notable aspect of the findings is that the magnitude of the average marginal effects of BMI on healthcare utilisation increases substantially under the IV models. For GP usage the probit models suggest that a one unit increase in BMI increases the probability of a GP visit by 0.7 percentage points (ppts), whereas under the IV models the marginal effect increases to 2.5 ppts. For inpatient stays BMI is significant at the one per cent level under the IV model and the effect again is larger. A notable aspect of this model is that while BMI status has no effect on lifetime inpatient stays on children up until they are 9 years old, it has a significant and large effect between when children are 13 years. Using the IV approach, a one unit increase in BMI increases the probability of an inpatient stay by 2.1 ppts. To explore potential non-linearities in the effect of BMI,5 and given that a large focus in policy circles surrounds the problems of childhood obesity specifically, Table 4 presents the average marginal effects of the overweight and obesity categories relative to the non-overweight category for the 13 year old children. The categories are calculated based on the Cole et al. (2000) cut-offs for girls and boys. In general, the probit models suggest small and/or non-significant effects whereas the IV models suggest large and significant effects for both children who are overweight or obese. Under the IV models, the probabilities of a GP visit and an inpatient stay are respectively 20 and 16 ppts higher for an overweight child compared to their non-overweight peers. The corresponding marginal effects for an obese child are 35 and 27 ppts.

5 We tried to estimate models for healthcare utilisation for children at 9 and 13 years including both continuous BMI and BMI squared as endogenous regressors but the models did not converge.

88

E. Doherty et al. / Economics and Human Biology 27 (2017) 84–92

Table 2 Models of Healthcare Utilisation for Children aged 9 Years. GP Usage

Child Characteristics Child BMI

Probit No Controls

Probit With Controls

IV Probit With Controls

Probit No Controls

Probit With Controls

IV Probit With Controls

0.003 (0.002)

0.003 (0.002) 0.028* (0.0152) 0.018 (0.0138) 0.009 (0.018) 0.061** (0.028) 0.058 (0.067) 0.026 (0.037)

0.013 (0.011) 0.023 (0.016) 0.023 (0.015) 0.007 (0.018) 0.058** (0.028) 0.064 (0.06) 0.026 (0.037)

0.004 (0.002)

0.004* (0.002) 0.076*** (0.015) 0.035*** (0.013) 0.01 (0.018) 0.041 (0.029) 0.075 (0.068) 0.071** (0.035)

0.003 (0.011) 0.076*** (0.016) 0.034** (0.015) 0.010 (0.019) 0.040 (0.029) 0.076 (0.069) 0.071** (0.035)

0.001 (0.001) 0.048* (0.025) 0.004 (0.103) 0.011 (0.099) 0.026 (0.099) 0.091 (0.100) 0.012 (0.020) 0.064*** (0.026) 0.009 (0.027) 0.006** (0.002)

0.001 (0.002) 0.047* (0.025) 0.008 (0.103) 0.005 (0.100) 0.018 (0.099) 0.083 (0.100) 0.017 (0.021) 0.073*** (0.027) 0.008 (0.027) 0.006** (0.002)

0.007*** (0.001) 0.041 (0.025) 0.157* (0.090) 0.167* (0.086) 0.211*** (0.086) 0.215*** (0.087) 0.0007 (0.021) 0.018 (0.025) 0.025 (0.026) 0.002 (0.002)

0.007*** (0.002) 0.042* (0.025) 0.158* (0.091) 0.168* (0.087) 0.212*** (0.087) 0.216*** (0.088) 0.0001 (0.021) 0.019 (0.028) 0.025 (0.027) 0.002 (0.002)

0.105*** (0.023) 0.058*** (0.019) 0.000 (0.00) 0.025* (0.015) 6566

0.105*** (0.023) 0.058*** (0.019) 0.000 (0.000) 0.026* (0.015) 6566

0.082*** (0.023) 0.050*** (0.019) 0.000 (0.000) 0.013 (0.015) 6566

0.082*** (0.023) 0.050*** (0.019) 0.000 (0.000) 0.013 (0.015) 6566

Female Birth Weight When Born: On time: (37–41 weeks) Somewhat early: (33–36 weeks) Very early: (32 weeks or less) Irish Citizen

Mother’s Characteristics Age Chronic Illness Fair Good Very Good Excellent Second Level Third Level Single Depression

Household Characteristics Medical Card Health Insurance Household Income Urban Observations

Inpatient Stays

6566

6566

Notes: The table presents estimated average marginal effects. In the IV models, the child BMI variable is instrumented with mother’s BMI. In the non-IV models child BMI is included directly in the models as an explanatory variable. Standard errors are presented in parentheses. *** denotes statistically significant at 1%, ** denotes statistically significant at 5%, and * denotes statistically significant at 10%.Source: Analysis of GUI data.

4. Extensions 4.1. Is the persistence of overweight or obesity important? To explore whether the persistence of overweight or obesity is important, we undertake additional subgroup analysis on healthcare usage for children when they 13 years old  see Table 5.6 The subgroups are developed to describe (1) children who are overweight or obese at age 9 years but are not overweight/obese at 13 years old (2) children who are overweight or obese at 13 years but are not overweight/obese at 9 years and (3) children who are

6 For the subgroups we combine the overweight and obese categories to reduce the likelihood that the results are affected by small sample size problems.

either overweight or obese at both 9 and 13 years old. These are compared to children who are not overweight/obese at both ages. The F-statistic from the first stage IV regression is 29.15. In the data, approximately 69% of children are not overweight/obese in both waves. Almost 7% of children are overweight or obese at 9 years but not at 13 years and approximately 8% of children are overweight or obese at 13 but not when they are 9 years old. Approximately 16% of children are overweight or obese in both waves. Consistent with earlier findings, the size of the marginal effects are larger using the IV approach. For GP visits, children who are overweight or obese at both time points have a significantly higher likelihood, at the one per cent level, of using GP services. In the case of inpatient stays, children who are overweight or obese at 9 years but who are not overweight by the time they are 13 years, are significantly more likely to have had an inpatient stay. The effects

E. Doherty et al. / Economics and Human Biology 27 (2017) 84–92

89

Table 3 Models of Healthcare Utilisation for Children aged 13 Years. GP Usage

Child Characteristics Child BMI

Probit: No Controls

Probit: With Controls

IV Probit: With Controls

Probit: No Controls

Probit: With Controls

IV Probit: With Controls

0.009*** (0.002)

0.007*** (0.002) 0.012 (0.017) 0.007 (0.015) 0.045** (0.020) 0.039 (0.033) 0.039 (0.078) 0.078* (0.042)

0.025*** (0.008) 0.010 (0.020) 0.016 (0.016) 0.042** (0.020) 0.035 (0.032) 0.040 (0.073) 0.078* (0.041)

0.003 (0.002)

0.004* (0.024) 0.081*** (0.017) 0.039*** (0.015) 0.013 (0.209) 0.064** (0.033) 0.054 (0.078) 0.027 (0.044)

0.021*** (0.009) 0.098*** (0.019) 0.046*** (0.016) 0.011 (0.020) 0.065** (0.032) 0.054 (0.073) 0.029 (0.042)

0.0006 (0.002) 0.083*** (0.025) 0.13 (0.087) 0.128 (0.082) 0.161* (0.083) 0.199*** (0.084) 0.003 (0.026) 0.031 (0.030) 0.061* (0.026) 0.0005 (0.002)

0.001 (0.002) 0.074*** (0.025) 0.119 (0.084) 0.119 (0.079) 0.150* (0.080) 0.190*** (0.081) 0.017 (0.026) 0.051 (0.031) 0.048* (0.027) 0.001 (0.003)

0.004 (0.002) 0.053** (0.026) 0.027 (0.091) 0.010 (0.087) 0.06 (0.088) 0.074 (0.089) 0.009 (0.026) 0.027 (0.031) 0.022 (0.026) 0.004 (0.003)

0.004* (0.002) 0.046* (0.026) 0.031 (0.088) 0.006 (0.084) 0.054 (0.085) 0.071 (0.086) 0.022 (0.027) 0.007 (0.033) 0.031 (0.026) 0.003 (0.003)

0.155*** (0.023) 0.072*** (0.020) 0.000 (0.000) 0.013 (0.017) 5924

0.148*** (0.023) 0.075*** (0.020) 0.000 (0.000) 0.013 (0.017) 5924

0.083*** (0.024) 0.015 (0.021) 0.000* (0.000) 0.030* (0.018) 5924

0.079*** (0.023) 0.020 (0.021) 0.000* (0.000) 0.028* (0.017) 5924

Female Birth Weight When Born: On time: (37–41 weeks) Somewhat early: (33–36 weeks) Very early: (32 weeks or less) Irish Citizen

Mother’s Characteristics Age Chronic Illness Fair Good Very Good Excellent Second Level Third Level Single Depression

Household Characteristics Medical Card Health Insurance Household Income Urban Observations

Inpatient Stays

5924

5924

Notes: The table presents estimated average marginal effects. In the IV models, the child BMI variable is instrumented with mother’s BMI. In the non-IV models child BMI is included directly in the models as an explanatory variable. Standard errors are presented in parentheses. *** Denotes statistically significant at 1%, ** denotes statistically significant at 5%, and * denotes statistically significant at 10%.Source: Analysis of GUI data. Table 4 Models of Healthcare Utilisation for Children aged 13 Years Using Categorical BMI Measures. GP Usage

Child Overweight Child Obese Observations

Inpatient Stays

Probit: No Controls

Probit: With Controls

IV Probit: With Controls

Probit: No Controls

Probit: With Controls

IV Probit: With Controls

0.032 (0.023) 0.099*** (0.037) 5924

0.027 (0.022) 0.067* (0.037) 5924

0.203*** (0.067) 0.349*** (0.106) 5924

0.012 (0.023) 0.047 (0.038) 5924

0.019 (0.022) 0.043 (0.038) 5924

0.162** (0.074) 0.273** (0.12) 5924

Notes: The table presents estimated average marginal effects. The models also control for all of the covariates listed in Table 3 (results available on request). Standard errors are presented in parentheses. *** denotes statistically significant at 1%, ** denotes statistically significant at 5%, and * denotes statistically significant at 10%.Source: Analysis of GUI data.

on inpatient stays are largest for children who are overweight or obese at both waves. It appears, therefore, that even though BMI does not have a significant impact on healthcare utilisation for children when they are 9 years, being overweight or obese from 9 years does

have a significant effect on healthcare utilisation at 13 years. The magnitudes of the effects are largest for children who are overweight or obese at both ages. This makes intuitive sense as it is plausible that these children are at the higher end of the BMI distribution.

90

E. Doherty et al. / Economics and Human Biology 27 (2017) 84–92

Table 5 Models of Healthcare Utilisation of Children aged 13 Years: Modelling the Effects of Persistence of Overweight or Obesity over Time. GP Usage Probit: No Controls Child overweight/obese at 9, not at 13 0.020 (0.039) Child overweight/obese at 13, not at 9 0.024 (0.034) Child overweight/obese at 9 and 13 0.066 years (0.024) 5657 Observations

Inpatient Stays Probit: With Controls

IV Probit: With Controls

Probit: No Controls

Probit: With Controls

IV Probit: With Controls

0.018 (0.038) 0.008 (0.033) 0.055** (0.024) 5657

0.131** (0.059) 0.151** (0.066) 0.260*** (0.087) 5657

0.029 (0.039) 0.044 (0.034) 0.031 (0.025) 5657

0.038 (0.037) 0.008 (0.033) 0.039 (0.024) 5657

0.140** (0.060) 0.121* (0.072) 0.226** (0.095) 5657

Notes: The table presents estimated average marginal effects. The models also control for all of the covariates listed in Table 3 (results available on request). The sample size decreases in these models due to missing data on wave 1 Child BMI status for the wave 2 cohort of children. Standard errors are presented in parentheses. *** denotes statistically significant at 1%, ** denotes statistically significant at 5%, and * denotes statistically significant at 10%.Source: Analysis of GUI data.

4.2. Mechanism by which BMI may affect healthcare utilisation

4.3. Robustness checks

We explore a number of possible reasons for why there is a higher likelihood of healthcare utilisation for overweight and obese children at 13 years compared to normal weight children. In the study, the child’s primary caregiver7 is asked whether the child has any ‘on-going physical or mental health problems, illness or disability’. In total, 9.7% of caregivers reported that their child had an on-going chronic illness. Breaking this down by BMI status, 9.4% of non-overweight children, 9.5% of overweight children and over 16% of obese children were reported to have a chronic illness by their caregiver. The primary caregiver was also asked to select which best describes the child’s health in the last year: “Very healthy, no problems; Healthy, but a few minor problems; Sometimes quite ill; Almost always unwell”. In response, 79%, 76% and 66% of non-overweight, overweight and obese children, respectively, are reported by their caregivers as being very healthy. This suggests a much lower health status for obese children relative to their nonoverweight or overweight peers. The primary caregiver is also asked whether the child had any accidents requiring hospital treatment in the previous 12 months. Approximately 14% of normal weight children, 13% of overweight children and 15% of obese children had an accident that required hospital treatment in the previous 12 month period. The primary caregiver is also asked to indicate whether the child had any episodes of chest wheezing in the previous 12 months. Approximately 14% of normal weight children, 17% of overweight children and 23% of obese children had episodes of wheezing in the previous 12 months. While the GUI survey does not contain more detailed information on health conditions, it is possible that some of these factors might explain higher healthcare utilisation for overweight and obese children at 13 years. At the same time, there may be other reasons (such as depression) that may capture higher usage but which we do not have information on. While we do not observe any differences in healthcare utilisation by BMI status for 9 year old children, there are also some differences in reported health status. For instance, 9% of nonoverweight children, 11% of overweight children and 12% of obese children were reported to have a chronic illness by their caregiver at 9 years. Similarly, 76% of normal weight children, 74% of overweight children and 66% of obese children are reported as being very healthy by their caregiver. We don’t observe any difference in the percentage of children who are reported to have accidents by BMI status at 9 years8 and there is no question related to wheezing in the 9 year old data.

In additional robustness checks we re-estimated the models presented in Tables 2 and 3 using BMI Z-scores for girls and boys (using standard deviations calculated from the GUI dataset) and did not find that our results were significantly different. We also estimated models using alternative subsets of control variables to determine how sensitive the coefficients on BMI are to their exclusion. Furthermore, we estimated several versions of the IV models with (a) all the mother characteristic variables removed, (b) with both the mother and household characteristics removed, and (c) with all controls removed except for instrumented BMI. We found that our results on BMI were not significantly different from those presented in Tables 2 and 3. Given some attrition between waves we also re-estimated our models from the first wave on the sample from the second wave, using the second wave’s probability weights, and our results do not change from those presented in Table 2.

7

In almost 99% of cases the primary caregiver is the child’s mother. The wording of the question at 9 years relates to accidents over the child’s lifetime. 8

5. Discussion This paper sought to identify the causal effect of child overweight and obesity, classified by BMI status, on healthcare utilisation at two points during childhood (when children are 9 and 13 years). The paper explicitly addressed the potential endogeneity in child BMI status by using maternal BMI as an instrumental variable in the econometric models of healthcare utilisation. This approach is in line with other studies that have used BMI of biological relatives as an instrument, albeit in different contexts. The paper found no significant effects of BMI status on healthcare utilisation when children were 9 years old. This finding was robust to the inclusion of various covariates and the use of IVs to address endogeneity in child BMI status. A notable finding of these results was that the binary probit models substantially underestimated the size of the effects of child BMI status on healthcare utilisation. More specifically, in the IV models, the magnitude of the effects of child BMI status on healthcare utilisation increased substantially and similar results were found when BMI status was modelled continuously or as a categorical variable. The results suggest that previous studies that did not address potential endogeneity of child BMI status were likely to have substantially underestimated the effect of BMI on healthcare utilisation or costs. Similarly, Cawley and Meyerhoefer (2012) found that studies that did not address endogeneity in adult BMI status likely substantially underestimated the effect of adult BMI on medical care costs in the United States. Our results differ from some previous studies given that we did not find a significant relationship between healthcare utilisation amongst younger children (Lynch et al., 2015; Bianchi-Hayes et al., 2015; Hayes et al., 2016). This may point to differences in the way

E. Doherty et al. / Economics and Human Biology 27 (2017) 84–92

that different healthcare systems are organised and in the way that they treat overweight or obese children. In Ireland, unlike most European countries, out-of-pocket payments are required for the majority of citizens to cover user fees for GP services. The exception to this are individuals who are eligible for a medical card, typically on the basis of an income-means test, which allows for free access to GP care. Notably within this context, the Irish government have explicitly recognised the inequities of the Irish healthcare system and have identified as a key goal the introduction of free GP care to all children until they are 18 years old. To date this goal has been partially fulfilled with, since 2014, free GP care introduced for all children under 6 years in Ireland. While this is projected to lead to a rise in the use of GP services by children, it may also result in more overweight and obese children accessing healthcare services at an earlier age. However, until the relevant data becomes available, it is only possible to speculate on what these impacts might be. Our study has a number of strengths. As well as being one of the first studies to estimate the causal effect of child BMI status on healthcare usage, it also benefits from the availability of independently measured BMI for mothers and children, rather than measures based on self-reported weight and height. This is important as there is substantial evidence that questions the accuracy of self-reported measures of BMI (Gosse, 2014; Elgar and  ska et al., 2015), which can substantially Stewart, 2008; Łopuszan bias estimates of healthcare usage. Moreover, a number of previous studies have recommended against the use of self-reported data for children under the age of 14 (Jansen et al., 2006; Beck, 2012; Himes and Faricy, 2001; Seghers and Claessens, 2010). A further strength of our paper is that it examines healthcare utilisation at two points in time for children, which allows us to examine whether patterns of utilisation differ by overweight or obese status at both ages and to compare patterns of utilisation for those who were overweight or obese at both ages to those who were overweight or obese at one age only. This latter analysis is particularly informative for policy makers as it goes to highlight the nature of the pathway through which such children (aged between 9 and 13 years) interact with the healthcare system. Finally, we employ an extremely rich and nationally representative sample allowing us to control for many covariates and factors that may explain healthcare utilisation. Our results, therefore, allow us to provide a comprehensive assessment of the effect of child BMI status on healthcare utilisation during childhood (aged 9) and early adolescence (aged 13). In terms of limitations, there was a proportion of respondents for which there was a non-response on BMI-related variables. While these non-response rates are comparatively small they could bias the results if individuals at the higher end of the BMI distribution did not complete the survey. Also, evidence from other cohorts suggests that higher BMI is associated with higher attrition rates, which could bias estimates. While this could be an issue in the current study, one reassuring finding is that we do not observe a fall in the proportion of mothers or children in the overweight and obese categories in wave 2 of the study.9 We also applied sampling weights to try and correct for potential problems associated with attrition. Another limitation of the study is that we have not examined the frequency of contact but rather only whether the child has used a particular healthcare service or not. While this might appear as an overly simplistic analysis, we were primarily interested here in understanding whether children with higher BMI were more likely to engage with the healthcare service. In further analysis, it might be useful to examine the frequency of interaction, as well as extending the analysis to examine the use of

9 We observe the proportions in the overweight and obese categories increasing in wave 2.

91

additional types of healthcare services. Furthermore, while the models presented in this paper control for a wide range of child, mother and household characteristics, some of these variables (such as child health status, medical card status etc.) may themselves be endogenous and therefore their estimates should be interpreted with caution. To ensure that our results on BMI were not affected by the presence of additional endogenous variables as regressors, we ran robustness checks where we removed these controls. A final limitation to our analysis is that it was constrained by the availability of only two waves of data. Ideally data covering a longer time horizon would be employed, which could allow for the effects of childhood overweight/obesity on healthcare utilisation to be considered in late adolescence and early adulthood. The findings from this paper raise important questions for policy makers and healthcare professionals alike. In particular, the healthcare burden of child overweight and obesity may be much more significant than previously thought and the timing of its impact appears to manifest itself in early adolescence rather than in early childhood. Taken together, this evidence may be suggestive of the need for a more proactive policy approach to tackling overweight and obesity in early childhood in primary care, as this may go to reduce the costs of GP services and more expensive hospital care in early adolescence. Evidence on the clinical and cost effectiveness of such a policy approach would be required. More broadly, our findings are particularly relevant given that the Irish healthcare system has gone through a significant period of budget cuts as a result of the recent economic recession. Moreover, projected demographic changes towards an ageing population are also expected to significantly impact on the healthcare system. While the economic costs of adult overweight and obesity have been recognised (Dee et al., 2015), until now there has not been a realisation of the consequences to the healthcare system of childhood overweight and obesity in Ireland. The weight of evidence internationally also suggests that children who are overweight or obese incur significantly more lifetime costs, both on the healthcare system and on the wider economy, relative to normal weight children (Finkelstein et al., 2014; Sonntag et al., 2016). Furthermore, a potential worrisome finding in tackling childhood overweight and obesity is that previous studies have shown that a substantial portion of Irish parents misclassify their children as normal weight when they are actually either overweight or obese (Cullinan and Cawley, 2017). This makes it more difficult for policy makers to target parents to help stem the increases in the rates of childhood overweight and obesity. Cawley (2015) discusses a number of potential avenues to help combat the observed rises in obesity that could be applied during childhood. The results from this paper suggest that if policies are implemented that can reduce the rates of childhood overweight and obesity, this could significantly benefit the healthcare system.

Acknowledgement Michelle Queally was supported by a Health Research Board Interdisciplinary Capacity Enhancement Postdoctoral Fellowship (ICE-2015-1026). References  ska, M., Lipowicz, A., Kołodziej, H., Szklarska, A., Bielicki, T., 2015. SelfŁopuszan reported versus measured body height and weight in Polish adult men: the risk of underestimating obesity rates. Anthropol. Anz. 72, 263–277. Andersen, R., Newman, J.F., 2005. Societal and individual determinants of medical care utilization in the United States. Milbank Q. 83 (4) doi:http://dx.doi.org/ 10.1111/j.1468-0009.2005.00428.x. Au, N., 2012. The health care cost implications of overweight and obesity during childhood. Health Serv. Res. 47, 655–676.

92

E. Doherty et al. / Economics and Human Biology 27 (2017) 84–92

Batscheider, A., Rzehak, P., Teuner, C.M., Wolfenstetter, S.B., Leidl, R., von Berg, A., Berdel, D., Hoffmann, B., Heinrich, J., 2014. Development of BMI values of German children and their healthcare costs. Econ. Hum. Biol. 12, 56–66. Beck, J., 2012. Accuracy of self-reported height and weight in children aged 6 to 11 years. Prevent. Chronic Dis. 9, E119. Bianchi-Hayes, J., Calixte, R., Huang, J., Cataldo, R., Wong, A., Pati, S., 2015. Healthcare utilization by obese and overweight children. J. Pediatr. 166, 626–631. Breitfelder, A., Wenig, C.M., Wolfenstetter, S.B., Rzehak, P., Menn, P., John, J., Leidl, R., Bauer, C.P., Koletzko, S., Röder, S., 2011. Relative weight-related costs of healthcare use by children—results from the two German birth cohorts, GINIplus and LISA-plus. Econ. Hum. Biol. 9, 302–315. Carey, F.R., Singh, G.K., Brown III, H.S., Wilkinson, A.V., 2015. Educational outcomes associated with childhood obesity in the United States: cross-sectional results from the 2011–2012 National Survey of Children’s Health. Int. J. Behav. Nutr. Phys. Activity 12, S3. Cawley, J., Meyerhoefer, C., 2012. The medical care costs of obesity: an instrumental variables approach. J. Health Econ. 31, 219–230. Cawley, J., 2004. The impact of obesity on wages. J. Hum. Resour. 39, 451–474. Cawley, J., 2015. An economy of scales: a selective review of obesity's economic causes, consequences, and solutions. J. Health Econ. 43, 244–268. Cole, T.J., Bellizzi, M.C., Flegal, K.M., Dietz, W.H., 2000. Establishing a standard definition for child overweight and obesity worldwide: international survey. BMJ 320, 1–6. Cullinan, J., Cawley, J., 2017. Parental misclassification of child overweight/obese status: the role of parental education and parental weight status. Econ. Hum. Biol. 24, 92–103. Cullinan, J., Gillespie, P., 2016. Does overweight and obesity impact on self-Rated health? evidence using instrumental variables ordered probit models. Health Econ. 25, 1341–1348. Dee, A., Callinan, A., Doherty, E., O'Neill, C., McVeigh, T., Sweeney, M.R., Staines, A., Kearns, K., Fitzgerald, S., Sharp, L., 2015. Overweight and obesity on the island of Ireland: an estimation of costs. BMJ Open 5, e006189. Elgar, F.J., Stewart, J.M., 2008. Validity of self-report screening for overweight and obesity: evidence from the Canadian Community Health Survey. Can. J. Public Health/Revue Canadienne de Sante'e Publique 423–427. Finkelstein, E.A., Trogdon, J.G., 2008. Public health interventions for addressing childhood overweight: analysis of the business case. Am. J. Public Health 98, 411–415. Finkelstein, E.A., Graham, W.C.K., Malhotra, R., 2014. Lifetime direct medical costs of childhood obesity. Pediatrics 133, 854–862. Gosse, M., 2014. How accurate is self-reported BMI? Nutr. Bull. 39, 105–114. Hampl, S.E., Carroll, C.A., Simon, S.D., Sharma, V., 2007. Resource utilization and expenditures for overweight and obese children. Arch. Pediatr. Adolesc. Med. 161, 11–14. Hayes, A., Chevalier, A., D'Souza, M., Baur, L., Wen, L.M., Simpson, J., 2016. Early childhood obesity: association with healthcare expenditure in Australia. Obesity 24, 1752–1758. Himes, J.H., Faricy, A., 2001. Validity and reliability of self-reported stature and weight of US adolescents. Am. J. Hum. Biol. 13, 255–260. Jansen, W., Van de Looij-Jansen, P., Ferreira, I., De Wilde, E., Brug, J., 2006. Differences in measured and self-reported height and weight in Dutch adolescents. Ann. Nutr. Metab. 50, 339–346. John, J., Wenig, C.M., Wolfenstetter, S.B., 2010. Recent economic findings on childhood obesity: cost-of-illness and cost-effectiveness of interventions. Curr. Opin. Clin. Nutr. Metab. Care 13, 305–313. John, J., Wolfenstetter, S.B., Wenig, C.M., 2012. An economic perspective on childhood obesity: recent findings on cost-of-illness and cost-effectiveness of interventions. Nutrition 28, 829–839. Kline, B., Tobias, J.L., 2008. The wages of BMI: Bayesian analysis of a skewed treatment–response model with nonparametric endogeneity. J. Appl. Econ. 23, 767–793. Layte, R., McCrory, C., 2011. Growing Up in Ireland-National Longitudinal Study of Children: Overweight and Obesity Among 9-Year-Olds. GUI, Dublin.

Layte, R., Nolan, A., 2015. Eligibility for free GP care and the utilisation of GP services by children in Ireland. Int. J. Health Econ. Manage. 15, 3–27. Llewellyn, A., Simmonds, M., Owen, C., Woolacott, N., 2016. Childhood obesity as a predictor of morbidity in adulthood: a systematic review and meta-analysis. Obes. Rev. 17, 56–67. Lynch, B.A., Rutten, L.J.F., Jacobson, R.M., Kumar, S., Elrashidi, M.Y., Wilson, P.M., Jacobson, D.J., Sauver, J.L.S., 2015. Health care utilization by body mass index in a pediatric population. Acad. Pediatr. 15, 644–650. McCrory, C., Layte, R., 2012. Breastfeeding and risk of overweight and obesity at nine-years of age. Soc. Sci. Med. 75, 323–330. Murray, A., McCrory, C., Thornton, M., Williams, J., 2009. Growing Up in Ireland National Longitudinal Study of Children Technical Report 1: Design, Instrumentation and Procedures for the Child Cohort. GUI, Dublin. Murray, L., Halligan, S., Cooper, P., 2010. Effects of postnatal depression on mother– infant interactions and child development, 2nd edition The Wiley-Blackwell Handbook of Infant Development, 2. , pp. 192–220. Ng, M., Fleming, T., Robinson, M., Thomson, B., Graetz, N., Margono, C., Mullany, E.C., Biryukov, S., Abbafati, C., Abera, S.F., 2014. Global, regional, and national prevalence of overweight and obesity in children and adults during 1980–2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet 384 (9945), 766–781. Nolan, A., Smith, S., 2012. The effect of differential eligibility for free GP services on GP utilisation in Ireland. Soc. Sci. Med. 74, 1644–1651. Pelone, F., Specchia, M.L., Veneziano, M.A., Capizzi, S., Bucci, S., Mancuso, A., Ricciardi, W., de Belvis, A., 2012. Economic impact of childhood obesity on health systems: a systematic review. Obes. Rev. 13, 431–440. Rokholm, B., Baker, J.L., Sørensen, T.I.A., 2010. The levelling off of the obesity epidemic since the year 1999–a review of evidence and perspectives. Obes. Rev. 11, 835–846. Roodman, D., 2009. Estimating fully observed recursive mixed-process models with cmp. Stata J. 11, 159–206. Rosner, B., Cook, N.R., Daniels, S., Falkner, B., 2013. Childhood blood pressure trends and risk factors for high blood pressure the NHANES experience. Hypertension 62, 247–254. Seghers, J., Claessens, A.L., 2010. Bias in self-reported height and weight in preadolescents. J. Pediatr. 157, 911–916. Skinner, A.C., Mayer, M.L., Flower, K., Weinberger, M., 2008. Health status and health care expenditures in a nationally representative sample: how do overweight and healthy-weight children compare? Pediatrics 121, e269–e277. Skinner, A.C., Perrin, E.M., Skelton, J.A., 2016. Prevalence of obesity and severe obesity in US children, 1999–2014. Obesity 24, 1116–1123. Smith, G.D., Sterne, J.A., Fraser, A., Tynelius, P., Lawlor, D.A., Rasmussen, F., 2009. The association between BMI and mortality using offspring BMI as an indicator of own BMI: large intergenerational mortality study. BMJ 339, b5043. Sonntag, D., Ali, S., Lehnert, T., Konnopka, A., Riedel-Heller, S., König, H.H., 2015. Estimating the lifetime cost of childhood obesity in Germany: results of a Markov Model. Pediatr. Obesity 10, 416–422. Sonntag, D., Ali, S., De Bock, F., 2016. Lifetime indirect cost of childhood overweight and obesity: a decision analytic model. Obesity 24, 200–206. Stock, J.H., Wright, J.H., Yogo, M., 2012. A survey of weak instruments and weak identification in generalized method of moments. J. Bus. Econ. Stat. 20, 518–529. Thornton, M., et al., 2016. Growing Up in Ireland: Design, Instrumentation and Procedures for the Child Cohort at Wave Two (13 Years). GUI, Dublin. Trasande, L., Chatterjee, S., 2009. The impact of obesity on health service utilization and costs in childhood. Obesity 17, 1749–1754. Uppal, V., Mansoor, S., Furuya, K.N., 2016. Pediatric non-alcoholic fatty liver disease. Curr. Gastroenterol. Rep. 18, 1–9. Walsh, B., Cullinan, J., 2015. Decomposing socioeconomic inequalities in childhood obesity: evidence from Ireland. Econ. Hum. Biol. 16, 60–72. Wenig, C.M., 2012. The impact of BMI on direct costs in Children and Adolescents: empirical findings for the German Healthcare System based on the KiGGSstudy. Eur. J. Health Econ. 13, 39–50. World Health Organization, 2016. Report of the Commission on Ending Childhood Obesity. (Available at http://www.who.int/en/).