The impact of childhood obesity on human capital in high‐income countries: A systematic review

Current evidence of the impact of childhood obesity on human capital development does not point in a consistent direction, and its interpretation is challenging. We carried out a systematic review of studies from high‐income countries that used robust causal inference approaches to assess the impact of childhood overweight and obesity on outcomes typically linked to human capital development in economics. Global Health, Medline and EconLit were used to search for peer‐reviewed papers. Three reviewers independently assessed study quality using the Newcastle‐Ottawa Scale. Nineteen papers representing 22 studies met the inclusion criteria. Included studies were categorized based on three components of human capital: cognitive performance (n = 18), measured through test scores; educational attainment (n = 3), through grade progression and college completion; and labour market outcomes (n = 1), through wages. We find that childhood overweight and obesity hinder education outcomes, with effects mostly observed at older ages of exposure measurement (12+ years). Girls with overweight and obesity experienced larger negative effects and more often than boys. Future research should elucidate the pathways through which childhood obesity impacts human capital development, to support policies that may mitigate those impacts, thus averting social costs that are currently widespread, increasing and unaccounted for.


| INTRODUCTION
The global number of children with overweight or obesity has increased more than tenfold, from 11 million in 1975 to 124 million in 2016. 1 Evidence suggests that the longer a person has obesity, the larger the excess morbidity and costs are in adulthood, with costs ranging from increased healthcare costs to reduced productivity in the workforce as a result of increased sick days. 2 The global economic cost of obesity was estimated at $2.0 trillion US dollars in 2012, including the loss of productive life, direct healthcare expenditures and the investments to lessen these costs. 3 For example, employees with obesity are likely to be less productive at work due to increased health issues (i.e. arthritis, fatigue, or depression) and related work absenteeism. 2 Investment costs include expenditures on public health programmes as well as commercial weight management and fitness products and plans. 4 Further, obesity is an intergenerational phenomenon, meaning that obesity is likely transmitted from parent to child in a cyclic manner. 5 Given the life-course and intergenerational effects of childhood overweight and obesity, governments around the world have been devising policies and programmes to curb the childhood obesity epidemic. 6,7 The case for government action to address childhood obesity would be further strengthened by robust evidence of its detrimental impacts on key social and economic outcomes such as education, employment or social capital, helping also to identify the best periods in which to intervene. 8 For ease of reading, we use the term 'childhood obesity' to refer to childhood and adolescent overweight and obesity (unless otherwise specified). Therefore, childhood obesity refers to anyone who is under 18 years of age and has an age-and sex-adjusted body mass index (BMI) z-score greater than or equal to the 85th percentile. While there is reasonably established evidence of the effects of childhood obesity on later health, evidence of the effects on social and economic outcomes is mixed and largely relies on studies that use less robust inference designs. 2,[9][10][11][12] Human capital is defined as 'the agency of human beingsthrough skill and knowledge as well as effort-in augmenting production possibilities.' 13 (p1959) Human capital development is represented by a cumulative production function framework that combines cognitive and noncognitive inputs. [14][15][16][17] We have conceived this systematic review in line with a human capital theoretical approach in which an individuals' social and economic outcomes are the result of a dynamic and cumulative process of human capital development over his or her life-course. 18 As such, human capital includes skills and knowledge that give an individual returns-be it economically or socially-that allow them to be valuable in a workforce. A recent review concludes that early childhood circumstances can have relatively substantial negative impacts in adulthood, though impacts are heterogeneous due to differences in the child's inputs and family environments. 19 In addition, research finds strong evidence of a negative association between childhood health and later socioeconomic outcomes (mainly educational and employment-related outcomes), but evidence on the longterm effects, especially those of childhood obesity, remains undeveloped. [19][20][21] This is because estimating the effects of childhood obesity on human capital outcomes is complicated by a range of potential observed and unobserved confounders and mediators (i.e., socioeconomic status, parents' education, obesogenic environments, and genetic makeup), issues of reverse causation and uncertainty over the time lag between exposure and outcome. 20,21 There is no clear consensus on the effect of childhood obesity on human capital. While previous literature suggests a negative relationship between adult obesity and labour market outcomes, the magnitude and statistical significance of this relationship depends on the gender and race or ethnicity of participants. [22][23][24][25][26] Literature on the effects of childhood obesity and educational outcomes suggests a negative relationship-though again, results vary according to gender, race/ethnicity, age and location of participants. [27][28][29][30] Some studies report no significant effects as well-demonstrating that the results are sensitive to the specific context of the data and model specifications used. [31][32][33] A study by Palermo and Dowd 34 used fixed effect models to investigate the effect of childhood obesity on cognitive and noncognitive outcomes and concluded that obesity in children and adolescents negatively affects noncognitive but not cognitive outcomes. 34 In 2011, a review by Suhrcke and de Paz Nieves 20 concluded that obesity and overweight are negatively related with negative educational outcomes though the 'evidence is contradictory concerning the gender-differentiated effect of these risk factors, and endogeneity issues also persist as obstacles in the estimation of causality.' 20 (p13) In the latter review, only a limited number of studies used longitudinal data and implemented econometric methods to control for the biases produced both by confounding and reverse causality. However, in most reviews, the majority of included studies were cross-sectional, meaning that they are prone to confounding, including from reverse causation. For example, the review by Caird et al 35 concluded that increased weight in childhood and adolescence was weakly related with decreased educational attainment. However, 3 years later, a different review concluded that the relationship between obesity and academic achievement was not clear-except for adolescent females, who experienced a negative relationship. 36 This, in addition to the inclusion of cross-sectional studies, means that the evidence on the effect of obesity on educational outcomes remains contradictory.
Another reason for the uncertainty of the effect of childhood obesity on human capital is that no previous reviews have investigated the impact on human capital as a multifaceted concept, including education and labour market outcomes, as the majority of the literature reviews on the topic included only educational outcomes. The review by Gondek et al 37 is the most similar to this review, though they focused on the social and economic impacts associated with any type of ill health (not just obesity) at any life stage (not just childhood). The authors also included studies that used longitudinal data, regardless of whether they used a rigorous framework or not. 37 Altogether, past reviews have shown inconsistent findings on the association between childhood obesity and subsequent outcomes depending on the age of exposure and methods used. We thus contribute to the literature by focusing on childhood obesity only and its effect on multiple human capital outcomes, and by only reporting results from studies with a sound inferential design, as defined in Section 2.2.
The aim of this review is to assess whether existing studies provide evidence that childhood obesity has an impact on human capital on middle and late childhood (including adolescence) or adulthood.
The review is designed to summarize the evidence generated by studies based on causal inference approaches and assess the strength of this evidence. By focusing on the effect of obesity in childhood on future outcomes, our review sheds light into the effect of obesity on later human capital. were searched, as well as the free text. Other keywords for weight status like 'overweight' are included in the medical subject headings term obesity ( Figure S1). To capture human capital and the variety of associated social and economic outcomes of interest, we included keywords for cognitive performance (test scores, literacy, mathematics), educational achievement (highest academic qualification, educational attainment), labour market outcomes (employment, unemployment, wages, employment disability), social capital (social relationships, interpersonal relationships, partnership status, social support, trust) and, finally, social participation (social engagement, voluntary work, membership to organization, voting). We note that while our original research question included social participation and social capital outcomes, no relevant papers were found. Thus, our review is solely on human capital outcomes.

| Selection criteria
The searches were restricted to peer-reviewed studies written in English published after 1980. Conference papers, dissertations, metaanalyses and working papers were excluded. Included studies had to assess the impact of childhood obesity on a wide set of outcomes including educational and labour market outcomes. Examples of educational outcomes can include standardized cognitive ability scores, grade point averages (GPAs) and subject scores. Cognitive performance is most often evaluated with ability tests 38 ; it is a measure of function in various cognitive domains (i.e., memory or executive function) and a proxy for educational achievement. We acknowledge that outcomes such as GPA may reflect more than an objective measurement of a student's cognitive performance, most likely capturing a child's classroom homework and participation scores, which could be subjected to teacher's bias. However, because scores like GPA are mostly an average of exam scores, and hence reported as a continuous scale, we group GPA as a cognitive performance outcome.
Educational attainment measures a students' academic progression or attainment. It includes domains such as high-school completion, higher-education completion, formal qualifications and degrees.
Labour market outcomes considered here include aspects associated with participation and productivity, such as employment, unemployment, work-related absences and wages or salaries. Included studies had to use longitudinal data and use methods that accounted for both reverse causation and unobserved confounding. Since our review draws on literature from different fields, such as economics, epidemiology and medical, we did not want to limit our search strategy by including specific methods. The aforementioned fields have different definitions of 'causal' inference methods, which is why we did not prespecify the types of inferential methods we were going to include in our search strategy (see Table S1). Instead, our goal with our search criteria was to be able to capture any quantitative study that met the rest of our criteria of having longitudinal data and took steps to deal with issues of reverse causation and unobserved confounding. Observational studies are excluded from the main analyses of this review as they lack robust and rigorous inference methods to control for the endogeneity of childhood obesity.
Only studies assessing childhood overweight and/or obesity, in children up to age 18 years, were eligible for inclusion. We excluded secondary analyses of datasets from low-and middle-income countries because of the rapidly transforming relationship between prevalence of childhood and adolescent overweight and obesity and the socioeconomic environment that the child is exposed. For example, a 2013 review concluded that the relationship between educational attainment and obesity is impacted by country's economic development level; high-income countries exhibit an inverse relationship, whereas lower-income countries exhibit a positive association between obesity and educational achievement. 39 We used the World Bank list of economies, last updated in June 2018, to identify the income level of the countries under study. 40 Further, studies of children with other comorbid conditions, such as type 2 diabetes, were excluded to avoid the possibility of confounding the relationships of interest. The complete inclusion and exclusion criteria are provided inTable S1.

| Data extraction
Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses, Figure 1 presents the flow diagram of the screening process. 41 The complete search strategy produced 4,168 articles after removing duplicates. After screening by title, this was narrowed down to 600 articles. Two reviewers screened title and abstracts for relevance. Concurrence was 90%. Disagreements were resolved through discussion and, when needed, by a third reviewer. After screening by abstract and title, two reviewers completed a full-text review on the 268 remaining articles. The interrater agreement for full-text review was assessed using Cohen's kappa (κ = 0.76). After full-text screening, 19 papers were selected for this review, representing 22 studies.

| Data synthesis
The 22 studies included in this review were categorized by the three distinct outcomes of interest: cognitive performance, educational attainment and labour market outcomes. Therefore, this manuscript is focusing on education and employment outcomes only. It was not possible to undertake a formal meta-analysis because of a high degree of heterogeneity in study populations, study designs and outcomes.

| Risk of bias
Three reviewers (ABS, MCH and EA) worked independently to appraise the methodological quality and risk of bias information for all papers using a modified Newcastle-Ottawa Quality Assessment Form for Cohort Studies (NOS); disagreements were resolved through discussion between the three reviewers. 42 The NOS has been recommended to assess the methodological quality of nonrandomized studies of the effect of interventions in systematic reviews using three domains-selection, comparability and outcome. 42 We followed the authors of various systematic reviews on obesity who modified the NOS. 37,[43][44][45] The original NOS has four criteria in the 'Selection' domain and a comparability domain, but because our exposed and unexposed participants come from the same cohort, we did not include these domains. Instead, we follow Gondek et al 37 and create an 'Adjustment' domain to evaluate the study's control of confounders, which consists of two questions. We did not assess papers according to the 'Design' domain because all our studies were longitudinal and therefore this domain was inapplicable for our review. 37 Studies were awarded a maximum of three points in the selection domain, based on the representativeness of the sample (internal validity), ascertainment of exposure (child weight status measured) and evidence that the outcome of interest was controlled for at baseline (i.e., academic achievement balanced between participants with and without overweight or obesity at baseline). Papers were awarded a maximum of two points for adjustment. One of these points could be F I G U R E 1 Preferred reporting items for systematic reviews and meta-analyses flow chart of study selection earned if the study controlled for the main confounders associated with childhood overweight and respective outcomes (sex, gender, socioeconomic status, parent or home inputs and intelligence). The other point was given if the paper included covariates that could mediate the relationship of interest, such as mental health and selfefficacy (in addition to including basic covariates such as age, gender, socioeconomic status). [46][47][48] Finally, a paper could earn two points in the outcome domain: one for ascertainment of outcome (objective measurements of cognitive performance or educational attainment) and one for reporting the sample attrition rates and/or discussing the implications of follow up rates.

| RESULTS
This review is based on 19 eligible papers that equate to 22 relevant studies as three papers cover more than one cohort or outcome. Of these, there are 18 studies on cognitive performance, three on educational attainment and one on labour market outcomes. Table 1 provides an overview of the results. Key summary information for each paper with the respective reference, including methods, data, location, outcome and exposure details, growth reference chart used, overall results and risk of bias, is presented in Table 2. The sample characteristics of each study including sample size, sex distribution, age at baseline and follow up, obesity prevalence at baseline, period considered, number of data waves used in analyses and results categorized by outcome are summarized in Table 3. We present point estimates from the studies that report significant findings in Table S2. We include point estimates only where studies report significant findings for their most robust models-even when this model is not the author's pre-  18 (17) Risk of bias

| Educational attainment
Three studies investigated the impact of childhood obesity (measured at ages ranging from 9 through 17 years) on later educational attainment (measured at ages ranging from 12 through 30 years). As per cognitive performance, both exposure and outcome variables varied across studies, which makes it difficult to compare the effect sizes of associations. Two studies measured exposure at age 12 years: one included both continuous (pounds and BMI) and categorical (overweight including obese) outcomes, reporting significant negative effects for both males and females, but stronger for the latter. 60 The other used overweight and obese categories separately, finding significant negative effects only for White and Asian-American females with overweight and obesity. 57 The third educational attainment study used obesity measured at age 7 to 9 years as the exposure of interest, without differentiating by sex, and did not find significant effects. 31 Educational attainment measures vary across studies, including students' grade retention, university completion and probability of graduating from high school, which may contribute to the heterogeneity of findings.

| Labour market outcomes
Only one study included in this review measured the effect of childhood obesity on later labour market outcomes. 58 64 suggesting that children with obesity in middle childhood were likely to remain with obesity in adolescence and therefore experience the negative effects.

| Sex as an effect modifier
Existing evidence consistently shows that sex may be an effect modifier in the relationship between childhood obesity and education outcomes. Specifically, females experienced larger negative effects compared with their male peers. Sixteen studies disaggregated their analyses by sex. Of these 16 studies, seven studies reported nonsignificant effects, though all of these studies had used a young age of exposure that could explain these nonsignificant effects. More importantly, nine of the sex-stratified studies reported significant effects, and of these, the majority (n = 6) reported that this effect was either more significant, larger or only observed by females.
Five studies presented their results with and without disaggregating by sex, which allowed us to further investigate the effect of sex. Three of these studies reported no significant effect, though all three used young exposure ages. The other two of these studies reported significant negative effects for the full sample then found differential negative effects when disaggregating by sex. The study by Ding et al 17 found an overall significant negative relationship between obesity and GPA, but when disaggregated by sex, the effect completely attenuated for males while the significant negative effect became slightly larger for females. The other study that reported significant negative results found that the effect became larger and more significant for females when disaggregating by sex. 56 Our findings regarding the differential impact of sex on the relationship of interest T A B L E 3 (Continued) is in line with the findings of the reviews by Cohen et al 39 and Martin et al. 36 Martin et al 36 concluded that the association between childhood obesity and academic achievement varied by sex, age and school subject. This review showed a significantly negative association between overweight and maths achievement for adolescent females, but not for younger females and males. 36 The review Cohen et al 39 found a negative association between educational attainment and obesity, with 'stronger social patterning among women.' 39 (p989) 3.4 | Differences by race and ethnicity

| Study quality (risk of bias)
Results of the risk of bias assessment are available in Table S3. Two of the three low-risk papers reported significant negative results. Lowrisk papers used objective procedures to secure child exposure weight and outcome measurements. These papers also verified low risk of selection bias by showing that the outcome difference was not present at baseline or was appropriately controlled for (i.e., by regressing the child's baseline cognitive performance score in the equation).
Additionally, these papers appropriately adjusted for both known and 'extra' confounders related to children's mental health. Low-risk papers also recognized and appropriately addressed issues of cohort attrition bias. As all three of these low-risk papers used weight exposure as measured between about ages 5 and 10 years, future studies should investigate using appropriate methods with later ages of exposure.
Most of the evidence in this review was at medium risk of bias,  53,65 The finding that children with overweight or obesity measured at older ages, such as in early-or mid-adolescence, are more likely to suffer negative effects also resulted from some of the previous reviews. 39 One possible explanation is that the impacts of childhood obesity are cumulative over time; a hypothesis also embodied in what is commonly referred to as the 'life course' approach. 66  did. Overweight is defined as having an age-and sex-adjusted BMI zscore greater than or equal to the 85th percentile; obesity is when it is greater than or equal to the 95th percentile. The main consequence of using different definitions is that the prevalence of obesity and overweight in the sample would change depending on the definition chosen. With a more restrictive definition, only the children with the higher BMI would be considered overweight or obese, meaning that the effects are more likely to be ascertained, but the reduced prevalence makes the results less likely to be statistically significant. However, when the prevalence of obesity is low, regardless of the definition used, those who have obesity may be more stigmatized, which may contribute to explaining why studies detected significant effects even with relatively low obesity rates, such as 6% in the study by Black et al 50 and 8% in Ding et al. 17 Many studies and reviews overlook the prevalence of obesity at exposure, but considering the sample's baseline obesity prevalence is important in interpreting results.

| Generational differences
Generational differences between cohorts are a cause of heterogeneity across studies. There is a 42-year gap in the year of birth of some of the cohort members in the reviewed studies. Not only does the prevalence of childhood obesity change, but also the social norms and expectations placed on cohort members' outcomes (e.g., educational attainment). As a result, the relationship between exposure and outcome of interest may be significantly different for a child in the NLSY79 cohort, born in 1957, compared with a child from the ECLS-K or ALSPAC cohorts, born in the early 1990s.

| Theoretical frameworks
The studies assessed in this review relied on different theoretical frameworks, and most were not informed by any frameworks, which likely contributed to inconsistencies in methods and results. Kaestner and Grossman 31 used a standard economic model of child quality and concluded that childhood obesity was not significantly associated with cognitive performance nor educational attainment outcomes, instrumenting children's weight status with lagged weight to obtain a model consistent with their production function. They argued that other studies reached different conclusions because those studies 'used empirical specifications that were not consistent with common theoretical formulations of the education (human capital) production function.' 31 (p660)

| CONCLUSION
This systematic literature review provides evidence that childhood obesity can negatively affect human capital development. This evidence is disproportionately based on US data, educational outcomes and entirely on data from high-income countries. Our evidence is also limited by the type of study outcomes we found, which may mean our results are not generalizable for all dimensions of human capital.