Associations Between Different Cortisol Measures and Adiposity in Children: A Systematic Review and Meta-Analysis

This study examined associations between hair, salivary, serum, and urinary cortisol concentration with adiposity-related indicators in children, and explored their potential effects modification by age, sex, cortisol measurement method, and country developmental context. We systematically searched PubMed, Web of Science, and Embase for studies examining at least one of the four aforementioned cortisol with objectively measured adiposity-related outcomes in children. Meta-analyses of cross-sectional studies revealed that hair cortisol concentration was associated with fat mass index (FMI)-standard deviation score (SDS)/FMI z-score (pooled-β = 0.04, 95% CI: 0.01, 0.08) and BMI/BMI z-score (pooled-β = 0.15, 95% CI: 0.06, 0.25), and these associations were significant among children aged ≤ 12 years (pooled-β = 0.15, 95% CI: 0.05, 0.26) and >12 years (pooled-β = 0.13, 95% CI: 0.04, 0.22), children from developed countries (pooled β = 0.12, 95% CI: 0.03, 0.21) and developing countries (pooled-β = 0.193, 95% CI: 0.188, 0.198), and in studies extracting cortisol via LC-MS/MS (pooled-β = 0.18, 95% CI: 0.06, 0.29) but not ELISA (pooled-β = 0.08, 95% CI: −0.06, 0.22). Meta-analyses of both cohort and cross-sectional studies revealed non-significant associations of morning salivary cortisol concentration and total daily cortisol output with BMI/BMI z-score. Serum cortisol concentration was not associated with BMI or waist circumference. Meta-analysis of urinary cortisol concentration and adiposity was hindered by insufficient data. These findings further corroborate understanding of chronic stress’ physiological contribution to increased pediatric obesity risk. Systematic Review Registration [https://www.crd.york.ac.uk/prospero/#recordDetails], identifier [CRD42020215111].


INTRODUCTION
Childhood obesity persists as a global public health crisis (1)(2)(3)(4). Recent research has identified stress as an important risk factor for childhood adiposity (5)(6)(7)(8). Stress is a negative emotional experience accompanied by predictable biochemical, physiological, cognitive, and behavioral changes directed toward altering the stressful event or accommodating to its effects (7); such changes may further serve to increase childhood obesity risk (7). Measurement of stress is inherently complex and requires consideration of multiple dimensions, including the social, psychological, and physiological (9). Given the inherent limitations of using subjective, self-reported measures for stress, considerable literature has established the use of physiological biomarkers for the objective assessment of stress for research. However, associations between physiological measures of stress and adiposity-related indicators in children are inconsistent, preventing a unified understanding of the stress processes in childhood obesity and subsequent design of related interventions.
The hypothalamic-pituitary-adrenal (HPA) axis is the most widely studied physiological stress system. When an individual perceives stress, a physiological cascade occurs in the HPA axis, and its main downstream hormone "cortisol" has been viewed as the "gold standard" biomarker with which to assess stress (6,10). Alterations in HPA axis may be reflected in changes in the level and diurnal trajectory of cortisol secretion (11). Cortisol can facilitate obesity by stimulating unhealthy eating behaviors and promoting fat deposition (7). Moreover, visceral adipose tissue itself is rich in 11β-hydroxysteroid dehydrogenase type I, which converts inactive cortisone to cortisol (12). Therefore, a potential bidirectional relationship between cortisol and adiposity outcomes may exist. However, in this study, our primary focus will be placed on examinations of cortisol on adiposity outcomes in children.
It is possible for laboratories to utilize blood, urine, saliva, and hair to measure cortisol (13). For many years, cortisol was obtained primarily from serum or urine, but more recent approaches have sampled saliva and hair for less invasive monitoring of HPA functioning, and each measure reflects bodily cortisol levels. Serum cortisol concentration measures the total cortisol (14). Salivary cortisol concentration is usually used to assess the circadian rhythm of cortisol (e.g., cortisol awakening response) and the secretion of cortisol under stress-induced conditions (e.g., the total output of cortisol) (15). Urine samples will generally capture HPA activity over a period of only 24 h or less. In contrast, hair cortisol concentration (HCC) have increasingly been used to assess the long-term presence and/or accumulation of cortisol in children (16,17).
More research is needed to evaluate and understand the associations between different cortisol measures for stress with adiposity-related outcomes in children. However, the literature on such associations is very limited (18)(19)(20). To date, only one systematic review (of n =26 studies) has provided the evidence on associations between HCC and obesity in children, finding a modest positive correlation between HCC and anthropometric measures including body mass index (BMI), BMI z-score, waist circumference (WC), and body fat (21). However, the meta-analyses of reviewed studies did not exclude those relying on self-reported weight status and did not distinguish between cross-sectional and longitudinal studies. Moreover, studies have suggested that individual (e.g., age and sex) and environmental contextual factors (e.g., country development status) may modify associations between cortisol and adiposity outcomes in children (22). For example, a study found that association between cortisol and increased BMI were stronger in early adolescence than in late adolescence (23). Another study showed that altered cortisol balance modified the net lipogenetic/lipolytic in various adipose tissue depots in a sex-dependent manner in the periphery, therefore contributing to the differential associations between cortisol and adiposity outcomes (24). Furthermore, lower socio-economic status of a country was a predictor of higher cortisol levels and obesity risk (25,26). These findings indicate that these background factors may modify the associations between cortisol and adiposity-related outcomes in children. Interestingly, no studies have heretofore examined how different cortisol measures may vary in their associations with obesity by different sociodemographic or socio-economic factors in children.
Therefore, this systematic review and meta-analysis aimed to examine associations of different cortisol measures -hair, saliva, serum, and urine -with various adiposity-related outcomes in children, and to further explore the potential modification of these associations by external contextual factors including child age, sex, cortisol measurement method, and country developmental context. These findings will synthesize the body of evidence surrounding associations between different cortisol measures and pediatric obesity, and advance the understanding of child stress biomarker research.

METHODS
This study was developed and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and other recommended practice standards (e.g., Johnson and Hennessy, 2019).

Literature Search
A systematic search was performed in three electronic bibliographic databases-PubMed, Web of Science, and Embase-for relevant studies published from inception to October 2021. We developed a search strategy for databases based on keywords of seminal articles we had previously identified. Search strategies included all combinations of terms related to cortisol, anthropometric measures, and children (Supplementary Table 1).
Hand searching of references was conducted to uncover any potentially overlooked studies. Articles identified from the reference lists were further screened and evaluated using the same study criteria. Reference searching was repeated on all newly identified articles until no additionally relevant articles were found.

Study Selection
Studies that met all of the following criteria were included: (1) was cross-sectional, case-control, or longitudinal; (2) studied children under 18 years old without mental disorders or any diagnosed chronic conditions (e.g., hypertension, cardiovascular disease); (3) examined naturally occurring cortisol, assayed from either urine, saliva, hair or blood, as exposure variables; (4) analyzed objectively measured adiposity-related outcomes; (5) reported statistical associations between cortisol and adiposity-related outcomes; (6) were published in English; and (7) were peerreviewed publications. When multiple articles reported on the same data, the article with the largest sample size and results most relevant to this review was retained. Two authors assessed all identified studies for eligibility independently and disagreements were resolved through discussion.

Data Extraction and Preparation
A standardized form was developed to collect information from selected studies. Data extracted included that on: (1) the study (e.g., first author, publication year, study design, cortisol measure[s] used, adiposity-related outcome[s] assessed, the country site of study, and the country site's developmental context [developed vs. developing]), (2) the sample (e.g., participant age, sex, race/ethnicity), and (3) effect sizes. Acceptable adiposity-related outcomes included BMI/BMI z-score/BMIstandard deviation score (BMI-SDS), waist circumference (WC), percentage body fat (PBF), fat mass index (FMI)-SDS/FMI z-score, free fat mass index (FFMI), and waist to height ratio (WtHR), and truncal distribution of fat mass (TDFM). Data were extracted independently by two authors and discrepancies were resolved through discussion.

Study Quality Assessment
Two authors independently assessed the quality of eligible articles using the U.S. National Heart, Lung, and Blood Institute's Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies (27). This assessment tool rates studies based on 14 criteria. For each criterion, a score of one was assigned for "yes" and zero otherwise (i.e., "no, " "not applicable, " "not reported, " or "cannot determine"). Overall quality was rated based on the total score of the scale, with 0-3, 4-7, and 7-14 reflecting poor, fair, and good quality, respectively. Discrepancies on study quality ratings were also resolved through discussion (Supplementary Table 2).

Statistical Analysis
A meta-analysis was performed to estimate the pooled associations between different cortisol measures and adiposityrelated outcomes in children. Study heterogeneity was assessed using the I 2 index and Tau-squared (T 2 ). The level of heterogeneity represented by I 2 was interpreted as modest (I 2 ≤ 25%), moderate (25% < I 2 ≤ 50%), substantial (50% < I 2 ≤ 75%), or considerable (I 2 > 75%) (28). A randomeffects model was applied because of assumed clinical and methodological heterogeneity among the studies (29).
Pre-specified subgroup analyses were conducted to test potential modifying effects of age, sex, country developmental context (i.e., developed vs. developing), and cortisol measurement method [i.e., enzyme-linked immunosorbent assay (ELISA) vs. liquid chromatography tandem-mass spectrometry (LC-MS/MS) vs. chemiluminescence immunoassay (CLIA) vs. electrochemiluminescence immunoassay (ECLIA) vs. Radioimmunoassay (RIA) vs. dissociation-enhanced lanthanide fluorescence immunoassay (DELFIA) vs. a time-resolved fluorescence immunoassay (TRFIA)]. Sensitivity analyses were conducted to investigate the influence of a single study on the overall pool estimation by omitting one study at a time.
Publication bias was assessed by visual inspection for symmetry/asymmetry of contour-enhanced funnel plots and Egger's tests. All statistical analyses were conducted in Stata 14 with specific meta-analysis commands (i.e., metan and metareg) (College Station, TX, United States). All analyses used two-sided tests and p < 0.05 was considered statistically significant.

Study Selection
The search identified 8,627 articles of which 38 (31 cross-sectional articles and seven longitudinal articles) were included in this systematic review, with a sample size of 18,667 children. Twentyfour articles were included in the meta-analysis (Figure 1). For testing potential modifying effects, nine (20,(30)(31)(32)(33)(34)(35)(36)(37) of the 24 articles were further divided into 18 separate studies given differences in age, sex, indicators of adiposity, and cortisol measurement method, thus, in sum, 33 separated studies were included for meta-analysis. Study characteristics are shown in Table 1.

Hair Cortisol Concentration and Adiposity-Related Outcomes Among Children
Nineteen articles encompassing 11,067 children reported on associations between HCC and adiposity-related outcomes, with three longitudinal articles, 11 articles among children aged ≤ 12 years old, 16 from developed countries, 13 using 3 cm hair samples, and ten extracting cortisol by ELISA and eight by LC-MS/MS. All articles measured BMI/BMI z-score/BMI-SDS, and six of them also measured WC, PBF, FMI-SDS/FMI z-score, and WtHR (Table 1).

Salivary Cortisol Concentration and Adiposity-Related Outcomes Among Children
Sixteen articles with 3,462 children examined associations between salivary cortisol concentration and adiposityrelated outcomes, including 13 cross-sectional articles and five longitudinal articles (two articles reported both crosssectional and longitudinal results). Fourteen of the 16 articles examined children ≤ 12 years old, twelve articles took place in developed countries, five articles examined cortisol as AUCi (area-under-the-curve-increase) and two reported AUCg (area under the curve with respect to ground), and eleven articles used ELISA for cortisol extraction. All these articles measured BMI/BMI z-score and four also measured WC and PBF ( Table 1).

Serum Cortisol Concentration and Adiposity-Related Outcomes Among Children
Six cross-sectional articles encompassing 4,265 children examined associations between serum cortisol concentration     TDFM was assessed with dual energy X-ray absorptiometry Correlation between urine cortisol and TDFM was: r = −0.28 for total children, r = 0.09 for boys and r = 0.25 for girls No covariates were reported BMI, body mass index; WC, waist circumference; PBF, percentage body fat; BMI-SDS, BMI standard deviation score; SDS, standard deviation score; FMI, fat mass index; FFMI, free fat mass index; WtHR, waist to height ratio; CDC, Centers for Disease Control and Prevention; CI, confidence interval; M, mean; NR, not reported;β, beta coefficient; ELISA, enzyme-linked immunosorbent assay; CLIA, chemiluminescence immunoassay; HPLC-MS/MS, high-performance liquid chromatography-tandem mass spectrometry; LC-MS/MS, liquid chromatography tandem mass spectrometry; ECLIA, electrochemiluminescence immunoassay; RIA, radioimmunoassay; DELFIA, dissociation-enhanced lanthanide fluorescence immunoassay; TRFIA, a time-resolved fluorescence immunoassay; TSST-C, Trier Social Stress Test for Children; AUC, area under the curve with respect to ground; AUCi, saliva cortisol area-under-the-curve-increase. # The two cohort studies that showed associations between hair cortisol and weight status were not included in the meta-analysis because the explanation of effect size (β) is different. # These studies were not included in the meta-analysis because the meaning of effect size (β) was different as those of the other studies. *These studies were not included in the meta-analysis because the effect sizes were not reported. $ These studies were not included in the meta-analysis because the studies reported the homologous association < 2. a and b: The studies data was extracted from one publication by gender, a for boys and b for girls, respectively. c and d: The studies data was extracted from one publication in the age groups.
x, y, and z: Study data were extracted from one publication according to different measurement times -x for morning, y for afternoon and z for evening. In meta-analysis, we only included the association between morning cortisol and weight status.
Frontiers in Nutrition | www.frontiersin.org and adiposity-related outcomes. All were based in developed countries. Three articles were among children aged ≤ 12 years old and four articles extracted cortisol by RIA. Two articles measured BMI/BMI z-score, while others measured WC, PBF, visceral fat, and TDFM ( Table 1). Pooled results showed that serum cortisol concentration was not correlated with WC (pooled-r = −0.01, 95% CI: −0.10, 0.09) from two cross-sectional studies ( Table 4 and Figure 4). Meta-analysis of serum cortisol concentration and other adiposity-related outcomes were not possible due to insufficient statistical data.

Urinary Cortisol Concentration and Adiposity-Related Outcomes Among Children
Two articles of 246 children examined associations between urinary cortisol and adiposity-related outcomes. Both were from developed countries and extracted cortisol by RIA. One study measured TDFM while the other measured BMI/BMI z-score and PBF (Table 1). Of the two articles, one reported that the correlations between urinary cortisol and TDFM was r = 0.14 (−0.24, 0.49) for all children and by sex, for boys: r = 0.09, and for girls: r = 0.25. The other article reported that urinary cortisol to be positively associated with BMI (β = 3.54, 95% CI: 1.12, 5.97) and BMI z-score (β = 0.56, 95% CI: 0.16, 0.96), but not with PBF (β = 2.60, 95% CI: −0.65, 5.85). Further subgroup meta-analysis was not possible because necessary statistics were not available.

Sensitivity Analysis and Assessment of Publication Bias
Respective sensitivity analyses were conducted to examine associations of HCC, salivary cortisol, and serum cortisol concentration with adiposity-related outcomes. Only when the study by Chu et al., 2017 was removed from the meta-analyses of cross-sectional studies did the non-significant correlations between morning salivary cortisol concentration and BMI/BMI z-score become significant (r = 0.35, 95% CI: 0.10, 0.60; Supplementary Table 3). The Egger's tests and funnel plots indicated no publication bias within our evaluated study parameters ( Table 2 and Supplementary Figure 1). FMI, fat mass index; BMI, body mass index; SDS, standard deviation score; ELISA, Enzyme-Linked Immunosorbent Assay; LC-MS/MS, liquid chromatography tandem mass spectrometry; CLIA, chemiluminescence immunoassay. We had searched two cohort studies that reported the associations (β, 95% CI) between hair cortisol concentration and adiposity, one of the studies showed that the associations between hair cortisol concentration and BMI: β (95% CI) = 4.62 (1.41, 7.83) (p < 0.01), the other one's effective size has different meaning. Thus, we were unable to perform a meta-analysis. a These studies did not report the associations between cortisol and adiposity for boys and girls, respectively. Thus, we did not do the subgroups analysis across genders, country context, or age groups. b Among the six studies showed association (r) between hair cortisol concentration and weight status, only one reported the association for boys and girls. We divided the article into two studies and one study only showed the association for girls. c These studies were all from developed countries and the participants were ≤12 years old, thus, we did not perform sub-group meta-analysis across country context and age groups. d In the sub-group meta-analysis, only one study was included and the effect size was the one reported in the original study. e The Egger's tests was used to indicate the existence of publication bias. If p-value < 0, it was indicated that publication bias was existed, otherwise, no publication bias existed. Numbers in bold indicate significance.

DISCUSSION
This is the first systematic review and meta-analysis to synthesize and evaluate the associations between different cortisol measures and adiposity-related outcomes in children. We found that most of our included studies examined the associations of either HCC or salivary cortisol concentration with adiposity-related outcomes, and most studies were from developed countries. However, results from our meta-analysis indicated that only HCC, the cortisol measure that serves as an indicator of longterm stress and cumulative cortisol activity, was positively associated with objectively measured adiposity-related outcomes (i.e., FMI-SDS/FMI z-score, BMI, BMI z-score) in children. Salivary, serum, and urinary cortisol measures were not consistently associated with these adiposity-related outcomes, especially after adjustment for covariates, and/or lacked sufficient data for meta-analyses. For HCC, meta-analysis of results from cross-sectional studies showed it to be robustly and positively associated with objectively measured adiposity-related outcomes in children, including FMI-SDS/FMI z-score and BMI/BMI z-score. The age-(≤12 years vs. >12 years) and country developmental context-stratified (developing countries vs. developed countries) analyses also supported these positive adjusted associations. Our meta-analyses result also revealed HCC to be positively correlated with WC without adjusting for covariates. These observations support the role of chronic stress or chronically elevated levels of cortisol in the development and maintenance of both general and central obesity in children. These findings are consistent with the results of a previous systematic review (21). Cortisol increases fat accumulation via glucocorticoid receptors, which have a particularly high density in visceral adipose tissue (38). Moreover, cortisol can increase food intake, especially of energy dense "comfort foods" (39), which can further contribute to increased obesity risk. The positive pooled effect sizes between HCC and adiposity-related outcomes corroborate the importance of considering chronic stress exposures over more acute stress measures when designing or evaluating childhood obesity interventions as well as in treating obesity (7).
Notably, our meta-analyses revealed the novel importance of HCC measurement method, the choice of which modified adjusted cross-sectional associations between HCC and BMI/BMI z-score in children. Only HCC extracted by LC-MS/MS, not ELISA, was associated with BMI/BMI z-scores. Immunoassays such as ELISA tend to yield higher but less accurate HCC than LC-MS/MS, possibly because ELISA overestimate steroid content given antibody cross-reactivity (40). Rather, LC-MS/MS offers superior specificity and sensitivity with its multi-analyte capabilities, making it the preferred method for HCC analysis in high-quality clinical research (41). Additionally, thanks to the high sensitivity for cortisol in hair provided by mass spectrometers, only small samples of hair are needed to run LC-MS/MS, which is conducive for large epidemiological studies among pediatric populations. Future studies should measure HCC by LC-MS/MS, and more longitudinal work is necessary to examine long-term associations.
Twelve of the 17 studies measuring HCC used hair 3 cm proximal to the scalp. Based on an average hair growth rate of 1 cm per month, such samples can reflect the cumulative cortisol and cortisone secretion of HPA axis in the previous 3 months (42). It follows then that most studies using HCC are, either consciously or not, accounting for chronic stress over the past 3-months in children. Other studies have also suggested that researchers could retrospectively examine cortisol production for a particular preceding time period when stress could have been more salient (43). However, other studies have observed HCC to decrease gradually along the length of hair shaft as distal hair samples may suffer a Age of the children in the three studies were ≤12 years, and they did not report the association between salivary cortisol concentration and BMI/BMI z-score for boys or girls. Thus, we did not do the subgroups analysis across gender and age groups. b In the sub-group meta-analysis, only one study was included, so the data reported in the original study was presented. c The Egger's tests was used to indicate the existence of publication bias. If p-value < 0, it was indicated that publication bias was existed, otherwise, no publication bias existed. Numbers in bold indicate significance. more insults (e.g., repeated water and soap exposure) (44). Future study designs should consider these attributes and explore ways to incorporate HCC measures so as to capture cortisol levels encompassing several months. This will serve to further elucidate associations between chronic stress and childhood obesity. In contrast to the long-term inference enabled by HCC, salivary cortisol concentration is more reflective of HPA reactivity and the stress response facilitated by laboratory settings (30). Seven (30,31,34,35,(45)(46)(47) of the 13 studies (23,30,31,34,35,(45)(46)(47)(48)(49)(50)(51)(52) used AUC i to assess increases in salivary cortisol after administering the Trier Social Stress Test for Children (TSST-C) (53). Though AUC i of salivary cortisol was correlated with BMI prior to adjusting for covariates, the adjusted associations were not significant for cross-sectional or longitudinal studies, for studies that measured salivary cortisol by ELISA or LC-MS/MS or for studies from developing or developed countries.
Rather than AUCi of salivary cortisol, the other six studies (23, 48-52) measured morning salivary cortisol to indicate the cortisol awakening response (54). However, we found neither unadjusted nor adjusted associations between morning salivary cortisol concentration and adiposity-related outcomes to be significant. These findings suggest that both cortisol awakening response and cortisol reactivity to acute stress challenge tasks are not associated with adiposity-related outcomes in children. Correspondingly, recent longitudinal studies found that obesity predicted greater changes in cortisol awakening response and cortisol reactivity to challenge in early to middle childhood, not that cortisol awakening response and cortisol reactivity predicted increased likelihood of obesity over the same time period (31). In our review, only four of the 13 included studies  were longitudinal, precluding similar inferences on the direction of these associations. More longitudinal studies are needed to understand these associations. Given the current evidence base, serum cortisol concentration was not observed to be correlated with WC and BMI in children. For urinary cortisol and adiposity-related outcomes, limited studies and data precluded further meta-analyses. However, we did have two studies examine these associations, both supporting significant positive associations between urinary cortisol and BMI (34). Still, these studies' cross-sectional designs and solitary existence demonstrate the need for more efforts to confirm serum and urinary cortisol associations in childhood obesity.
The present systematic review and meta-analysis expands the knowledge base concerning stress biomarker utility in pediatric adiposity research by providing pooled effect sizes for different cortisol measures against objectively measured adiposity-related outcomes. These findings may help health professionals and policymakers better understand how different cortisol measures reflect underlying stress processes and how stress may contribute to adiposity in children. This review also comprehensively investigated the effects of potential moderators on cortisol-adiposity associations, such as age, sex, cortisol measurement method, and country development context. These latter findings provide insights on how to measure HCC more precisely, and how to better understand obesogenic effects of stress in different socio-demographic and economic contexts. Furthermore, examining the pooled effect sizes separately using unadjusted and adjusted models provides a more comprehensive picture of the cortisol with adiposity.
Nonetheless, some limitations should be considered in the interpretation of our results. First, sex-stratified analyses of adjusted associations between HCC and adiposity-related outcomes were not possible given limited statistics available. Second, the generalizability of our findings is limited as we included only studies published in English, most of our included studies were from developed countries, and we excluded studies focusing on children with mental disorders or chronic diseases. Third, most studies were observational in nature, precluding causal interpretations. Fourth, while our findings provide insights on physiological stress processes and adiposity-related outcomes, the sources of stress could not be identified beyond chronicity and acuteness and are thus unable to inform actionable recommendations for childhood obesity prevention efforts; such can be the efforts of future work. Fifth, the number of studies included in some subgroup analyses were small as only limited eligible studies were available, especially for salivary and serum cortisol; more studies utilizing these biomarkers are needed. Last, as several original studies with <50 participants were included in the meta-analysis, the small samples reduced the power to find significant associations between cortisol and adiposityrelated outcomes.
After consideration of the four cortisol measures of hair, saliva, serum, and urine in children, this study provides important evidence supporting a positive relationship between HCC and objectively measured adiposity-related outcomes. Similar findings were found for children aged ≤12 years and >12 years, and for children from developing and developed countries. These findings provide direct evidence of the physiological stress processes that contribute to increased risk of adiposity-related outcomes in children, and corroborate the need to focus on chronic stress in childhood obesity intervention efforts.

DATA AVAILABILITY STATEMENT
The original contributions presented in this study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding authors.