==== Front JAMA Netw Open JAMA Netw Open JAMA Network Open 2574-3805 American Medical Association 36961462 10.1001/jamanetworkopen.2023.4625 zoi230171 Research Original Investigation Online Only Psychiatry Association Between Ghrelin and Body Weight Trajectory in Individuals With Anorexia Nervosa Association Between Ghrelin and Body Weight Trajectory in Anorexia Nervosa Association Between Ghrelin and Body Weight Trajectory in Anorexia Nervosa Kim Youngjung R. MD PhD 1 Lauze Meghan S. BS 2 Slattery Meghan MSN 2 Perlis Roy H. MD MSc 1 Holsen Laura M. PhD 3 4 Breithaupt Lauren PhD 5 Stern Casey M. BA 5 Fava Maurizio MD 6 Thomas Jennifer J. PhD 5 Lawson Elizabeth A. MD 2 Misra Madhusmita MD MPH 7 Eddy Kamryn T. PhD 5 1 Center for Quantitative Health, Department of Psychiatry, Massachusetts General Hospital and Harvard Medical School, Boston 2 Neuroendocrine Unit, Department of Medicine, Massachusetts General Hospital and Harvard Medical School, Boston 3 Division of Women’s Health, Department of Medicine, Brigham and Women’s Hospital, Boston, Massachusetts 4 Department of Psychiatry, Brigham and Women’s Hospital and Harvard Medical School, Boston, Massachusetts 5 Eating Disorders Clinical and Research Program, Department of Psychiatry, Massachusetts General Hospital and Harvard Medical School, Boston 6 Department of Psychiatry, Massachusetts General Hospital and Harvard Medical School, Boston 7 Department of Pediatrics, Division of Pediatric Endocrinology, Massachusetts General Hospital and Harvard Medical School, Boston Article Information Accepted for Publication: February 7, 2023. Published: March 24, 2023. doi:10.1001/jamanetworkopen.2023.4625 Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2023 Kim YR et al. JAMA Network Open. Corresponding Author: Youngjung R. Kim, MD, PhD, Center for Quantitative Health, Department of Psychiatry, Massachusetts General Hospital and Harvard Medical School, 185 Cambridge St CPZN 6300, Boston, MA 02114 (ykim82@mgh.harvard.edu). Author Contributions: Dr Kim had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Drs Lawson, Misra, and Eddy contributed equally to the manuscript. Concept and design: Thomas, Lawson, Misra, Eddy. Acquisition, analysis, or interpretation of data: All authors. Drafting of the manuscript: Kim, Eddy. Critical revision of the manuscript for important intellectual content: All authors. Statistical analysis: Kim. Obtained funding: Lawson, Misra, Eddy. Administrative, technical, or material support: Kim, Lauze, Breithaupt, Stern, Thomas, Lawson. Supervision: Perlis, Fava, Thomas, Lawson, Misra, Eddy. Conflict of Interest Disclosures: Dr Kim has not done any personal consulting; any consulting she has done has been on behalf of Massachusetts General Hospital (MGH). Dr Perlis reported receiving personal fees from Genomind, Belle Artificial Intelligence, Psy Therapeutics, Burrage Capital, and Circular Genomics outside the submitted work. Dr Breithaupt reported receiving personal fees from Otsuka Pharmaceutical outside the submitted work. Dr Fava reported receiving research support from Acadia Pharmaceuticals, Aditum Bio Management Company, Allergan, Alkermes, Altimate Health Corporation, Alto Neuroscience, Ancora Bio, Angelini S.p.A., Aptinyx, Arbor Pharmaceuticals, Avanir Pharmaceuticals, Axsome, Benckiser Pharmaceuticals, BioClinica, Biogen, BioHaven, BioShin Ltd, Cambridge Science Corporation, Centrexion Therapeutics Corporation, Cerecor, Cybin IRL Ltd, Eliem Therapeutics Ltd, Gate Neurosciences, GenOmind, Gentelon, Happify, Johnson & Johnson, Lundbeck, Marinus Pharmaceuticals, Methylation Sciences, Millennium Pharmaceutics, Minerva Neurosciences, Neuralstem, Neurocrine Biosciences, NeuroRX, Novaremed, Novartis, Otsuka, Pfizer, Premiere Research International, Praxis Precision Medicines, Protagenic Therapeutics, Relmada Therapeutics, Reckitt, Shenox Pharmaceuticals, Stanley Medical Research Institute, Taisho, Takeda, University of Michigan, Vistagen, WinSanTor, Xenon Pharmaceuticals, National Institute of Drug Abuse, National Institutes of Health, National Institute of Mental Health, and Patient-Centered Outcomes Research Institute outside the submitted work; lecturing at Global Medical Education Mood Disorders Summit, November 2020; having equity holdings in Compellis, Neuromity, Psy Therapeutics, and Sensorium Therapeutics; holding patents for Sequential Parallel Comparison Design, licensed by MGH to Pharmaceutical Product Development, (US_7840419, US_7647235, US_7983936, US_8145504, US_8145505); and having patent application for a combination of Ketamine Plus Scopolamine in Major Depressive Disorder, licensed by MGH to Biohaven; having patents for pharmacogenomics of Depression Treatment with Folate (US_9546401, US_9540691); and having copyrights for the MGH Cognitive & Physical Functioning Questionnaire, Sexual Functioning Inventory, Antidepressant Treatment Response Questionnaire, Discontinuation-Emergent Signs & Symptoms, Symptoms of Depression Questionnaire, and SAFER from Belvio, Lippincott, Williams & Wilkins, Wolkers Kluwer, and World Scientific Publishing Co. Dr Fava has not done any personal consulting; any consulting he has done has been on behalf of MGH, except for Sensorium Therapeutics. The most up-to-date lifetime disclosures for Dr Fava can be viewed online (https://mghcme.org/faculty). Dr Lawson reported receiving grants from Tonix Pharmaceuticals, serving on a scientific advisory board and receiving stock options from OXT Therapeutics, and receiving personal fees from UpToDate (royalties) outside the submitted work. Dr Misra reported receiving personal fees from AbbVie, Ipsen, and Sanofi outside the submitted work. No other disclosures were reported. Funding/Support: The parent study was supported by grant R01 MH103402 from the National Institute of Mental Health (Drs Misra, Eddy, and Lawson). This work was supported in part by grants K99 MH127366 (Dr Kim), K24 MH120568 (Dr Lawson), R01 MH116270 (Dr Perlis), R01 MH123804 (Dr Perlis), K23 MH127465 (Dr Breithaupt), and R03 MH126143 (Dr Breithaupt) from the National Institute of Mental Health and grants P30 DK040561 (Dr Lawson) and R01 DK104772 (Dr Holsen) from the National Institute of Diabetes and Digestive and Kidney Diseases, the Brain Behavior Research Foundation (Dr Breithaupt), the Phyllis & Jerome Lyle Rappaport Foundation (Dr Kim), and the MGH Department of Psychiatry (Dr Kim). Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. Disclaimer: Dr Perlis is an Associate Editor for JAMA Network Open but was not involved in the editorial review or decision on this manuscript. Data Sharing Statement: See Supplement 2. Additional Contributions: We are grateful to all study participants for their involvement in the study. We thank our colleagues at the MGH Eating Disorders Clinical and Research Program, Division of Pediatric Endocrinology, Division of Adolescent Medicine, and Neuroendocrine Unit, for their support of participant recruitment, sample collection, and experimental work. 24 3 2023 3 2023 24 3 2023 6 3 e23462527 10 2022 7 2 2023 Copyright 2023 Kim YR et al. JAMA Network Open. https://creativecommons.org/licenses/by/4.0/ This is an open access article distributed under the terms of the CC-BY License. jamanetwopen-e234625.pdf Key Points Question Is there an association between circulating levels of the orexigenic hormone ghrelin and body weight trajectories in individuals with anorexia nervosa? Findings In this 18-month cohort study of 68 girls and young women, elevated baseline ghrelin levels were associated with prospective weight gain in anorexia nervosa. Meaning This study offers evidence of association between ghrelin and longitudinal weight outcomes in individuals with anorexia nervosa; further studies are warranted to confirm this association and evaluate the potential clinical utility of ghrelin in anorexia nervosa. This cohort study examines the use of endogenous ghrelin as a baseline factor associated with longitudinal weight trajectories in individuals with anorexia nervosa compared with healthy controls. Importance Individuals with anorexia nervosa maintain extremely low body weights despite elevations in the circulating orexigenic hormone ghrelin. Whether circulating levels of endogenous ghrelin are associated with weight gain in anorexia nervosa is unknown. Objective To examine the association between baseline ghrelin and future weight change in individuals with anorexia nervosa. Design, Setting, and Participants This prospective cohort study was conducted between April 1, 2014, and March 31, 2020, in the US. Girls and women aged 10 to 22 years were recruited from the greater Boston area from community and area treatment centers, enrolled, and followed up for 18 months. Statistical analyses were performed between January and August 2022. Exposures Presence or absence of anorexia nervosa and elevations in endogenous ghrelin. Main Outcomes and Measures Changes in age- and sex-standardized body mass index percentiles from baseline to 9- and 18-month follow-up were the main outcomes of interest. Results A total of 68 girls and young women (11 [16%] Asian, 4 [6%] Hispanic or Latina, 51 [75%] White [non–Hispanic or Latina], and 2 [3%] other race or ethnicity), including 35 with anorexia nervosa and 33 healthy controls of similar Tanner stage, were included in this study. Anorexia nervosa and healthy control groups were not statistically different by race and ethnicity, Tanner stage, number completing follow-up visits, and the duration between baseline and follow-up visits. At baseline, individuals with anorexia nervosa were slightly older (median [IQR], 20.1 [18.5-21.0] vs 18.7 [14.7-19.4] years; P = .005), had lower body mass index percentiles (median [IQR], 2.4 [0.3-4.7] vs 52.9 [40.4-68.3]; P < .001), and had elevated circulating ghrelin area under the curve composite index (median [IQR], 1389.4 [1082.5-1646.4] vs 958.5 [743.0-1234.5] pg/mL; P = .003) compared with healthy individuals. In linear mixed-effects regression analyses, baseline ghrelin was associated with prospective weight gain after adjusting for diagnosis, age, race, and duration of follow-up (odds ratio, 2.35; 95% CI, 1.43-3.73; P = .004). Conclusions and Relevance In this cohort study, endogenous ghrelin was associated with longitudinal weight gain in individuals with anorexia nervosa. Further studies are warranted to confirm this result and examine its potential clinical utility in treatment development. ==== Body pmcIntroduction Anorexia nervosa (AN) is a debilitating chronic illness with a premature mortality rate 6 times that of the general population—one of the highest among psychiatric illnesses.1,2 The mainstay of treatment includes interventions aimed at restoring body weight, which are challenged by illness-driven motivations to remain thin. No medications are currently approved by the US Food and Drug Administration for AN; innovative therapies modulating endocrine signals involved in energy homeostasis, such as ghrelin,3 are being investigated. Ghrelin is a peptide hormone secreted primarily from the stomach that acts on the growth hormone secretagogue receptor 1a in the hypothalamus to promote food intake and weight gain in animal models.4,5,6 Surprisingly, even though intravenous ghrelin administration increases food intake in the short term,7 elevations in endogenous ghrelin have not been associated with prospective weight gain in humans,8,9,10 except in infants born small for gestational age, whose rates of weight gain in the first year of life were positively associated with persistent elevations in ghrelin following glucose administration.11 In AN, intravenous ghrelin administration increases food intake,12 and endogenous ghrelin levels are significantly elevated,13 related to increased secretory bursts.14 Whether such high levels of circulating ghrelin have a role in long-term weight homeostasis is unknown. Understanding this association is critical because there is an urgent need for novel therapeutics to assist with weight restoration in this population. In this study, we tested endogenous ghrelin as a baseline factor associated with longitudinal weight trajectory in individuals with AN compared with healthy controls (HCs). Methods This cohort study used data collected as part of a larger parent study examining multidimensional aspects of food motivation pathways, including hormones, neural circuitry, and psychological symptoms, as mediators of eating disorder trajectories for 18 months in female adolescents and young adults with low-weight eating disorders, including AN. A summary of the parent study protocol, including eligibility criteria and study procedures, appears in the eMethods in Supplement 1. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.15 Data Collection Data collection took place between April 1, 2014, and March 31, 2020, at Massachusetts General Hospital (MGH) with institutional review board approval. All participants provided written informed consent or parental consent with assent of minors younger than 18 years. Participants were recruited from clinical programs at MGH and beyond, including the MGH Eating Disorders Clinical and Research Program, specialized treatment centers in the greater Boston area, student health centers at local universities, and community advertisements. Female participants aged 10 to 22 years were eligible if they met the Diagnostic and Statistical Manual of Mental Disorders (Fifth Edition) (DSM-5)16 criteria for AN with insufficient energy intake relative to requirements due to restrictive eating behaviors characteristic of AN, resulting in a significantly low body weight, or were HCs without a lifetime history of eating disorders with similar Tanner stages. DSM-5 defines low body weight for adults with AN as body mass index (BMI) less than 18.5 (calculated as weight in kilograms divided by height in meters squared) but does not define a cutoff BMI percentile for children and adolescents with AN, and low body weight in this study was defined as no greater than the 10th BMI percentile for age and sex.17,18 Screening for AN included evaluation to confirm the lack of any organic conditions that could account for the low body weight. Study participants had no history of diabetes, gastrointestinal conditions or procedures, recent systemic hormone use, or pregnancy, which could impact ghrelin levels. Participants were medically and psychiatrically stable with sufficiently mitigated imminent risk of harm, without dangerous and active medical issues (eg, severe or rapidly progressing electrolyte imbalances or anemia) or acutely elevated psychiatric risk (eg, active suicidality). See the eMethods in Supplement 1 for details of selection criteria. Data were collected at baseline, 9-month, and 18-month study visits. Each visit started with anthropometric measurements and a fasting blood draw, after which participants were asked to eat an approximately 400-kcal mixed meal standardized for macronutrient content, with approximately 20% fat, 60% carbohydrates, and 20% protein. At 0.5, 1.0, and 2.0 hours after meal initiation, blood samples were collected. Blood was fractionated to obtain plasma, and plasma total ghrelin was measured using an enzyme-linked immunosorbent assay (interassay and intra-assay coefficients of variation, 6.6%, 1.3%; EZGRT-89K, Millipore Sigma). Statistical Analysis To test the association between baseline ghrelin and longitudinal changes in body weight, we filtered the data set for the availability of (1) body weight measurements at baseline and one or both follow-up visits and (2) baseline ghrelin measurements. We tested our a priori hypothesis with a 2-sided significance level of P < .05 using linear mixed models (LMMs) estimating the weight change from baseline to either follow-up visit as a repeated-measures outcome of interest with a random intercept for each individual with ghrelin as the main independent variable of interest. To compare body weight changes in a developmentally diverse cohort, we derived standardized BMI percentiles and z scores using the lambda-mu-sigma (LMS) method to account for the changing skewness of population growth distributions during development, with the participant’s age, sex, measured weight in kilograms, and measured height in meters, together with the US population data.19 For individuals older than 20 years (26 of 68 [38%] at baseline, 30 of 63 [48%] at 9 months, and 40 of 61 [66%] at 18 months), population data were extrapolated to derive LMS-standardized BMI percentiles. The primary outcome of interest, the weight change index, was set a priori as the fold change in BMI percentile from baseline to follow-up (ie, ratio) to account for the significance of weight gain in the at-risk AN population with extremely low BMI percentiles, together with clinical considerations taking into account the challenges in gaining weight for participants with AN vs HCs. As an example, BMI percentile changes from 1 to 5 in participants with AN and from 51 to 55 in HCs both result in an increase of 4, but the fold-change method factors in the clinical significance of weight gain in individuals with low body weights (ie, [5 / 1] × 100 or 500% in AN) vs those with healthy weight (ie, [55 / 51] × 100 or 108% in HCs), respectively. The BMI percentiles (which are always positive) rather than equivalent z scores (may be positive, zero, or negative) were used after LMS standardization to compute fold changes using nonzero and nonnegative values. Dynamic ghrelin measurements across fasting and postprandial time points of the baseline visit were integrated to derive a single composite index (ie, total area under the curve [AUC]) using the trapezoid rule.20 Ghrelin AUC was used as the main independent variable of interest for the univariate LMMs. Age, self-reported race, and diagnostic group at baseline visit and duration of follow-up specific to the visit for the weight change index were used as covariates in the primary multivariable LMM. We included race as a variable in this study (and, later in the sensitivity analysis, ethnicity as well) to account for racial (and ethnic) differences in body weight regulation. Included racial groups were American Indian or Alaska Native, Asian, Black or African American, White, and other. Ethnicity groups included HIspanic or Latina and not Hispanic or Latina. Diagnostic group was included as an interactor given clinical considerations of how chronic starvation in AN may result in associated elevations in ghrelin. With ghrelin AUC as the main independent variable, no covariates had significant effects on the model fit, including the interaction term, in a stepwise backward deletion of variables. Covariate terms were scaled to mitigate the effects of any 1 term having an undue influence on the model. To minimize bias, no outliers were removed in the main LMM, and joint assessments of nonindependent covariates (ie, diagnosis and baseline BMI percentiles) were reserved for sensitivity analyses. Odds ratios (ORs) are easier to interpret than LMM estimates for clinical outcomes such as changes in body weight, which can be dichotomized to derive clinically informative outcome categories (ie, 10% weight gain); however, dichotomization of a continuous outcome can result in loss of information. We derived OR estimates for longitudinal weight gain positively associated with ghrelin AUC directly from LMMs.21 To test the robustness of the primary analysis, we performed a complementary set of sensitivity analyses using LMMs with (1) outlier removal, (2) different sets of covariates with ghrelin AUC, (3) same set of covariates with individual ghrelin measurements, (4) alternatively defined outcome variable for longitudinal change in body weight, (5) alternate assumptions of variance estimators, and (6) alternatively defined random-effects terms. Because ORs were derived for fixed-effect terms only, 3 of the 17 LMMs from sensitivity analyses that tested how different random-effect model terms affect the model fit are not shown in the summary table for clarity of presentation of other sensitivity analyses. Each random-effect term was reduced, and likelihood ratio tests of model reductions were assessed with sequential analysis of variance decomposition of random effects. In LMMs with alternate random effect terms, we observed associations consistent with the main LMM (P < .05 for all 3 LMMs). Moreover, multivariable linear regression models (LMs) for 9- and 18-month visits were tested to assess the sensitivity of repeated-measures approach in LMMs. In addition, we tested the sensitivity of our approach to derive the ORs directly from LMMs by dichotomizing the continuous outcome with a range of cutoff values for percentage of weight gain in a logistic regression approach with univariate generalized linear mixed-effects models (GLMMs) with baseline ghrelin AUC as the positively associated independent variable.21 Lastly, to explore the possibility of large effects of ghrelin AUC diluting effects of associated diagnostic groups in the LMM, we performed subgroup analyses with subsets of data limited to participants with AN or HCs. We limited subgroup analysis to LMMs with 3 terms, with ghrelin AUC and sets of 2 covariate terms from the main model (not including diagnosis). All multivariable LMMs and LMs were assessed with stepwise backward deletion of single terms to assess how dropping each term would affect model fit. Outliers beyond 2 SDs of the mean weight change index were assessed but not removed except when indicated (ie, sensitivity analysis with outlier removal and models using subsets of data as in simple LMs for each follow-up and LMMs for subgroup analyses) to mitigate the potential for amplified outlier effects in models with fewer data points as well as to prevent uneven outlier effects in models with data subsets, which could introduce undue bias (ie, outliers of weight outcome only in the AN group or greater proportion of outliers at 9 months than 18 months). Similar to the a priori hypothesis tested with the primary analysis, post hoc hypothesis testing for sensitivity and subgroup analyses used a 2-sided significance level of P < .05 together with corrections for multiple comparisons using the Benjamini-Hochberg method to mitigate the risk of chance discoveries (type I errors). Even so, sensitivity and subgroup analyses should be interpreted as exploratory given the post hoc nature. Statistical analyses were performed between January and August 2022 with R and RStudio, versions 4.0.0 and 0.99.902 (R Foundation for Statistical Computing).22 Additional details are reported in the eMethods in Supplement 1. Results A total of 198 individuals consented or assented to participate and 76 (38.4%) did not complete the baseline study visit: 47 failed the screen (see reasons in the eMethods in Supplement 1), 11 were lost to follow-up, and 18 withdrew consent or assent before completing the baseline visit. Of the 122 who completed baseline visit procedures, 9 were lost to follow-up and 8 withdrew consent or assent before completing the 9-month study visit, resulting in data from 105 individuals (86.1%) who were followed up longitudinally for at least 9 months. Of those followed up longitudinally, 70 met the criteria for the AN or HC group. Weight or ghrelin data were not available for 2 individuals with AN, resulting in available data from a total of 68 individuals. Baseline Characteristics A total of 68 girls and young women (11 [16%] Asian, 4 [6%] Hispanic or Latina, 51 [75%] White [non–Hispanic or Latina], and 2 [3%] other race or ethnicity), including 35 with AN (median [IQR] age, 20.1 [18.5-21.0] years) and 33 HCs (median [IQR] age, 18.7 [14.7-19.4] years) of similar Tanner stage, were included in this study. The AN group was significantly older than the HC group by 1.6 years (95% CI, 0.5-2.8 years; P = .005). The AN and HC groups were comparable in race and ethnicity, Tanner stages, number completing follow-up, and duration of follow-up. For individuals with AN, we observed lower BMI percentiles at all visits (between-group difference for baseline visit, −51.8; 95% CI, −58.8 to −46.1; P < .001) (Table 1) and higher baseline visit ghrelin AUC (between-group difference in medians, 327.9 pg/mL; 95% CI, 119.6-543.9 pg/mL × 2 hours; P = .003). The AN group gained weight with median BMI percentile fold changes of 1.9 (IQR, 0.6-22.1) at approximately 9 months and 5.8 (IQR, 2.7-36.1) at approximately 18 months, whereas the HC group had relatively stable weights over time with median BMI percentile fold changes of 1.0 (IQR, 0.9-1.1) at approximately 9 months and 1.1 (IQR, 0.9-1.2) at approximately 18 months. See Table 1 for clinical and demographic characteristics of the study participants. Table 1. Study Participant Characteristics Characteristic AN (n = 35) HC (n = 33) Difference (95% CI)a P valueb Age, median (IQR), y 20.1 (18.5 to 21.0) 18.7 (14.7 to 19.4) 1.6 (0.5 to 2.8) .005 Tanner stage, median (IQR) Breast development 5 (5 to 5) 5 (4 to 5) 0 (0 to 0) .17 Genital development 5 (4 to 5) 5 (4 to 5) 0 (0 to 0) .81 Race and ethnicityc Other than White Any 7 (20) 6 (18) NA .79 American Indian or Alaska Native 0 0 NA Asian 6 (17) 5 (15) NA Black or African American 0 0 NA Other 1 (3) 1 (3) NA White 28 (80) 27 (82) NA Hispanic or Latina 3 (9) 1 (3) NA Not Hispanic or Latina 25 (71) 26 (79) NA Completed visitsd Baseline 35 (100) 33 (100) NA .38 9 mo 31 (89) 32 (97) NA Without 9-mo follow-up 4 (11) 1 (3) NA 18 mo 31 (89) 30 (91) NA Without 18-mo follow-up 4 (11) 3 (9) NA Both follow-ups 27 (77) 29 (88) NA Duration of follow-up from baseline, median (IQR), mod 9 mo 9.1 (8.6 to 9.5) 9.1 (8.8 to 9.4) 0.1 (−0.3 to 0.4) .65 18 mo 18.5 (17.8 to 19.4) 18.0 (17.7 to 18.7) 0.5 (−0.1 to 1.2) .14 BMI percentiled Baseline 2.4 (0.3 to 4.7) 52.9 (40.4 to 68.3) −51.8 (−58.8 to −46.1) <.001 9 mo 4.7 (1.8 to 13.9) 52.0 (40.4 to 71.1) −45.0 (−53.1 to −36.7) <.001 18 mo 11.3 (3.5 to 26.5) 54.0 (43.4 to 69.7) −40.7 (−49.1 to −30.3) <.001 Weight change index, fold change in BMI percentile from baseline visit 9 mo 1.9 (0.6 to 22.1) 1.0 (0.9 to 1.1) 0.9 (0.0 to 4.4) .06 18 mo 5.8 (2.7 to 36.1) 1.1 (0.9 to 1.2) 4.6 (2.5 to 13.1) <.001 Baseline visit ghrelin measurements, pg/mL Fastinge 821.5 (668.4 to 1013.4) 583.5 (374.3 to 852.6) 216.6 (75.7 to 352.9) .002 Postprandiale 0.5 h 667.0 (563.3 to 814.2) (n = 33) 467.4 (373.9 to 641.6) (n = 32) 167.0 (40.7 to 274.2) .008 1 h 611.5 (505.1 to 705.8) (n = 30) 475.8 (349.2 to 607.4) (n = 31) 119.9 (19.8 to 224.3) .02 2 h 608.9 (464.5 to 727.1) (n = 35) 413.4 (325.1 to 604.0) (n = 33) 164.3 (65.4 to 263.9) .003 Composite AUC, pg/mL × 2 hf 1389.4 (1082.5 to 1646.4) 958.5 (743.0 to 1234.5) 327.9 (119.6 to 543.9) .003 Abbreviations: AN, anorexia nervosa; AUC, area under the curve; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); HC, healthy control; NA, not applicable. a Groups were compared using the Wilcoxon rank sum test. Group differences in medians computed with the Hodges-Lehmann estimation are shown with nonparametric 95% CIs, computed as the median of the set of differences between each value in the AN subgroup and each value in the HC subgroup. Two-tailed P < .05 was considered significant. b To compare race and ethnicity, nonzero and nonoverlapping categories of Asian, White Hispanic or Latina, White Not Hispanic or Latina, and other were used with the χ2 test. Similarly, to compare number of follow-up completers per visit, nonoverlapping categories of individuals who completed follow-up at month 9 but not at month 18, completed follow-up at month 18 but not at month 9, and completed both follow-up visits were used with the χ2 test. c Race and ethnicity data were self-reported using fixed categories. For non-White individuals, ethnicity breakdown is not shown because all individuals identified as non–Hispanic or Latina. d Numbers of participants with visit interval and BMI percentile data at respective study visits correspond to the numbers of participants completing each visit. e Total ghrelin measures at time points surrounding the test meal at baseline visit. Fasting and 2-hour postprandial ghrelin measurements were available for all participants, and sample sizes of available data for 0.5 and 1 hour after the test meal are indicated. f Composite AUC index for all available ghrelin measurements spanning 2 hours were calculated using the trapezoid method. Associations Between Baseline Ghrelin and Longitudinal Weight Trajectory We applied LMMs to estimate the longitudinal weight gain associated with ghrelin AUC. Univariate LMM indicated a statistically significant association (OR, 1.81; 95% CI, 1.29-2.49; P = .005). Adjusting for covariation with diagnosis, age, race, and follow-up duration in the main multivariable LMM, the OR of the association was 2.35 (95% CI, 1.43-3.73; P = .004). We observed no statistically significant associations for the covariate terms in the main LMM. Results of the main LMM are in Figure 1; raw data not adjusted for covariation appear in eFigure 1 in Supplement 1. Figure 1. Association Between Baseline Ghrelin and Prospective Change in Body Weight Ghrelin area under the curve (AUC) is plotted against the weight change index, defined as body mass index percentile (surrogate of body weight) at follow-up relative to baseline. Weight change index values shown are model adjusted for age, race, and interval duration between baseline and follow-up. Data from 68 individuals (124 data points) are shown. AN indicates anorexia nervosa; HC, healthy control. We tested the validity and robustness of the multivariable LMM assessing our a priori hypothesis through an exploratory series of complementary sensitivity analyses. We found statistically significant positive associations between baseline ghrelin and longitudinal weight gain in all 27 regression models (Table 2). Validity of deriving ORs from LMMs was assessed with GLMMs with dichotomization of the outcome with 21 different weight cutoffs (eFigure 2 in Supplement 1). Table 2. Sensitivity Analyses for the Final Modela Regression model OR (95% CI) P value Reference Main model Univariate 1.81 (1.29-2.49) .005 Multivariableb 2.35 (1.43-3.73) .004 Outlier removal in the data set Outliers − SD methodc 4.49 (2.61-7.34) <.001 Ghrelin measurements instead of ghrelin AUC Fasting 2.17 (1.31-3.50) .01 0.5 h 2.81 (1.70-4.48) <.001 1 h 4.27 (2.47-7.00) <.001 2 h 2.17 (1.32-3.47) .01 Alternatively defined weight change indexes as outcome of interest BMI z Score delta change 2.63 (1.60-4.20) <.001 Percentile delta fold change 2.35 (1.43-3.73) .004 Percentile log fold change 2.96 (1.79-4.74) <.001 Alternative sets of covariates Tanner stage By breast development instead of age 2.43 (1.49-3.85) .002 By genital development instead of age 2.42 (1.49-3.82) .002 Ethnicity Instead of race 2.50 (1.53-3.97) .002 With race as nonoverlapping categories 2.37 (1.45-3.77) .004 BMI percentiles at baseline instead of diagnostic group 2.31 (1.44-3.61) .003 Alternative assumptions of variance estimators Restricted maximum likelihood instead of maximum likelihood 2.35 (1.43-3.73) .006 Linear models using data from specified follow-up visit onlyd Ghrelin at 9 mo Fasting 5.28 (2.31-10.67) <.001 0.5 h 10.77 (4.35-22.21) <.001 1 h 8.42 (3.43-17.37) <.001 2 h 11.52 (4.65-23.80) <.001 AUC 8.32 (3.51-16.84) <.001 Ghrelin at 18 mo Fasting 2.42 (1.14-4.81) .02 0.5 h 3.29 (1.55-6.36) .001 1 h 2.60 (1.22-5.11) .01 2 h 3.83 (1.74-7.62) <.001 AUC 3.02 (1.43-5.90) .003 Abbreviations: AUC, area under the curve; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); OR, odds ratio. a P values are from the Wald χ2 test for the fixed effects of respective ghrelin term in the linear mixed-effects regression model (LMM) and are not adjusted for multiple comparisons. Multiple comparisons adjustment of the false discovery rate did not change results. For all LMMs in sensitivity analyses, except when testing for effects of removing the outliers, no outliers were removed to avoid introducing bias. b All parameters were kept the same as the reference main multivariable model of the study with single change as described for each sensitivity analysis. c From the data set with 124 data points, 6 outlier weight change indexes (<5%) from individuals with anorexia nervosa (AN) were removed before modeling. For 1 individual, data from both visits were identified as outliers; for 3 individuals, data from only 1 follow-up visit were identified as outliers; and for 1 individual who only had single follow-up data, this was identified as an outlier, resulting in the final sample size of 33 for the AN subgroup. No healthy control participants had weight change values in the extremes (n = 33). d Linear regression models (LMs) were used to confirm the findings from the LMMs using weight change index at a single follow-up visit only, adjusted for diagnostic group, age, race, and follow-up duration. The LMs were built with a data set after removal of the 6 outlier data points before performing the analyses. There were 4 and 2 outlier data points in the 9- and 18-month follow-up data with the final data set from 59 and 61 individuals, respectively. Lastly, we conducted subgroup analyses to test the association between diagnostic group and weight trajectory that may have been overshadowed by ghrelin AUC in the main model. We found statistically significant positive associations between ghrelin and longitudinal weight gain only in the AN subgroup and not for HCs, which was a consistent finding across 60 LMMs tested for subgroups (12 LMMs with ghrelin AUC in Figure 2; 48 other LMMs in eFigure 3 in Supplement 1), with no changes after correction for multiple comparisons (false discovery rate P < .025). Figure 2. Subgroup Analyses of Baseline Ghrelin and Prospective Change in Body Weight In models stratified by subgroups, ghrelin area under the curve (AUC) was positively associated with future weight gain only in the anorexia nervosa (AN) subgroup, unadjusted and after adjusting for any 2 combinations of the covariates from the final model. Outliers were removed for subgroup analyses resulting in data points from 33 participants with AN and 33 healthy controls (HCs). Subgroup models could not be adjusted for the same set of covariates in the main 5-term model given the sample sizes. Given the exploratory nature of subgroup analysis, analysis expanded the hypothesis testing to include linear mixed models (LMMs) that replace ghrelin AUC with individual ghrelin measurements, which are shown in eFigure 3 in Supplement 1. Findings from the 60 LMMs were consistent with the models with ghrelin AUC term as shown here. BMI indicates body mass index. Discussion This prospective cohort study found that baseline ghrelin was positively associated with longitudinal weight gain in girls and young women with AN in community settings during approximately 18 months of follow-up. Of note, robust associations in our extensive series of sensitivity analyses support the validity of results regardless of modeling with ghrelin measurements from fasting or postprandial time points or the composite index and regardless of how longitudinal change in body weight is defined. Although the primary analysis did not find statistically significant results of the diagnostic grouping jointly with ghrelin, exploratory subgroup analyses support the specificity of association only in AN. Our finding that ghrelin is associated with longitudinal weight gain in AN is in stark contrast to prior longitudinal investigations in other human populations, and this is the first, to our knowledge, to report this association in those older than infants. Prior studies8,9,10 have consistently demonstrated the lack of association between ghrelin and weight trajectory despite the short-term effects of ghrelin in stimulating food intake in humans7 and preclinical evidence for a weight regulatory role of ghrelin. The only prior human study11 that observed a positive association between ghrelin and prospective weight gain was in infants born small for gestational age, with the association observed only with ghrelin measurements following glucose administration but not for fasting ghrelin. Although we observed robust associations across fasting and postprandial time points, we note that AN and small for gestational age cohorts have low body weights relative to expected for developmental stage based on population norms. It is possible that having very high levels of circulating ghrelin in AN (an observation in agreement with prior studies13) together with vulnerability from extremes of low body weights may exert an effect on energy homeostasis and modulate weight-regulatory mechanisms in this vulnerable population. Large-scale genomic investigations have identified heritable risk factors for AN, which implicate genetic loci associated with metabolic regulation.23 Ghrelin is known to have multifaceted roles in regulating glucose metabolism and is being investigated as a potential new avenue of therapeutics development for metabolic diseases such as type 2 diabetes.24 In addition, genes encoding components of the ghrelin signaling pathway have been studied as risk factors for AN, with 1 study25 reporting that specific single-nucleotide polymorphisms of the ghrelin gene were associated with longitudinal weight recovery in AN. As such, effects of ghrelin signaling may be associated with the course of illness in AN, but this should be interpreted with caution because evidence regarding effects of ghrelin signaling on longitudinal weight gain remains limited. For instance, a prior randomized clinical trial3 evaluated the effects of a ghrelin receptor agonist, relamorelin, on body weight outcome for 22 women with AN (mean age, 28.9 years). Ghrelin agonist treatment resulted in greater weight increases after 4 weeks compared with placebo, with means (SEMs) of 0.86 (0.40) and 0.04 (0.28) kg, but between-group difference in means was not statistically significant (P = .07). This finding is in line with our findings regarding the positive association between endogenous levels of ghrelin and longitudinal weight gain. Even so, it remains unknown whether ghrelin signaling has a direct effect on weight gain in AN, and additional studies to understand ghrelin biology specific to AN are warranted given the urgent need for new treatments because of the lack of US Food and Drug Administration–approved medications for this illness, which has one of the highest mortality rates in psychiatry. Limitations This study has some limitations. Although the cohort age range spans the full developmental spectrum when AN typically presents, the generalizability of our results is limited by the relatively modest sample size and lack of racial, ethnic, and gender diversity. Larger longitudinal studies are needed to confirm the association of elevated endogenous ghrelin with weight trajectory in AN. We also acknowledge the limitations in deriving ORs from LMMs, because the method was developed without accounting for the variance of random effects, which could result in an overestimation in applications to LMMs despite the benefits over GLMMs. This limitation is mitigated by rigorous hypothesis testing directly with LMMs together with extensive and complementary sensitivity analyses, including a series with GLMMs that support the validity and robustness of our findings. Nevertheless, the use of ORs was intended to enhance the interpretability of the results for an outcome variable critical to AN in clinical settings. Conclusions In this cohort study, circulating levels of ghrelin in adolescent girls and young women with AN were positively associated with prospective weight gain during 18 months of follow-up. This study provides evidence supporting further research into the role of ghrelin biology in weight regulation in AN. Supplement 1. eMethods. Supplemental Methods eFigure 1. Association Between Baseline Ghrelin and Prospective Change in Body Weight Without Adjustment for Covariates eFigure 2. Sensitivity Analysis for Deriving Odds Ratios From Linear Regression Models eFigure 3. Expanded Subgroup Analyses of Baseline Ghrelin and Prospective Change in Body Weight Click here for additional data file. Supplement 2. Data Sharing Statement Click here for additional data file. ==== Refs References 1 The Social and Economic Cost of Eating Disorders in the United States of America: A Report for the Strategic Training Initiative for the Prevention of Eating Disorders and the Academy for Eating Disorders. June 2020. Accessed September 20, 2022. https://www.hsph.harvard.edu/striped/report-economic-costs-of-eating-disorders/ 2 Eddy KT, Tabri N, Thomas JJ, . Recovery from anorexia nervosa and bulimia nervosa at 22-year follow-up. J Clin Psychiatry. 2017;78 (2 ):184-189. doi:10.4088/JCP.15m10393 28002660 3 Fazeli PK, Lawson EA, Faje AT, . Treatment with a Ghrelin agonist in outpatient women with anorexia nervosa: a randomized clinical trial. J Clin Psychiatry. 2018;79 (1 ):17m11585. doi:10.4088/JCP.17m11585 29325236 4 Kojima M, Hosoda H, Date Y, Nakazato M, Matsuo H, Kangawa K. Ghrelin is a growth-hormone-releasing acylated peptide from stomach. Nature. 1999;402 (6762 ):656-660. doi:10.1038/45230 10604470 5 Tschöp M, Smiley DL, Heiman ML. Ghrelin induces adiposity in rodents. Nature. 2000;407 (6806 ):908-913. doi:10.1038/35038090 11057670 6 Nakazato M, Murakami N, Date Y, . A role for ghrelin in the central regulation of feeding. Nature. 2001;409 (6817 ):194-198. doi:10.1038/35051587 11196643 7 Wren AM, Seal LJ, Cohen MA, . Ghrelin enhances appetite and increases food intake in humans. J Clin Endocrinol Metab. 2001;86 (12 ):5992. doi:10.1210/jcem.86.12.8111 11739476 8 Ravussin E, Tschöp M, Morales S, Bouchard C, Heiman ML. Plasma ghrelin concentration and energy balance: overfeeding and negative energy balance studies in twins. J Clin Endocrinol Metab. 2001;86 (9 ):4547-4551. doi:10.1210/jcem.86.9.8003 11549706 9 Langenberg C, Bergstrom J, Laughlin GA, Barrett-Connor E. Ghrelin, adiponectin, and leptin do not predict long-term changes in weight and body mass index in older adults: longitudinal analysis of the Rancho Bernardo cohort. Am J Epidemiol. 2005;162 (12 ):1189-1197. doi:10.1093/aje/kwi338 16236994 10 Chao AM, Jastreboff AM, White MA, Grilo CM, Sinha R. Stress, cortisol, and other appetite-related hormones: prospective prediction of 6-month changes in food cravings and weight. Obesity (Silver Spring). 2017;25 (4 ):713-720. doi:10.1002/oby.21790 28349668 11 Iñiguez G, Ong K, Peña V, Avila A, Dunger D, Mericq V. Fasting and post-glucose ghrelin levels in SGA infants: relationships with size and weight gain at one year of age. J Clin Endocrinol Metab. 2002;87 (12 ):5830-5833. doi:10.1210/jc.2002-021206 12466394 12 Hotta M, Ohwada R, Akamizu T, Shibasaki T, Takano K, Kangawa K. Ghrelin increases hunger and food intake in patients with restricting-type anorexia nervosa: a pilot study. Endocr J. 2009;56 (9 ):1119-1128. doi:10.1507/endocrj.K09E-168 19755753 13 Seidel M, Markmann Jensen S, Healy D, . A systematic review and meta-analysis finds increased blood levels of all forms of Ghrelin in both restricting and binge-eating/purging subtypes of anorexia nervosa. Nutrients. 2021;13 (2 ):709. doi:10.3390/nu13020709 33672297 14 Misra M, Miller KK, Kuo K, . Secretory dynamics of ghrelin in adolescent girls with anorexia nervosa and healthy adolescents. Am J Physiol Endocrinol Metab. 2005;289 (2 ):E347-E356. doi:10.1152/ajpendo.00615.2004 15755766 15 von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP; STROBE Initiative. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet. 2007;370 (9596 ):1453-1457. doi:10.1016/S0140-6736(07)61602-X 18064739 16 American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders. 5th ed. American Psychiatric Association; 2013. 17 Walsh BT, Hagan KE, Lockwood C. A systematic review comparing atypical anorexia nervosa and anorexia nervosa. Int J Eat Disord. Published online December 12, 2022. doi:10.1002/eat.23856 36508318 18 Andersen SB, Lindgreen P, Rokkedal K, Clausen L. Grasping the weight cut-off for anorexia nervosa in children and adolescents. Int J Eat Disord. 2018;51 (12 ):1346-1351. doi:10.1002/eat.22977 30548931 19 Kuczmarski RJ, Ogden CL, Guo SS, . 2000 CDC Growth Charts for the United States: methods and development. Vital Health Stat 11. 2002;(246 ):1-190.12043359 20 Brouns F, Bjorck I, Frayn KN, . Glycaemic index methodology. Nutr Res Rev. 2005;18 (1 ):145-171. doi:10.1079/NRR2005100 19079901 21 Moser BK, Coombs LP. Odds ratios for a continuous outcome variable without dichotomizing. Stat Med. 2004;23 (12 ):1843-1860. doi:10.1002/sim.1776 15195319 22 Chambers JM. Software for Data Analysis: Programming With R. Springer; 2008. doi:10.1007/978-0-387-75936-4 23 Watson HJ, Yilmaz Z, Thornton LM, ; Anorexia Nervosa Genetics Initiative; Eating Disorders Working Group of the Psychiatric Genomics Consortium. Genome-wide association study identifies eight risk loci and implicates metabo-psychiatric origins for anorexia nervosa. Nat Genet. 2019;51 (8 ):1207-1214. doi:10.1038/s41588-019-0439-2 31308545 24 Poher AL, Tschöp MH, Müller TD. Ghrelin regulation of glucose metabolism. Peptides. 2018;100 :236-242. doi:10.1016/j.peptides.2017.12.015 29412824 25 Ando T, Komaki G, Nishimura H, ; Japanese Genetic Research Group for Eating Disorders. A ghrelin gene variant may predict crossover rate from restricting-type anorexia nervosa to other phenotypes of eating disorders: a retrospective survival analysis. Psychiatr Genet. 2010;20 (4 ):153-159. doi:10.1097/YPG.0b013e32833a1f0e 20421852