Association between fat mass index and depressive symptoms in U.S. adults: A cross-sectional analysis of NHANES 2011–2016 Authors: Haixia Wu, Shangjin Yang, Cheng Chen, Xin Wang, Wenlin Chen, Fei Ge Published: 2026 ABSTRACT Fat mass index (FMI) reflects adiposity normalized for height and may provide information beyond body mass index. Evidence on its relationship with depressive symptoms in U.S. adults remains limited. This cross-sectional study used data from adults participating in National Health and Nutrition Examination Survey 2011 to 2016. Fat mass index was calculated from dual-energy X-ray absorptiometry-derived fat mass divided by height squared. Depressive symptoms were defined using the Patient Health Questionnaire-9. Survey-weighted logistic regression, restricted cubic spline analysis, subgroup analyses, and machine-learning models were applied to evaluate the association and predictive performance. The final analysis included 7273 participants. After full adjustment, FMI was positively associated with depressive symptoms, with an odds ratio (OR) of 1.05 for each 1-unit increment (95% confidence interval [CI]: 1.03–1.08). In the quartile-based analysis, participants in Q4 showed greater odds of depressive symptoms compared with those in Q1 (OR = 1.50, 95% CI: 1.12–2.01). Restricted cubic spline analysis suggested an approximately linear positive association. In exploratory prediction models, FMI showed a higher AUC than body mass index, although the overall discrimination remained moderate. Higher FMI was associated with a higher prevalence of depressive symptoms among U.S. adults. FMI may serve as a useful adiposity-related indicator in epidemiological research, but prospective studies are needed before clinical predictive use can be recommended. 1. Introduction Major depressive disorder is a common mental health condition characterized by persistent low mood, anhedonia, and reduced motivation, with consequences for work or school performance, eating patterns, and sleep.[1] Globally, depression affects a substantial proportion of adults over the life course[2] and may follow a recurrent or chronic course that undermines social functioning and quality of life. Data from the Global Burden of Disease study showed an increase in depression cases, from 172 million in 1990 to 258 million in 2017,[3] and depression is projected to become a leading contributor to global disease burden by 2030. Recurrence also remains frequent despite therapeutic advances.[4] Obesity is another widespread condition with multiple contributing factors and meaningful health consequences.[5] Excess adiposity has been related to poorer quality of life and to disorders such as obstructive sleep apnea, type 2 diabetes, cardiovascular disease, osteoarthritis, and selected cancers. Although body mass index (BMI) is commonly used to define overweight and obesity according to WHO thresholds,[6] it does not measure body fat directly and may be affected by age, sex, ethnicity, and body composition. Fat mass index (FMI) may offer a more fat-specific index by capturing adipose burden standardized for height, particularly in groups with marked differences in fat and lean mass.[7–10] Evidence increasingly connects adiposity with depressive symptoms, but most population studies have relied on broad anthropometric measures such as BMI or waist circumference. These indices are practical, yet they cannot separate fat mass from lean mass. FMI, calculated by dividing fat mass by squared height, may therefore provide a more targeted measure of adiposity. Large population-based analyses that specifically address FMI in relation to depressive symptoms are still scarce. Using National Health and Nutrition Examination Survey (NHANES) 2011–2016, we examined whether FMI was associated with depressive symptoms in U.S. adults. We further evaluated the shape of this association and compared the exploratory model-based performance of FMI with BMI. 2. Materials and methods 2.1. Study population Data were obtained from NHANES 2011–2016, a nationally representative survey conducted by the National Center for Health Statistics. NHANES collects health-related information through interviews, physical examinations, and laboratory testing. The survey protocol was approved by the NCHS Ethics Review Board, and all participants provided written informed consent. This analysis included adults with complete data on FMI, depressive symptoms, and prespecified covariates. Participants with missing exposure, outcome, or key covariate information were excluded, leaving 7273 participants for analysis (Fig. 1). Because a complete-case approach was used, potential selection bias should be considered when interpreting the findings. Flowchart of the included study population. 2.2. FMI FMI was calculated as fat mass in kilograms divided by height in meters squared: FMI = FM (kg)/height (m2). Body fat mass was obtained from the NHANES DXA examination using a Lunar Prodigy Advance device (GE Healthcare, Madison) operated by trained technologists. Height was measured in the mobile examination center under standardized conditions. 2.3. Assessment of depression symptoms The PHQ-9 questionnaire served as the instrument for evaluating depressive symptoms. This 9-item instrument is based on DSM-IV symptom criteria, produces total scores from 0 to 27, and has been validated for depression screening. Participants with scores of 10 or higher were classified as having clinically relevant depressive symptoms, consistent with the reported 88% sensitivity and specificity at this cutoff.[11] 2.4. Covariates Potential covariates included demographic, socioeconomic, lifestyle, and health-related variables considered relevant to the FMI–depressive symptoms relationship. Demographic and socioeconomic variables included age, sex, race/ethnicity, education level, marital status, and PIR, all collected through standardized household interviews. Smoking status was dichotomized into never smokers and smokers. Individuals who had smoked at least 100 cigarettes were considered smokers. Alcohol use was classified as non-drinking or drinking, with drinking defined as consumption of at least 12 alcoholic beverages during the previous year. Health-related covariates included FMI, hypertension, coronary heart disease, congestive heart failure, chronic bronchitis, cancer or malignancy, and diabetes. FMI was expressed in kg/m2 and calculated as fat mass divided by squared height. Hypertension was defined by self-reported physician diagnosis or measured blood pressure of at least 140/90 mm Hg. Diabetes, coronary heart disease, chronic bronchitis, congestive heart failure, and cancer or malignancy were based on participants’ self-reported history of a diagnosis by a doctor or other health professional. 2.5. Statistical analysis Survey weights, strata, and clustering variables were incorporated in the analyses to account for the complex multistage design of NHANES. Continuous variables are reported as weighted means with standard deviations, and categorical variables as weighted counts and percentages. Weighted group comparisons used the Wilcoxon rank-sum test for continuous variables and Rao-Scott second-order adjusted chi-square tests for categorical variables. Survey-weighted multivariable logistic regression was used to estimate the association between FMI and depressive symptoms, with ORs and 95% CIs. FMI was examined as a continuous variable and as quartiles, using Q1 as the reference category. We fitted 3 sequential models. The first model was crude. The second model accounted for sex, age, race/ethnicity, marital status, education level, and PIR. The third model further incorporated height, coronary heart disease, chronic bronchitis, congestive heart failure, hypertension, cancer or malignancy, smoking, alcohol consumption, and diabetes. Height was added in the fully adjusted model to account for residual body-size variation beyond FMI. Because FMI already includes height standardization, this adjustment should be viewed as an additional analytic strategy rather than part of the FMI definition. For exploratory predictive analyses, we combined variable selection with machine-learning modeling. LASSO regression was used to select predictors and reduce overfitting, with the tuning parameter λ chosen by cross-validation. Discrimination was summarized by ROC curves and AUC values. Random forest and XGBoost models were fitted in addition to regression-based modeling to explore nonlinear patterns and interactions. Feature importance and SHAP values were used to aid interpretation. The predictive framework included 3 components. First, a random forest model was built from LASSO-selected variables, using bootstrap aggregation of decision trees; variable importance was assessed with mean decrease in Gini impurity. Second, an XGBoost model used iterative boosting with early stopping to limit overtraining, and SHAP values were used to describe feature contributions and directionality. Third, all analyses were conducted using R software, version 4.3.2. Statistical significance was defined as a 2-sided P-value < 0.05. 3. Results 3.1. Baseline characteristics of participants Tables 1 and 2 describe the complete-case analytic sample. Table 1 shows weighted characteristics stratified by depressive symptom status. The sample included 7273 participants, and the weighted prevalence of depressive symptoms was 7.8%. Participants with depressive symptoms showed a higher mean FMI than those without such symptoms (11.44 vs 9.67, P < .001). Participants with depressive symptoms were more often female, had at least ninth-grade education, reported lower income (PIR < 3.5), and commonly had a smoking history. Differences by depressive symptom status were also observed for sex, race, education, marital status, hypertension, diabetes, smoking, coronary heart disease, chronic bronchitis, congestive heart failure, cancer or malignancy, PIR, and FMI (all P < .05). Table 1Weighted baseline characteristics of the study participants.CharacteristicN*OverallN = 115,599,504†without DepressionN = 106,579,019†with DepressionN = 9020,485†P-value‡Sex7273<.001 Male3698 (51%)3488 (52%)210 (35%) Female3575 (49%)3193 (48%)382 (65%)Age727339.28 ± (11.71)39.17 ± (11.68)40.61 ± (11.99).033Race7273.005 Mexican1023 (9.6%)967 (9.9%)56 (6.4%) Other Hispanic742 (6.7%)653 (6.5%)89 (9.8%) Non-Hispanic White2760 (64%)2488 (64%)272 (65%) Non-Hispanic Black1525 (11%)1416 (11%)109 (10%) Other Race1223 (8.5%)1157 (8.4%)66 (8.6%)Education7273<.001 Less Than 9th388 (3.5%)344 (3.3%)44 (5.6%) 9–11th853 (9.2%)747 (8.8%)106 (14%) High School1577 (21%)1420 (20%)157 (28%) Some College2395 (34%)2190 (33%)205 (36%) College Graduate2060 (33%)1980 (34%)80 (17%)Marrige7273<.001 Married3524 (52%)3343 (54%)181 (34%) Widowed96 (1.2%)78 (1.0%)18 (3.8%) Divorced660 (9.2%)557 (8.6%)103 (16%) Separated247 (2.5%)200 (2.2%)47 (5.3%) Never married1946 (25%)1776 (24%)170 (28%) Living with partner800 (10%)727 (10%)73 (13%)Drink alcohol7273.594 No1741 (18%)1616 (18%)125 (17%) Yes5532 (82%)5065 (82%)467 (83%)Hypertension7273<.001 No5609 (78%)5233 (79%)376 (65%) Yes1664 (22%)1448 (21%)216 (35%)Diabetes7273<.001 No6718 (94%)6206 (94%)512 (90%) Yes555 (6.1%)475 (5.7%)80 (10%)Smoking7273<.001 No4338 (58%)4124 (60%)214 (35%) Yes2935 (42%)2557 (40%)378 (65%)Height7273169.48 ± (9.45)169.73 ± (9.44)166.52 ± (9.04)<.001Coronary heart disease7273.005 No7210 (99%)6632 (99%)578 (98%) Yes63 (0.8%)49 (0.7%)14 (2.4%)Chronic bronchitis7273<.001 No6946 (95%)6440 (96%)506 (87%) Yes327 (4.8%)241 (4.1%)86 (13%)Congestive heart failure7273.004 No7204 (99%)6629 (99%)575 (98%) Yes69 (0.7%)52 (0.6%)17 (2.2%)Cancer or malignancy7273.045 No6997 (95%)6448 (95%)549 (93%) Yes276 (5.2%)233 (5.0%)43 (7.4%)Income-poverty ration7273<.001 <1.32398 (24%)2078 (22%)320 (42%) 1.3–3.52551 (34%)2360 (34%)191 (36%) >=3.52324 (42%)2243 (44%)81 (21%)FMI72739.81 ± (4.34)9.67 ± (4.24)11.44 ± (5.11)<.001* N, unweighted sample size.† Weighted N, representing the estimated population size in the United States.‡ P-value calculated using appropriate weighted tests (Rao-Scott chi-square for categorical, Wilcoxon for continuous variables). Weighted baseline characteristics of the study participants. * N, unweighted sample size. † Weighted N, representing the estimated population size in the United States. ‡ P-value calculated using appropriate weighted tests (Rao-Scott chi-square for categorical, Wilcoxon for continuous variables). Table 2Weighted baseline characteristics of the study cohort stratified according to FMI quartiles.CharacteristicN*OverallN = 115,599,504†Q1N = 28,907,610†Q2N = 28,906,236†Q3N = 28,914,865†Q4N = 28,870,793†P-value‡Sex7273<.001 Male3698 (51%)1459 (77%)1100 (60%)778 (47%)361 (20%) Female3575 (49%)406 (23%)666 (40%)954 (53%)1549 (80%)Age727339.28 ± (11.71)35.02 ± (11.74)40.04 ± (11.52)41.11 ± (11.25)40.96 ± (11.26)<.001Race7273<.001 Mexican1023 (9.6%)155 (6.6%)227 (9.1%)295 (11%)346 (12%) Other Hispanic742 (6.7%)124 (5.0%)197 (7.3%)192 (7.1%)229 (7.6%) Non-Hispanic White2760 (64%)708 (65%)687 (66%)657 (65%)708 (62%) Non-Hispanic Black1525 (11%)423 (12%)273 (7.8%)340 (9.4%)489 (14%) Other Race1223 (8.5%)455 (12%)382 (9.6%)248 (7.8%)138 (4.7%)Education7273<.001 Less Than 9th388 (3.5%)76 (3.1%)87 (3.0%)115 (4.0%)110 (3.8%) 9–11th853 (9.2%)236 (10%)193 (8.4%)200 (9.0%)224 (9.2%) High School1577 (21%)391 (19%)362 (21%)367 (21%)457 (22%) Some College2395 (34%)570 (32%)509 (28%)585 (35%)731 (40%) College Graduate2060 (33%)592 (36%)615 (40%)465 (31%)388 (25%)Marrige7273<.001 Married3524 (52%)720 (41%)966 (59%)935 (57%)903 (52%) Widowed96 (1.2%)13 (0.8%)17 (0.9%)29 (1.6%)37 (1.5%) Divorced660 (9.2%)124 (6.1%)150 (8.3%)162 (11%)224 (12%) Separated247 (2.5%)55 (2.4%)48 (2.3%)59 (2.1%)85 (3.2%) Never married1946 (25%)720 (38%)411 (21%)359 (18%)456 (21%) Living with partner800 (10%)233 (12%)174 (8.4%)188 (10%)205 (10%)Drink alcohol7273<.001 No1741 (18%)335 (13%)347 (14%)459 (20%)600 (26%) Yes5532 (82%)1530 (87%)1419 (86%)1273 (80%)1310 (74%)Hypertension7273<.001 No5609 (78%)1642 (88%)1422 (81%)1301 (75%)1244 (67%) Yes1664 (22%)223 (12%)344 (19%)431 (25%)666 (33%)Diabetes7273<.001 No6718 (94%)1809 (98%)1671 (96%)1605 (94%)1633 (88%) Yes555 (6.1%)56 (2.4%)95 (4.1%)127 (5.7%)277 (12%)Smoking7273.201 No4338 (58%)1057 (57%)1024 (58%)1046 (57%)1211 (61%) Yes2935 (42%)808 (43%)742 (42%)686 (43%)699 (39%)Height7273169.48 ± (9.45)173.11 ± (8.48)171.06 ± (9.05)168.82 ± (9.40)164.93 ± (8.85)<.001Coronary heart disease7273.182 No7210 (99%)1855 (99%)1749 (99%)1712 (99%)1894 (99%) Yes63 (0.8%)10 (0.5%)17 (0.7%)20 (1.4%)16 (0.7%)Chronic bronchitis7273<.001 No6946 (95%)1813 (97%)1709 (96%)1659 (95%)1765 (92%) Yes327 (4.8%)52 (3.0%)57 (3.9%)73 (4.7%)145 (7.6%)Congestive heart failure7273.148 No7204 (99%)1857 (100%)1749 (99%)1717 (99%)1881 (99%) Yes69 (0.7%)8 (0.4%)17 (0.6%)15 (0.6%)29 (1.2%)Cancer or malignancy7273.081 No6997 (95%)1814 (96%)1708 (95%)1669 (95%)1806 (94%) Yes276 (5.2%)51 (3.7%)58 (5.3%)63 (5.2%)104 (6.5%)Income-poverty ration7273<.001 <1.32398 (24%)617 (25%)503 (20%)554 (21%)724 (27%) 1.3–3.52551 (34%)660 (33%)573 (31%)615 (36%)703 (37%) >=3.52324 (42%)588 (41%)690 (49%)563 (43%)483 (35%)Depression7273<.001 without Depression6681 (92%)1752 (94%)1658 (94%)1602 (93%)1669 (87%) with Depression592 (7.8%)113 (6.1%)108 (5.5%)130 (6.9%)241 (13%)* N, unweighted sample size.† Weighted N, representing the estimated population size in the United States.‡ P-value calculated using appropriate weighted tests (Rao-Scott chi-square for categorical, Wilcoxon for continuous variables). Weighted baseline characteristics of the study cohort stratified according to FMI quartiles. * N, unweighted sample size. † Weighted N, representing the estimated population size in the United States. ‡ P-value calculated using appropriate weighted tests (Rao-Scott chi-square for categorical, Wilcoxon for continuous variables). Table 2 summarizes the weighted baseline characteristics according to FMI quartiles. Participants in Q4 were more likely to be female, Non-Hispanic White, educated beyond the 11th grade, and from lower-income households. Alcohol consumption, hypertension, diabetes, and chronic bronchitis differed across FMI quartiles (P < .001). Compared with Q1, Q4 showed higher proportions of hypertension and diabetes. The prevalence of depressive symptoms increased across higher FMI categories. 3.2. Association between FMI and depressive symptoms Table 3 presents the survey-weighted logistic regression results. After full adjustment, FMI remained positively associated with depressive symptoms when modeled continuously, with an OR of 1.05 for every 1-unit increase (95% CI: 1.03–1.08, P < .001). When participants were grouped by FMI quartiles, those in the highest quartile showed higher odds than those in the lowest quartile (OR = 1.50, 95% CI: 1.12–2.01, P = .006). The quartile-based results also showed a significant upward trend across FMI categories (P for trend = .006). Table 3Association between FMI and depressive symptoms in U.S. adults: NHANES 2011–2016.OutcomesContinuous modelsCategorical modelORP valueQ1Q2Q3Q4P trendModel 11.08 (1.06–1.10)<0.0011 (Ref)0.90 (0.63–1.27)1.13 (0.80–1.60)2.21 (1.70–2.88).001Model 21.06 (1.04–1.08)<0.0011 (Ref)0.91 (0.63–1.31)0.98 (0.70–1.37)1.56 (1.18–2.04).002Model 31.05 (1.03–1.08)<0.0011 (Ref)0.91 (0.62–1.33)0.97 (0.69–1.37)1.50 (1.12–2.01).006 Model 1: Unadjusted. Model 2: Demographic covariate-adjusted (sex, age, ethnicity, marital status, education, PIR). Model 3: Incorporated Model 2 adjustments plus clinical/behavioral covariates (coronary heart disease, chronic bronchitis, congestive heart failure, hypertension, cancer/malignancy, smoking, alcohol use, diabetes). Association between FMI and depressive symptoms in U.S. adults: NHANES 2011–2016. Model 1: Unadjusted. Model 2: Demographic covariate-adjusted (sex, age, ethnicity, marital status, education, PIR). Model 3: Incorporated Model 2 adjustments plus clinical/behavioral covariates (coronary heart disease, chronic bronchitis, congestive heart failure, hypertension, cancer/malignancy, smoking, alcohol use, diabetes). Restricted cubic spline analysis showed a significant overall association between FMI and depressive symptoms (P-overall < .05), while the nonlinear term was not significant (P-nonlinear = .49). These findings indicate an approximately linear positive relationship between FMI and depressive symptoms (Fig. 2). Restricted cubic spline analysis of the FMI-depression relationship. FMI = fat mass index. 3.3. Subgroup analysis Subgroup analyses assessed whether the FMI-depressive symptoms association differed across demographic and clinical strata (Fig. 3). No significant interaction was observed for sex, education level, race, smoking, marital status, drinking, chronic bronchitis, congestive heart failure, hypertension, diabetes, PIR, or cancer or malignancy (all P-interaction > .05). Coronary heart disease showed evidence of interaction (interaction P < .05), suggesting possible effect modification. Within the gender-stratified restricted cubic spline analysis, the curve slope was steeper for females than males, revealing a significantly greater increase in prevalence of depressive symptoms at higher FMI levels among females (Fig. 4). The restricted cubic spline model stratified by race demonstrated notable discrepancies in the FMI and odds of depressive symptoms curves across diverse racial groups (Fig. 5). Similarly, the restricted cubic spline model by education level demonstrated variations in the curvature of the FMI curves across different educational groups (Fig. 6). Subgroup analyses of the FMI-depression association. Red text indicates the subgroup category used as the reference or the category in which the interaction term reached statistical significance. FMI = fat mass index. A restricted cubic spline model analyzed the FMI-depression association, with stratification by gender (1 = Male, 2 = Female). FMI = fat mass index. Race-stratified restricted cubic spline analysis of the association between FMI and depression (1 = Mexican, 2 = Other Hispanic, 3 = Non-Hispanic White, 4 = Non-Hispanic Black, 5 = Other Race). FMI = fat mass index. Education-level-stratified restricted cubic spline analysis of the FMI-depression association (1 = Less Than 9th, 2 = 9–11th, 3 = High School, 4 = Some College, 5 = College Graduate). FMI = fat mass index. 3.4. Model feature selection, construction, and interpretation Figures 7 and 8 display the LASSO regression results, identifying 16 significant predictors, including age, race, sex, congestive heart failure, cancer or malignancy, coronary heart disease, chronic bronchitis, diabetes history, alcohol consumption, education, height, hypertension, PIR, marital status, smoking, and FMI. These features were used in subsequent machine learning models for direct comparison. Ultimately, 10 key features were selected at the lambda.min and lambda.1se positions, which simplified the model and improved generalizability. Based on feature importance ranking, age, FMI, marital status, and sex emerged as the top predictors. The model demonstrated good classification performance, with an ROC AUC of 0.749. LASSO regression model and variable screening. LASSO regularization path for model optimization. Figure 9 summarizes the random forest model based on LASSO-selected variables. Age, FMI, marital status, and sex were the main predictors. With increasing tree numbers, the out-of-bag error declined and then stabilized, supporting model convergence. The random forest model had an AUC of 0.737. Random forest model analysis. Figure 10 shows the XGBoost model results. Smoking, PIR, FMI, chronic bronchitis, and sex were identified as leading contributors. The model yielded an AUC of 0.736, indicating moderate discrimination. SHAP analysis suggested that FMI contributed to model classification, with higher FMI values generally corresponding to a higher model-predicted probability of depressive symptoms. XGBoost model evaluation. To compare FMI with BMI as single adiposity indicators, ROC curves were examined (Fig. 11). FMI showed a higher AUC than BMI (P < .001), although the discrimination of both single indicators was limited. ROC curve comparison between FMI and BMI (Blue: BMI; Red: FMI). BMI = body mass index, FMI = fat mass index. Figure 12 shows the overlap of selected predictors across LASSO, random forest, and XGBoost models. The shared features suggest consistency across algorithms, but these findings should be interpreted as exploratory. Venn diagrams for 3 machine learning models (LASSO, RF, XGBoost). 4. Discussion Using data from U.S. adults, we assessed whether FMI was associated with depressive symptoms. Using NHANES 2011–2016 data, we found a positive association between FMI and depressive symptom prevalence after adjustment for demographic, lifestyle, and comorbidity factors. The quartile analysis showed a similar pattern. Exploratory LASSO, random forest, XGBoost, and SHAP analyses were used to evaluate model-based classification and feature contribution, but these analyses were supplementary rather than diagnostic. Depression and obesity are both prevalent and significant public health concerns globally.[12] Their relationship is complex and multidimensional.[13–15] Evidence confirms odds of depressive symptoms rises proportionally with obesity severity and metabolic dysfunction, where elevated adiposity further compounds this risk.[16] Research confirms reciprocal links exist between overweight (BMI 25–29.99) and depression across genders, with heightened associations observed in obesity (BMI ≥ 30).[17] Although BMI and waist circumference (WC) serve as universal diagnostic criteria for obesity, BMI exhibits limitations in evaluating obesity risk across varied body compositions, age groups, and ethnic populations.[18,19] Consequently, establishing more precise obesity metrics is imperative. The fat mass index (FMI), an innovative obesity measure, directly quantifies fat mass with height standardization to more accurately capture metabolic burden. Compared to BMI and WC, FMI offers 3 key advantages. First, enables objective comparisons across individuals with varying physiques. Second, precisely distinguishes lean mass from fat mass. Third, effectively identifies high-risk individuals with normal BMI but excessive body fat.[20] In pediatric populations, FMI enables more precise obesity classification. Freedman and colleagues revealed a curvilinear association of BMI with adiposity during childhood obesity evaluation, establishing FMI as the preferred metric.[21] Adult studies further confirm FMI’s significant association with metabolic syndrome and its high predictive validity in both males and females.[22,23] In patients with asthma, the combination of increased FMI and decreased fat-free mass index (FFMI) was associated with higher mortality after adjustment for covariates. Overall, these pieces of evidence suggest that FMI is a valuable forward-looking indicator.[24] The link between obesity and depression in adults has received relatively strong support. Previous research has demonstrated obesity increases prevalence of depressive symptoms by 18%[25]; A meta-analysis reports 23 to 36% higher self-reported depressive symptoms and 14 to 34% elevated risk of clinically diagnosed major depressive disorder in obese versus non-obese adults[26]; Findings from Mendelian randomization analyses suggest that obesity may contribute causally to depression.[27] Mounting evidence confirms that obesity’s psychiatric consequences stem primarily from suboptimal diets, physical inactivity, and visceral adipose tissue accumulation. Under high-fat diet conditions, prevalent among obese individuals, gut microbiota dysbiosis precipitates systemic inflammation and hypothalamic-pituitary-adrenal axis perturbation, ultimately propagating neural impairment.[28] According to Ouchi et al, excessive adipose tissue is not only a site of fat storage but also an endocrine-active tissue capable of producing inflammatory mediators, including interleukin-6 (IL-6) and tumor necrosis factor-alpha (TNF-α).[29] Moreover, sustained low-grade inflammation is mechanistically connected to depression pathophysiology through cytokine penetration of the blood-brain barrier, which subsequently disrupts key monoamine systems (serotonin, dopamine, norepinephrine).[30] Increased body fat also induces changes in leptin levels, further affecting mood regulation. Jitte et al confirmed that leptin resistance, often accompanying high body fat, impairs neuroendocrine signaling pathways governing mood.[31] Notably, insulin resistance, prevalent in obese populations, adversely impacts brain insulin signaling, potentially compromising cognitive and affective functions.[32] This study was strengthened by its large nationally representative sample and by the use of survey weights appropriate for NHANES analyses. We also adjusted for multiple demographic, lifestyle, anthropometric, and clinical variables. We further conducted subgroup analyses to evaluate the consistency of the association across strata, and restricted cubic spline analyses to assess whether the relationship departed from linearity. Machine learning models exhibit significant predictive power for functional outcomes among psychosis-risk or early-depression individuals.[33] Our analytical approach employed 3 machine learning algorithms to pinpoint key determinants modulating prevalence of depressive symptoms, yielding 16 critical predictors. In the supplementary predictive analysis, the models based on LASSO, random forest, and XGBoost respectively achieved AUC values of 0.749, 0.737, and 0.736, indicating their relatively moderate discrimination ability. To date, interpretable machine learning has been less frequently applied in prevalence of depressive symptoms assessment. Among these approaches, LASSO had the highest AUC. SHAP analysis suggested that FMI and PIR were leading contributors after smoking status. This pattern was consistent with the regression results, which showed an independent association between FMI and depressive symptoms. These findings support the exploratory value of interpretable machine learning but do not establish clinical utility. Overall, the findings indicate that higher FMI is associated with greater prevalence of depressive symptoms among U.S. adults. This association may be related to low-grade inflammation, leptin and insulin resistance, neuroendocrine dysregulation, and related pathways connecting excess adiposity with mood-related outcomes. Because the data are cross-sectional, directionality and causation cannot be determined. Our complementary predictive analyses should also be interpreted cautiously. The AUC values showed moderate, not strong, discrimination; therefore, the machine-learning findings should be regarded as supplementary and exploratory. Several limitations need to be acknowledged. First, the cross-sectional nature of the study does not allow causal inference or assessment of temporal sequence. Second, depressive symptoms were identified using the PHQ-9 rather than a structured psychiatric interview. Third, unmeasured or residual confounding may remain despite multivariable adjustment. Fourth, complete-case selection led to substantial exclusions and may have affected representativeness. Finally, although FMI performed numerically better than BMI, the AUC values were modest, and FMI should be considered an anthropometric correlate rather than a validated clinical biomarker. 5. Conclusions Among U.S. adults, higher FMI was associated with a greater likelihood of depressive symptoms. FMI may be considered an adiposity-related correlate for use in future epidemiological studies. Prospective studies and external validation are needed before any clinical predictive application is pursued. Acknowledgments The authors thank the National Center for Health Statistics and the NHANES participants for making these data available for public health research. Author contributions Conceptualization: Haixia Wu. Data curation: Haixia Wu. Methodology: Haixia Wu. Supervision: Fei Ge. Visualization: Shangjin Yang. Writing – original draft: Haixia Wu, Shangjin Yang, Cheng Chen, Xin Wang, Wenlin Chen, Fei Ge. Writing – review & editing: Haixia Wu, Shangjin Yang, Cheng Chen, Xin Wang, Wenlin Chen, Fei Ge.