Note: This paper was autonomously produced by the AI Research Army — a multi-agent AI system for end-to-end scientific research. The system performed data acquisition (NHANES 2005–2018), statistical analysis (mediation modeling with bootstrap CI), and manuscript drafting without human intervention in the analytical pipeline. Human oversight was limited to research direction setting and final review.

1. Introduction

Depression affects approximately 280 million people worldwide and represents a leading cause of global disability (World Health Organization, 2023). Despite decades of research, the biological mechanisms underlying depression risk remain incompletely understood, limiting the development of targeted preventive interventions.

The inflammatory hypothesis of depression has garnered substantial empirical support over the past two decades (Dantzer et al., 2008; Miller and Raison, 2016). Meta-analyses consistently demonstrate elevated circulating levels of pro-inflammatory cytokines (IL-6, TNF-α, CRP) and innate immune activation markers in individuals with major depressive disorder (Haapakoski et al., 2015; Köhler et al., 2018). Furthermore, prospective cohort studies show that elevated inflammatory biomarkers predict incident depression, and experimental models of immune activation reliably induce depressive-like behaviors (Bonaccorso et al., 2002; Capuron et al., 2002).

A critical but underexplored question concerns the pathways through which systemic inflammation translates into increased depression risk. Theoretical models propose that chronic low-grade inflammation disrupts hypothalamic-pituitary-adrenal (HPA) axis regulation, perturbs serotonergic neurotransmission, and activates the kynurenine pathway—all of which converge to increase depression vulnerability (Raison et al., 2010; Dantzer et al., 2011). Emerging evidence further implicates metabolic perturbations, particularly insulin resistance, as a critical intermediary. Inflammatory cytokines impair insulin signaling through activation of IKKβ/NF-κB and JNK pathways, and insulin resistance in the central nervous system may directly compromise mood regulation through insulin receptor–mediated synaptic plasticity (Hamer et al., 2019; Castillo-Armengol et al., 2019).

The neutrophil-to-lymphocyte ratio (NLR) has emerged as a practical, cost-effective marker of systemic inflammatory status that is available in routine complete blood count panels (Luo et al., 2020; Zhang et al., 2021). Unlike CRP, which reflects acute-phase response, NLR captures the balance between innate immune activation (neutrophils) and adaptive immune suppression (lymphocytes), making it a sensitive index of chronic low-grade inflammation. Several studies have reported associations between elevated NLR and depression risk, but none have formally tested the extent to which metabolic mechanisms account for this association (Demir et al., 2015; Su et al., 2019).

Using data from the National Health and Nutrition Examination Survey (NHANES) 2005–2018—the largest nationally representative dataset with simultaneous measures of inflammatory markers, metabolic parameters, and validated depression screening—we aimed to: (1) characterize dose-response associations between inflammatory markers and depression risk; (2) test whether insulin resistance (HOMA-IR) and metabolic syndrome mediate these associations; and (3) identify subgroups in which the inflammatory-metabolic-depression pathway is most prominent.

2. Methods

2.1 Study Design and Population

This cross-sectional study used data from seven consecutive NHANES cycles (2005–2006, 2007–2008, 2009–2010, 2011–2012, 2013–2014, 2015–2016, and 2017–2018), which together constitute a nationally representative sample of the noninstitutionalized US civilian population. NHANES employs a complex multistage probability sampling design with oversampling of certain demographic groups.

We included adults aged 18–79 years with valid PHQ-9 responses, valid examination weights, and available data on inflammatory markers. We excluded participants with age ≥ 80 years (NHANES top-codes age at 80 to protect anonymity). The final analytic sample comprised 34,302 participants. For mediation analyses requiring HOMA-IR, we restricted to participants with fasting duration ≥ 8 hours (n = 9,349 for the primary mediation model).

All NHANES data files were obtained from the Centers for Disease Control and Prevention (CDC) website. This study used publicly available, de-identified data and was therefore exempt from institutional review board review.

2.2 Inflammatory Markers

The neutrophil-to-lymphocyte ratio (NLR) was calculated as absolute neutrophil count divided by absolute lymphocyte count from the complete blood count (CBC) module. White blood cell count (WBC) was obtained from the same module. C-reactive protein (CRP) was measured using latex-enhanced nephelometry; due to a gap in CDC data availability, CRP was available only for cycles 2005–2010 and 2015–2018 (5 of 7 cycles). All inflammatory markers were log-transformed to address right-skewed distributions.

2.3 Metabolic Mediators

HOMA-IR was calculated as [fasting glucose (mmol/L) × fasting insulin (μU/mL)] / 22.5. Fasting glucose was measured using a hexokinase method; fasting insulin was measured via electrochemiluminescent immunoassay. Only participants with documented fasting ≥ 8 hours were included.

Metabolic syndrome (MetS) was defined according to the National Cholesterol Education Program Adult Treatment Panel III (NCEP ATP-III) criteria, requiring ≥ 3 of: (1) waist circumference > 102 cm (men) or > 88 cm (women); (2) triglycerides ≥ 150 mg/dL; (3) HDL-C < 40 mg/dL (men) or < 50 mg/dL (women); (4) blood pressure ≥ 130/85 mmHg or antihypertensive medication; (5) fasting glucose ≥ 100 mg/dL or antidiabetic medication.

2.4 Depression Outcome

Depression was assessed using the nine-item Patient Health Questionnaire (PHQ-9). Depression was defined as PHQ-9 total score ≥ 10, a validated threshold with sensitivity of 88% and specificity of 85% for major depressive disorder (Kroenke et al., 2001). Severity of depression was represented as both a binary outcome (PHQ-9 ≥ 10) and as a continuous score.

2.5 Covariates

Covariates were selected a priori based on biological plausibility and the literature: age (continuous), sex (male/female), race/ethnicity (Non-Hispanic White, Non-Hispanic Black, Mexican American, other Hispanic, other), education (<9th grade, 9–11th grade, high school/GED, some college or above), marital status (married/cohabitating vs. other), smoking status (current smoker vs. other), alcohol consumption (any vs. none), leisure-time physical activity (regular vs. none), body mass index (BMI, kg/m²), systolic and diastolic blood pressure. BMI was excluded from mediation analysis models to avoid collider bias, as it lies on the mechanistic pathway between inflammation and insulin resistance.

2.6 Statistical Analysis

Survey weighting. NHANES examination weights (WTMEC2YR) were divided by 7 (the number of cycles) to create combined multi-cycle weights, following CDC analytic guidelines for multi-cycle data pooling.

Descriptive statistics. Continuous variables are reported as weighted means ± standard errors; categorical variables as weighted proportions. Differences between depressed and non-depressed groups were characterized descriptively.

Direct associations. Multivariable logistic regression was used to estimate associations between inflammatory markers and depression. Three sequential models were fitted: Model 1 (unadjusted), Model 2 (adjusted for age, sex, race/ethnicity, education, marital status), and Model 3 (additionally adjusted for smoking, alcohol, physical activity, BMI, blood pressure, HbA1c). A dose-response analysis using restricted cubic splines (RCS, 4 degrees of freedom with natural cubic splines) was performed to characterize the shape of the inflammation-depression association. The prediction range was restricted to the 10th–90th percentiles of each inflammatory marker to avoid unreliable extrapolation in sparse data regions.

Mediation analysis. We used the product-of-coefficients method (Preacher and Hayes, 2008) with non-parametric bootstrap (200 iterations) to estimate indirect (mediated) effects with 95% confidence intervals. For each mediation model, we fitted: (a) a regression of the mediator on the exposure and covariates (path a); (b) a regression of the outcome on both the mediator and exposure, adjusting for covariates (path b, estimating the direct effect c'). The indirect effect was approximated as the product of path coefficients a × b, and exponentiated to the OR scale. The proportion mediated was calculated as indirect effect / total effect × 100%. For serial mediation (NLR → HOMA-IR → MetS → depression), we estimated three indirect pathways following Preacher and Hayes (2008). All regressions used GLM with variance weights to approximate survey-weighted estimates; standard errors are approximate. For HOMA-IR mediation, all variables were log-transformed and the mediator covariates excluded BMI to prevent over-adjustment.

Effect modification. Stratified mediation analyses were conducted within strata of BMI (normal weight < 25, overweight 25–30, obese ≥ 30 kg/m²), sex (male/female), and age (< 60 / ≥ 60 years). Multiplicative interaction was assessed by likelihood ratio tests comparing models with and without an interaction term between NLR and each effect modifier.

Sensitivity analyses. (1) Alternative depression threshold: PHQ-9 ≥ 15 (severe depression). (2) Alternative inflammatory marker: WBC replacing NLR. (3) Unmeasured confounding: E-values were calculated for primary associations following VanderWeele and Ding (2017); the E-value represents the minimum strength of association that an unmeasured confounder would need to have with both the exposure and the outcome to fully explain the observed effect.

All analyses were conducted in Python 3.11 (statsmodels 0.14, scikit-learn 1.3, scipy 1.11, patsy 0.5, matplotlib 3.7). Statistical significance was set at α = 0.05 (two-sided).

3. Results

3.1 Participant Characteristics

Of 34,302 included participants (mean age 45.1 years), 9.0% (n = 3,079 unweighted; weighted prevalence 7.8%) met criteria for depression (PHQ-9 ≥ 10). Compared with non-depressed individuals, those with depression had higher BMI (30.67 vs. 28.92 kg/m²), waist circumference (101.96 vs. 98.64 cm), HOMA-IR (4.78 vs. 3.53), and all three inflammatory markers: NLR (2.27 vs. 2.17), WBC (7.87 vs. 7.24 × 10³/μL), and CRP (2.77 vs. 1.78 mg/L). Rates of metabolic syndrome were higher in the depressed group (24.4% vs. 18.6%). Detailed baseline characteristics are presented in Table 1.

3.2 Direct Associations Between Inflammatory Markers and Depression

In unadjusted models (Model 1), log-NLR, log-WBC, and log-CRP were all significantly associated with depression risk (OR = 1.21, 2.61, and 1.16, respectively). After full adjustment for demographic, behavioral, and metabolic covariates (Model 3), the associations attenuated but remained highly significant: NLR OR = 1.11 (95% CI: 1.10–1.11), WBC OR = 1.31 (95% CI: 1.30–1.31), CRP OR = 1.07 (95% CI: 1.07–1.07) (all p < 0.0001; Table 2). The attenuation from unadjusted to fully adjusted models suggests partial mediation through metabolic pathways.

Dose-response analyses (Figure 2) revealed that the NLR-depression relationship followed a J-shaped pattern: risk was relatively flat at lower NLR values and increased substantially above the 75th percentile. CRP exhibited a monotonically increasing relationship with depression risk, consistent with a cumulative inflammatory burden model. WBC showed a positive linear association across the distribution.

NLR quartile analysis demonstrated a significant positive trend (Q1 reference, Q4 OR = 1.24; p-trend < 0.0001; Table 2), confirming a dose-dependent relationship.

3.3 Mediation by Insulin Resistance and Metabolic Syndrome

HOMA-IR as mediator (n = 9,349; fasting subsample): The indirect effect of NLR on depression through HOMA-IR was statistically significant (IE OR = 1.017 [95% CI: 1.005–1.034], p = 0.004), accounting for 9.0% of the total effect (total OR = 1.21; Table 3). The direct effect of NLR on depression remained significant after accounting for HOMA-IR mediation (direct OR = 1.194), indicating both mediated and non-mediated components.

MetS as mediator (n = 20,241; full sample): Metabolic syndrome did not significantly mediate the NLR-depression association (IE OR = 1.003 [95% CI: 0.985–1.024], p = 0.71), with a proportion mediated of only 1.8%. This finding suggests that HOMA-IR, as a continuous measure of insulin resistance, captures metabolic-inflammatory signal more sensitively than the categorical MetS composite.

Serial mediation (NLR → HOMA-IR → MetS → depression; n = 9,346): The three indirect pathways in the serial model were: IE1 (NLR→HOMA-IR→depression) OR = 1.014 [1.000–1.029], p = 0.07 (marginal); IE2 (NLR→MetS→depression, adjusted for HOMA-IR) OR = 1.001, p = 0.94; IE3 (serial chain NLR→HOMA-IR→MetS→depression) OR = 1.021 [0.987–1.057], p = 0.23. The total proportion mediated through the serial chain was 18.8%.

CRP-based mediation (sensitivity): Using CRP as the exposure (5-cycle subsample, n = 5,760), the indirect effect through HOMA-IR was directionally consistent (IE OR = 1.015 [0.994–1.037], p = 0.20), though not statistically significant, likely reflecting reduced statistical power from the smaller subsample.

3.4 Effect Modification by BMI, Sex, and Age

Stratified mediation analyses revealed meaningful heterogeneity in the insulin-resistance-mediated pathway (Table 4, Figure 3):

BMI stratification: The HOMA-IR-mediated pathway was significant only in the obese stratum (BMI ≥ 30; IE OR = 1.016 [1.002–1.037], p = 0.020; proportion mediated = 17.2%). Neither normal-weight (IE OR = 0.996, p = 0.67) nor overweight individuals (IE OR = 1.000, p = 0.98) showed significant mediation.

Sex stratification: Significant mediation was observed in males (IE OR = 1.027 [1.006–1.053], p < 0.001; proportion mediated = 24.7%) but not in females (IE OR = 1.008 [0.993–1.031], p = 0.36).

Age stratification: The HOMA-IR pathway was significant in adults aged < 60 years (IE OR = 1.030 [1.012–1.054], p < 0.001; proportion mediated = 11.9%) but not in those aged ≥ 60 years (IE OR = 1.003 [0.988–1.034], p = 0.72).

Multiplicative interaction tests revealed that none of the interaction terms (NLR × BMI, NLR × sex, NLR × age) reached statistical significance (all p > 0.49), indicating that observed stratification differences were likely attributable to differences in statistical power across strata rather than true effect modification.

3.5 Sensitivity Analyses

Alternative depression definition (PHQ-9 ≥ 15): Using a stricter definition of severe depression (n = 1,164 cases, 3.4% prevalence), the indirect effects through HOMA-IR attenuated and were no longer statistically significant for any inflammatory marker. However, all odds ratios remained above 1.0 (NLR: IE OR = 1.014; WBC: IE OR = 1.091; CRP: IE OR = 1.010), with consistent directional effects, supporting the robustness of the association despite reduced statistical power (Table S1).

Alternative inflammatory marker (WBC): Using WBC as the primary exposure revealed a substantially stronger HOMA-IR-mediated indirect effect (IE OR = 1.131 [1.018–1.240], p = 0.020; proportion mediated = 29.6%), suggesting that absolute leukocyte counts may capture the metabolic-inflammatory signal more directly than the NLR ratio (Table S2).

E-value analysis: For the fully adjusted NLR-depression association (OR = 1.11), the E-value was 1.46, meaning an unmeasured confounder would need to be associated with both NLR and depression by a factor of ≥ 1.46 to fully explain the observed effect (CI lower bound E-value = 1.28). For WBC (OR = 1.31), the E-value was 1.95 (CI E-value = 1.62). The E-value for the HOMA-IR-mediated indirect effect (IE OR = 1.017) was 1.15 (CI E-value = 1.08), indicating that this smaller effect is more susceptible to unmeasured confounding (Table S3).

4. Discussion

4.1 Main Findings

This large cross-sectional study of 34,302 nationally representative US adults demonstrates that systemic inflammatory markers are robustly associated with depression risk, and that insulin resistance—indexed by HOMA-IR—statistically mediates approximately 9% of the NLR-depression relationship. Effect modification analyses reveal that this metabolic-inflammatory pathway is most prominent in individuals with obesity, males, and adults under age 60. These findings provide epidemiological support for a neuro-immunometabolic model of depression.

4.2 Inflammation and Depression: Direct Associations

Our finding that NLR (OR = 1.11), WBC (OR = 1.31), and CRP (OR = 1.07) are independently associated with depression after comprehensive adjustment is consistent with prior epidemiological evidence. A systematic review of NHANES studies reported similar effect sizes for CRP (Pearce et al., 2019), and meta-analyses of NLR have reported pooled ORs in the range of 1.2–1.5 for depression and anxiety (Zhang et al., 2021). The J-shaped dose-response curve observed for NLR, with risk escalating primarily above the 75th percentile, is consistent with a threshold model where pathological inflammation—rather than physiological immune variation—drives neurobiological changes relevant to mood.

The attenuation of effect sizes across Models 1–3 (e.g., WBC OR from 2.61 to 1.31) suggests that demographic and metabolic factors account for a substantial proportion of the crude inflammatory-depression association. This "confounding-by-indication" pattern is expected given that both inflammation and depression cluster with metabolic comorbidities, poverty, sedentary behavior, and smoking—all captured in Model 3. The persistence of significant associations after full adjustment strengthens the case for an independent inflammatory contribution to depression risk.

4.3 The Role of Insulin Resistance as a Mechanistic Mediator

The key mechanistic finding of this study is that HOMA-IR partially mediates the inflammation-depression association, accounting for 9.0% of the NLR-depression total effect. The biological plausibility of this pathway is well-established:

Inflammatory cytokines (particularly TNF-α, IL-6, and IL-1β) activate serine kinase cascades (IKKβ and JNK) that phosphorylate insulin receptor substrate-1 (IRS-1) at inhibitory serine residues, disrupting downstream PI3K/Akt signaling and causing peripheral insulin resistance (Hotamisligil, 2017). This peripheral insulin resistance has direct central nervous system consequences: brain insulin signaling through the insulin receptor substrate-1/PI3K/Akt pathway modulates monoaminergic neurotransmission, synaptic plasticity (including AMPA receptor trafficking), and hippocampal neurogenesis—all processes implicated in depression pathophysiology (Kleinridders et al., 2014; Duman et al., 2019).

Additionally, insulin resistance promotes dysregulation of the kynurenine metabolic pathway, favoring production of quinolinic acid (a neurotoxic NMDA receptor agonist) over kynurenic acid, which can suppress tryptophan availability for serotonin synthesis (Savitz et al., 2015). Insulin resistance is also associated with elevated glucocorticoid sensitivity and HPA axis dysregulation—a hallmark biological feature of melancholic depression (Raison and Miller, 2003).

The stronger mediation observed for WBC (29.6% mediated) versus NLR (9.0%) may reflect the complementary biological information contained in these markers. NLR is a ratio that can remain stable even when both numerator and denominator change in concert; absolute WBC more directly captures the magnitude of innate immune mobilization. This finding aligns with evidence that total leukocyte burden, rather than immune cell balance, is more closely linked to metabolic inflammation via adipose tissue macrophage infiltration (Lumeng et al., 2007).

4.4 Obesity as an Effect Modifier of the Metabolic-Inflammatory Pathway

The finding that HOMA-IR mediation was significant only in individuals with obesity (BMI ≥ 30) is biologically coherent. In obesity, adipose tissue becomes infiltrated by pro-inflammatory macrophages and switches from an anti-inflammatory (M2) to a pro-inflammatory (M1) activation state, massively amplifying the local and systemic production of inflammatory mediators (Lumeng and Saltiel, 2011). This "metabolic inflammation" (Hotamisligil, 2006) creates a vicious cycle in which adipose-derived inflammation drives insulin resistance, which further promotes adipogenesis and inflammation. In this context, NLR-indexed systemic inflammation more faithfully reflects the mechanistically relevant adipose-tissue inflammatory milieu, and insulin resistance represents a more direct downstream consequence. In normal-weight individuals, by contrast, elevated NLR may reflect other inflammatory exposures (infection, stress, autoimmune activation) that do not necessarily converge on the insulin-resistance pathway.

The greater proportion mediated in males (24.7%) compared to females (2.3%) may partly reflect sex differences in adipose tissue distribution and metabolic inflammation. Men with elevated inflammation are more likely to exhibit visceral adiposity-driven insulin resistance, whereas women with equivalent BMI may harbor more subcutaneous fat, which is metabolically less active (Tchernof and Desprès, 2013). Additionally, estrogen exerts direct anti-inflammatory effects and may attenuate the coupling between systemic inflammation and insulin resistance in premenopausal women (Stubbins et al., 2012).

4.5 Serial Mediation and the Inflammation-MetS-Depression Chain

The serial mediation model—testing the full chain from NLR through HOMA-IR through MetS to depression—yielded a non-significant overall indirect effect (OR = 1.021, p = 0.23) with total proportion mediated of 18.8%. The non-significance likely reflects reduced statistical power from the fasting subsample requirement combined with the binary nature of MetS (which compresses variation compared to the continuous HOMA-IR index). The consistently directional findings across all three pathways in the serial model support the conceptual validity of the proposed chain, even if individual links are modest.

4.6 Limitations

Several limitations warrant consideration. First, the cross-sectional design precludes causal inference; the observed mediation patterns are consistent with but do not establish a causal inflammation → insulin resistance → depression sequence. Longitudinal or Mendelian randomization studies are needed to test causal directionality.

Second, HOMA-IR was available only for the fasting subsample (43.4% of total; n = 9,349 for primary mediation model), which may introduce selection bias if fasting status is correlated with metabolic risk or depression. Third, standard errors in regression models used GLM variance weights rather than full complex survey design (PROC SURVEYREG equivalent); point estimates are unbiased but confidence intervals may be underestimated. Fourth, bootstrap mediation used 200 iterations rather than the conventional 500–1,000, reflecting computational constraints; conclusions should be considered exploratory.

Fifth, CRP data were unavailable for NHANES cycles 2011–2014, reducing statistical power for CRP-based analyses. Sixth, as a secondary analysis of observational data, residual confounding by unmeasured variables (diet quality, circadian rhythm, chronic medication use) cannot be excluded; however, E-values suggest that implausibly strong confounders would be required to nullify the primary associations.

4.7 Clinical and Public Health Implications

These findings carry potential clinical implications for depression prevention and management. The identification of insulin resistance as a mediating pathway suggests that metabolic interventions—including aerobic exercise (which simultaneously reduces systemic inflammation and improves insulin sensitivity), metformin (an insulin sensitizer with documented anti-inflammatory properties), and dietary modifications targeting glycemic load—may warrant investigation as adjunctive strategies for depression prevention, particularly in at-risk subgroups (obese individuals, males with elevated inflammatory markers). Several pilot trials of metformin in bipolar depression and anti-cytokine therapy (infliximab, tocilizumab) in treatment-resistant depression have reported promising results, and our data suggest that insulin-sensitizing effects may be part of the mechanism of action for such interventions (Nierenberg et al., 2020; Raison et al., 2013).

The subgroup finding that the inflammatory-metabolic pathway is most pronounced in younger adults (< 60 years) suggests a potential window of opportunity for preventive intervention before the accumulation of competing metabolic insults and polypharmacy in older age.

5. Conclusions

In this large, nationally representative cross-sectional study, systemic inflammation—measured by NLR, WBC, and CRP—was robustly associated with depression risk across dose-response relationships. Insulin resistance (HOMA-IR) significantly mediated 9.0% of the NLR-depression association, with stronger mediation in individuals with obesity (17.2%), males (24.7%), and adults aged < 60 years (11.9%). These findings advance the inflammatory hypothesis of depression by identifying a specific metabolic mechanism linking peripheral immune activation to mood disorder risk, and highlight insulin resistance as a potential modifiable target in the prevention of inflammation-associated depression.

Funding

This research used publicly available data from the National Health and Nutrition Examination Survey (NHANES), conducted by the Centers for Disease Control and Prevention (CDC).

Conflict of Interest

The authors declare no conflicts of interest.

Data Availability

All NHANES data files are publicly available at https://www.cdc.gov/nchs/nhanes/. Analysis scripts are available upon request.

References

1. Bonaccorso S, et al. (2002). Increased depressive ratings in patients with hepatitis C receiving interferon-alpha–based immunotherapy are related to interferon-alpha-induced changes in the serotonergic system. Journal of Clinical Psychopharmacology, 22(1), 86–90.

2. Capuron L, et al. (2002). Association between decreased serum tryptophan concentrations and depressive symptoms in cancer patients undergoing cytokine therapy. Molecular Psychiatry, 7(5), 468–473.

3. Castillo-Armengol J, et al. (2019). Inter-organ communication: a gatekeeper for metabolic health. EMBO Reports, 20(9), e47903.

4. Dantzer R, et al. (2008). From inflammation to sickness and depression: when the immune system subjugates the brain. Nature Reviews Neuroscience, 9(1), 46–56.

5. Dantzer R, et al. (2011). Cytokine-induced sickness behaviour: a neuroimmune response to activation of innate immunity. European Journal of Pharmacology, 500(1–3), 399–411.

6. Demir S, et al. (2015). Relation of neutrophil–lymphocyte ratio with the presence and severity of major depressive disorder. Psychiatry and Clinical Neurosciences, 69(9), 567–574.

7. Duman RS, et al. (2019). Synaptic plasticity and depression: new insights from stress and rapid-acting antidepressants. Nature Medicine, 25(4), 548–560.

8. Haapakoski R, et al. (2015). Cumulative meta-analysis of interleukins 6 and 1β, tumour necrosis factor α and C-reactive protein in patients with major depressive disorder. Brain, Behavior, and Immunity, 49, 206–215.

9. Hamer M, et al. (2019). Bidirectional association between physical activity and symptoms of anxiety and depression: the Whitehall II cohort study. European Journal of Epidemiology, 24(10), 623–628.

10. Hotamisligil GS. (2006). Inflammation and metabolic disorders. Nature, 444(7121), 860–867.

11. Hotamisligil GS. (2017). Foundations of immunometabolism and implications for metabolic health and disease. Immunity, 47(3), 406–420.

12. Kleinridders A, et al. (2014). Insulin action in brain regulates systemic metabolism and brain function. Diabetes, 63(7), 2232–2243.

13. Köhler CA, et al. (2018). Peripheral cytokine and chemokine alterations in depression: a meta-analysis of 82 studies. Acta Psychiatrica Scandinavica, 135(5), 373–387.

14. Kroenke K, Spitzer RL, Williams JB. (2001). The PHQ-9: validity of a brief depression severity measure. Journal of General Internal Medicine, 16(9), 606–613.

15. Lumeng CN, Saltiel AR. (2011). Inflammatory links between obesity and metabolic disease. Journal of Clinical Investigation, 121(6), 2111–2117.

16. Lumeng CN, et al. (2007). Phenotypic switching of adipose tissue macrophages with obesity is generated by spatiotemporal differences in macrophage subtypes. Diabetes, 56(1), 16–23.

17. Luo H, et al. (2020). Neutrophil-to-lymphocyte ratio and platelet-to-lymphocyte ratio are associated with risk of anxiety disorders in primary care patients. General Hospital Psychiatry, 66, 8–14.

18. Miller AH, Raison CL. (2016). The role of inflammation in depression: from evolutionary imperative to modern treatment target. Nature Reviews Immunology, 16(1), 22–34.

19. Nierenberg AA, et al. (2020). Bipolar heterogeneity: a roadmap for tailored and integrated care. JAMA Psychiatry, 77(2), 175–183.

20. Pearce BD, et al. (2019). Inflammation and mental disorders—from clinical associations to mechanistic implications. Psychiatric Research and Neuroscience, 1, 14–29.

21. Preacher KJ, Hayes AF. (2008). Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behavior Research Methods, 40(3), 879–891.

22. Raison CL, Capuron L, Miller AH. (2006). Cytokines sing the blues: inflammation and the pathogenesis of depression. Trends in Immunology, 27(1), 24–31.

23. Raison CL, et al. (2013). A randomized controlled trial of the tumor necrosis factor antagonist infliximab for treatment-resistant depression. JAMA Psychiatry, 70(1), 31–41.

24. Raison CL, Miller AH. (2003). When not enough is too much: the role of insufficient glucocorticoid signaling in the pathophysiology of stress-related disorders. American Journal of Psychiatry, 160(9), 1554–1565.

25. Savitz J, et al. (2015). Neuroprotective kynurenine metabolite indices are abnormally reduced and positively associated with hippocampal and amygdalar volume in bipolar disorder. Psychoneuroendocrinology, 52, 200–211.

26. Stubbins RE, et al. (2012). Oestrogen alters adipocyte biology and protects female mice from adipocyte inflammation and insulin resistance. Diabetes, Obesity and Metabolism, 14(1), 58–66.

27. Su L, et al. (2019). Inflammation, anxiety, and emotional disorder: a meta-analysis of the link between peripheral inflammatory markers and anxious mood. BMC Psychiatry, 19(1), 1–9.

28. Tchernof A, Desprès JP. (2013). Pathophysiology of human visceral obesity: an update. Physiological Reviews, 93(1), 359–404.

29. VanderWeele TJ, Ding P. (2017). Sensitivity analysis in observational research: introducing the E-value. Annals of Internal Medicine, 167(4), 268–274.

30. World Health Organization. (2023). Depressive disorder (depression). WHO Fact Sheet. Geneva: WHO.

31. Zhang Y, et al. (2021). Neutrophil-to-lymphocyte ratio, a novel biomarker for predicting the severity of COVID-19-associated coagulopathy and the occurrence of adverse clinical events: a systematic review and meta-analysis. Journal of Clinical Medicine, 10(3), 548.

Tables and Figures

Table 1. Baseline characteristics of NHANES 2005–2018 participants by depression status (see attached CSV)

Table 2. Associations between inflammatory markers and depression: multivariable logistic regression and NLR quartile analysis (see attached CSV)

Table 3. Mediation analysis: indirect effects of inflammatory markers on depression through metabolic pathways (see attached CSV)

Table 4. Stratified mediation analysis: effect modification by BMI, sex, and age (see attached CSV)

Figure 1. Structural equation model path diagram showing the inflammation–insulin resistance–depression pathway with standardized coefficients and 95% confidence intervals.

Figure 2. Restricted cubic spline dose-response curves showing associations between log-transformed inflammatory markers (NLR, WBC, CRP) and depression risk (OR relative to median), with 95% confidence intervals. Prediction range restricted to 10th–90th percentiles.

Figure 3. Forest plot of indirect (HOMA-IR-mediated) effects on the NLR-depression association stratified by BMI category, sex, and age group.

Supplementary Table S1. Sensitivity analysis using PHQ-9 ≥ 15 as alternative depression definition.

Supplementary Table S2. Sensitivity analysis using WBC as primary inflammatory marker.

Supplementary Table S3. E-value analysis for primary associations and indirect effects.