Introduction

Depressive disorder is a prevalent mental health condition characterized by prolonged periods of low mood or diminished interest in activities. In severe cases, it can lead to suicide, with an estimated 727,000 people worldwide losing their lives to suicide in 20211. In 2019, depression accounted for 37.3% of disability-adjusted life-years attributable to mental disorders and ranked second among the top causes of years lived with disability globally2. Despite the widespread use of antidepressant medications, nearly one-third of patients remain treatment-resistant3,4, underscoring the need to identify novel and modifiable factors beyond conventional neurotransmitter-based mechanisms.

Emerging evidence suggests that dyslipidemia may be involved in the pathophysiology of depression. Elevated triglyceride (TG) levels may contribute to depression through inflammatory pathways5,6. Similarly, hypercholesterolemia has been associated with increased inflammation and impaired serotonin neurotransmission, potentially contributing to depressive symptoms7. However, evidence of the link between lipids and depression remains inconclusive. One study reported an association between low total cholesterol (TC) levels and a higher risk of depression8, whereas others found no significant association9,10. Similarly, studies based on the National Health and Nutrition Examination Survey (NHANES) also yielded conflicting results regarding the association between TG levels and depression11,12,13.

Most prior studies have focused on individual lipid parameters rather than their potential interactive effects. Cardiovascular studies have demonstrated that the combination of high TG and low high-density lipoprotein cholesterol (HDL-C) exerts synergistic effects on disease risk14. Building on this evidence, we hypothesized that interactions between TG and cholesterol subtypes would be associated with depression. Therefore, this study aimed to evaluate the interactive effects of specific lipid pairs (TG/TC, TG/low-density lipoprotein cholesterol [LDL-C], and TG/HDL-C) on the odds of depression.

Methods

Study design and participants

The NHANES assesses the health and nutritional status of the non-institutionalized U.S. civilian population through household interviews, standardized physical examinations, laboratory testing, and detailed demographic and dietary data. The study protocol was approved by the National Center for Health Statistics Research Ethics Review Board, and all participants provided written informed consent. For this study, we pooled data from seven consecutive survey cycles (2005–2006 to 2017–2018) to increase the sample size and generate nationally representative estimates.

Study variables

Serum lipids, including TG, TC, HDL-C, and LDL-C, served as exposure variables, and detailed analytical protocols for their quantification are comprehensively documented in the NHANES Laboratory Procedure Manual.

Depression was measured using the Patient Health Questionnaire-9 (PHQ-9), a validated 9-point instrument quantifying symptom frequency (0 = never to 3 = nearly daily), during the preceding fortnight. Higher total scores correspond to greater depression severity15. In this study, participants were stratified into non-depression (score range, 0–9) and depression (score range, 10–27) groups based on a validated clinical cutoff10.

Covariates

Covariates included age stratification, sex, race/ethnicity, educational attainment, marital status, family poverty-to-income ratio (PIR) categories, body mass index (BMI) categories, smoking status, alcohol consumption status16, physical activity (PA), estimated glomerular filtration rate (eGFR), diabetes, hypertension, and use of antidepressants and antihyperlipidemic drugs. The classification criteria for the indicators are listed in Table 1.

Table 1 Weighted distributions of characteristics across depression subgroups.
Full size table

PA was quantified as weekly metabolic equivalent of task (MET)-minutes. MET values were computed using established methods17. Activity levels were categorized according to American guidelines: inactive (0 MET-min/week), insufficiently active (1–599 MET-min/week), and sufficiently active (≥ 600 MET-min/week)18. eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration 2021 equation (CKD-EPI 2021) and categorized according to established criteria19,20. Diabetes was defined by clinical diagnosis (self-reported physician diagnosis or current use of antidiabetic medications) or laboratory confirmation (fasting plasma glucose ≥ 7.0 mmol/L or HbA1c ≥ 6.5%)21. Hypertension was defined as physician-diagnosed hypertension, active antihypertensive medication use, or blood pressure ≥ 130/80 mmHg22. Antidepressant and antihyperlipidemic drug use was identified using a prescription medication questionnaire from NHANES.

Statistical analysis

The NHANES data analysis incorporated sampling weights to ensure national representativeness23. In this study, the fasting subsample 2-year mobile examination center weight (WTSAF2YR) was divided by 7 and used as the final weight. Differences in variables were assessed using weighted linear regression or Rao-Scott adjusted chi-squared tests.

Three survey-weighted generalized linear models (GLMs) were used to evaluate lipid-depression associations: Model 1 (unadjusted); Model 2 (adjusted for sociodemographic variables); and Model 3 (Model 2 additionally adjusted for smoking status, alcohol consumption, hypertension, diabetes, physical activity, and eGFR). The potential nonlinear relationship between lipid levels and depression were evaluated in Model 3 using survey-weighted restricted cubic spline (RCS) regression with 3 knots and pre-smoothing was applied to minimize overfitting prior to formal analysis. Variance inflation factors (VIFs) were calculated to assess multicollinearity, excluding variables with VIFs ≥ 524. Final VIFs are provided in Tables S1–S4.

This study employed survey-weighted GLMs (using the survey R package) to examine the interaction effects between continuous TG and various cholesterol levels, adjusting for the covariates in Model 3. For significant interaction terms, participants were dichotomized using lipid 2 thresholds corresponding to percentiles ranging from 2.5 to 97.5% in 2.5% increments. Within each subgroup, survey-weighted GLMs assessed the association between lipid 1 and depression odds, interaction testing was performed using likelihood ratio tests, with resulting P values adjusted for multiple comparisons via Bonferroni, Holm, and Benjamini–Hochberg methods. The optimal lipid 2 cutoff value was determined by selecting the points where the P values in the three multiple comparison adjustments are the smallest, with at least one being less than 0.05. In addition, to visualize the interaction effect between lipid parameters, the marginal effect of lipid 1 on depression across the observed range of lipid 2 was estimated as: βlipid 1 + βinteraction × lipid 2. Odds ratios (ORs) were calculated at varying levels of LDL-C, and corresponding 95% confidence intervals (CIs) were derived using the delta method under the complex survey design.

Sensitivity analyses were conducted under eight distinct analytical conditions to evaluate the robustness of the primary outcomes (Table 5). For conditions requiring handling of missing data, multiple imputations were performed using the mice R package. Detailed specifications of the multiple imputation procedure, including imputation methods, variables incorporated, and diagnostic checks, are provided in Table S5. All statistical procedures were performed using R version 4.3.3 (R Foundation) and Free Statistics platform v2.3, with P < 0.05 determining significance.

Results

Baseline characteristics

Following the screening protocol in Fig. 1, a total of 8962 eligible participants were included in the analysis. Significant differences were observed between depression and non-depression groups. Females predominated in the depression cohort (62.0%, P < 0.001). Although age showed no significant difference (P = 0.571), notable disparities were found in marital status, race, education, PIR, smoking, diabetes, BMI, PA, eGFR, TG, and HDL-C (all P < 0.05) (Table 1), collectively underscoring the multifactorial interplay between depression and sociodemographic, behavioral, and cardiometabolic determinants.

Figure 1
Figure 1
Full size image

Participant inclusion flowchart.

Baseline characteristics of included and excluded participants are shown in Table S6. Except for age group and education level, all standardized mean differences were < 0.20, indicating generally comparable distributions between groups. The observed differences in age and education are consistent with the NHANES study design, as the PHQ-9 questionnaire was administered only to participants aged ≥ 18 years.

Association between individual lipids and depression

In the crude analyses, TG showed a positive association with depression, whereas HDL-C demonstrated a negative association (TG: P = 0.036; HDL-C: P = 0.028). These associations were attenuated to marginal significance in Model 2 (TG, P = 0.057; HDL-C, P = 0.079) and became non-significant in Model 3 (Table 2). Subsequent survey-weighted RCS analyses of lipid parameters based on Model 3 revealed no evidence of nonlinear associations with depression (Fig. 2; all P values for nonlinearity > 0.05).

Table 2 Survey-weighted GLMs of associations between lipids and depression.
Full size table
Figure 2
Figure 2
Full size image

Association between lipids and odds of depression. Panels: (a) Triglycerides; (b) Total Cholesterol; (c) High Density Lipoprotein Cholesterol; (d) Low Density Lipoprotein Cholesterol.

Interaction analysis of TG and cholesterol subtypes on the odds of depression

Survey-weighted GLMs demonstrated a significant multiplicative interaction between continuous triglyceride (TG) and LDL-cholesterol (LDL-C) levels in their association with depression (interaction β = 0.158; P = 0.03) (Table 3). The marginal effect analysis showed that the association between TG and depression progressively strengthened as LDL-C increased (Fig. 3). This interaction indicated that the relationship between TG and depression depended on LDL-C, prompting a search for relevant LDL-C thresholds. The optimal stratification point was identified at LDL-C = 3.6 mmol/L (Bonferroni, Holm, and Benjamini–Hochberg adjusted P were 0.061, 0.061, and 0.023, respectively, see Table S7). In contrast, no statistically valid TG cutoff value emerged to modify LDL-C–depression associations (all adjusted P > 0.05) (Table S8).

Table 3 Interaction effects of lipid profiles on depression.
Full size table
Figure 3
Figure 3
Full size image

Marginal effect of TG across LDL-C levels.

The participants were stratified into subgroups based on predefined LDL-C cutoff: LDL-C levels ≥ 3.6 mmol/L, TG was found to have a positive association with depression (OR 1.78, 95% CI 1.30–2.42; P < 0.001), with a significant interaction effect (Pinteraction = 0.002) (Table 4). These findings underscore the lipid-level-dependent synergism in the odds of depression.

Table 4 Subgroup analysis of the association between TG and depression.
Full size table

Sensitivity analyses

Across all tested conditions, a consistently positive association between triglycerides (TG) and depression was observed in the LDL-C ≥ 3.6 mmol/L subgroup (adjusted ORs 1.41–1.79; all P ≤ 0.005), whereas no significant association was detected in the LDL-C < 3.6 mmol/L subgroup (adjusted ORs 0.90–1.07; all P > 0.05). Interaction tests were significant in four of eight analytic conditions (Pinteraction ≤ 0.03) (Table 5).

Table 5 Sensitivity analyses.
Full size table

To further minimize potential optimism bias, bootstrap resampling under the complex survey design was conducted using 1,000 replicate weights. The replicate-weighted GLM, based on the fully adjusted Model 3 and incorporating the TG × LDL-C group interaction term, demonstrated a significant interaction (OR 1.73, 95% CI 1.25–2.41, P = 0.002). These findings further support the robustness of the identified LDL-C cutoff.

Discussion

Individual analyses revealed no significant association between lipid components and depression. In contrast, when exploring the interaction effects, a significant interaction between LDL-C and TG levels on depression was identified. Quantile-driven analysis revealed a significant positive association between TG levels and depression specifically when LDL-C exceeded 3.6 mmol/L. Conversely, no such association was observed below this LDL-C threshold. By identifying LDL-C ≥ 3.6 mmol/L as a key effect modifier amplifying the TG-depression link, our findings provide an exploratory cutoff value for advancing ‘precision psychiatry’ approaches. Although the magnitude of some associations was modest, statistical significance does not necessarily imply clinical importance at the individual level. However, given the high prevalence of dyslipidemia and depression in the general population, even relatively small increases in relative risk may have meaningful implications at the population level, potentially affecting a substantial number of individuals.

Our analysis revealed a positive interaction between LDL-C and TG levels and the odds of depression. Critically, neither 3.388 mmol/L nor 3.465 mmol/L—values bracketing the conventional 3.4 mmol/L clinical threshold25,26—demonstrated statistically significant adjusted interaction P values, confirming that 3.4 mmol/L does not represent the optimal modifying threshold for TG-depression associations. Instead, statistical significance emerged at LDL-C = 3.595 mmol/L (Padjusted < 0.05), establishing 3.6 mmol/L as a meaningful modifying threshold for TG-depression relationships in this population, with robustness confirmed through sensitivity analyses.

The interaction analysis revealed that LDL-C significantly modified the association between TG and depression which may operate through sequential neuroinflammatory mechanisms. Elevated LDL-C levels impair blood–brain barrier integrity27,28,29, facilitating the entry of TG-derived free fatty acids into the central nervous system. These fatty acids activate toll-like receptor 4 receptors on microglia30,31, triggering neuroinflammatory cascades that exacerbate depression pathogenesis, particularly during the concomitant elevation of both lipids. Furthermore, both LDL particles and TG-rich lipoproteins are apolipoprotein B-containing lipoproteins, and elevated LDL-C levels may reflect an increased burden of circulating atherogenic particles. This cumulative apolipoprotein B-containing lipoprotein load has been implicated in endothelial dysfunction and systemic low-grade inflammation, potentially predisposing individuals to neurovascular impairment.

Although the association between LDL-C and depression is theoretically expected to vary significantly across TG levels, our analysis failed to identify statistically significant cutpoints after multiple comparison adjustment. As shown in Table 3, the negative main effect of LDL-C combined with its positive interaction with TG suggests that the net association may manifest as protective, null, or detrimental effects depending on TG concentrations—a finding consistent with prior reports of context-dependent lipid-depression relationships32,33,34. The absence of significant TG thresholds could stem from reduced statistical power in extreme percentiles or undetected nonlinearities.

This study has several strengths. First, combining a nationally representative sample with rigorously implemented inclusion and exclusion criteria and comprehensive sensitivity analyses strengthened the validity of our findings. Second, to our knowledge, this is the first study to demonstrate the synergistic interaction between TG and LDL-C levels on the odds of depression in a cross-sectional cohort. Third, our results further underscore the necessity of incorporating interactive effects into the evaluation of associations between lipid profiles and disease risk.

Nonetheless, several limitations should be acknowledged. First, depression may influence lipid metabolism through behavioral and biological mechanisms. Depressive symptoms are associated with altered dietary intake, reduced physical activity, and dysregulation of the hypothalamic–pituitary–adrenal axis, all of which may contribute to adverse lipid profiles. Therefore, reverse causality cannot be excluded in this cross-sectional analysis. Second, as the study population was derived from U.S. adults, caution is warranted when extrapolating these findings to populations with different racial and ethnic compositions. Finally, LDL-C was calculated using the Friedewald equation in the included NHANES cycles. Because this method does not yield valid LDL-C values when TG are ≥ 400 mg/dL, individuals with very high TG concentrations were excluded. Consequently, the findings may not be generalizable to populations with TG ≥ 400 mg/d. Future research should address these limitations to strengthen the clinical and mechanistic interpretations.

Conclusion

This study demonstrates that TG exhibit a significant positive association with depression specifically at LDL-C levels ≥ 3.6 mmol/L. Clinicians should prioritize comprehensive lipid profiling over single-metric evaluations when assessing the susceptibility to depression. Future large-scale prospective studies are critical to validate the interaction cutoffs and establish causality, which could reshape preventive strategies for depression.