As previously reported, from July 2020 to June 2021, 63,564 adults completed routine health screenings at West China Hospital’s Health Management Center (11). After applying exclusion criteria: 49.0% (n = 31,233) were removed due to incomplete depression assessments, followed by 5% (n = 1,570) with missing hematological biomarkers (RDW/HB). Subsequent exclusions comprised age <18 years (n = 5), cancer diagnoses (n = 127), and cardiovascular conditions (n = 202), resulting in 30,427 eligible participants (Figure 1). Specifically, cardiovascular disease (CVD) and cancer were excluded because these conditions can substantially alter hematologic indices (including HRR) and risk of elevated depressive symptoms through systemic inflammation, metabolic disturbances, treatment effects, and comorbidity burden. Furthermore, they introduce complex bidirectional relationships (e.g., depression as a risk factor for CVD and a prognostic factor in cancer) that obscure the independent association between HRR and depression (12, 13). Comparative analysis revealed no meaningful differences in baseline characteristics between the excluded individuals and the final group (Supplementary Table S1).

This investigation analyzed a single-institution clinical dataset from the Health Management Center of West China Hospital (Chengdu, China) collected during in-person health examinations; it was not a secondary analysis of NHANES or any other open repository. All inclusion/exclusion criteria (including a priori exclusion of cardiovascular disease and cancer due to strong bidirectional confounding on hematological indices and depression) were prespecified before analysis, and the participant flow and comparability of excluded vs. included samples are documented. The Ethics Committee of West China Hospital approved this research (No. 2018–303), with all participants providing written consent.

Depressive symptoms were evaluated using the Chinese version of the Zung Self-Rating Depression Scale (SDS), originally developed by Zung in 1965 (14). The SDS, a 20-item validated instrument, was administered to assess depressive symptoms experienced over the preceding 7 days. Each item employs four frequency-based responses: (1) never, (2) sometimes, (3) frequently, and (4) always. The raw score (sum of all items, range 20–80) was converted to a standardized score by multiplying by 1.25. In this study, we implemented a threshold of 53 for depression screening (standardized score ≥53 indicating clinically significant symptoms) (15). The Zung Self-Rating Depression Scale (SDS) was administered by trained clinicians using standardized protocols, with a cutoff of ≥53 selected based on validation studies in Chinese populations demonstrating robust sensitivity/specificity (16). While differing cutoffs exist across settings, this threshold aligns with our center’s standardized practice calibrated against clinical evaluations. Questionnaires were completed during routine health examinations. However, while SDS has acceptable psychometric properties in Chinese populations, it is a screening tool rather than a diagnostic interview; therefore it provides information on symptom burden rather than a clinical DSM/ICD diagnosis established by structured interview. In this study we operationalized the outcome as “elevated depressive symptoms (SDS ≥ 53),” consistent with our center’s screening practices and published validation studies. To mitigate administration variability, participants self-administered the SDS within 20–30 min under standardized conditions; for individuals with visual impairments or limited literacy (<6 years of education), trained staff neutrally read items without interpretive guidance to ensure unbiased responses.

After fasting overnight for 10–12 h, peripheral blood samples were collected in the morning by experienced nurses at the Health Management Center of West China Hospital using EDTA anticoagulant tubes. Hematological parameters including hemoglobin (Hb) and red blood cell distribution width (RDW) were analyzed following standardized laboratory protocols (17). The hemoglobin-to-red blood cell distribution width ratio (HRR) was calculated as hemoglobin (g/dL) ÷ RDW (%) measured by automated hematology analyzers, and for clarity, HRR values were uniformly reported with two decimal places throughout the manuscript. Covariates encompassed demographic factors age, sex, occupation and education level, lifestyle indicators (smoking status and alcohol use), clinical parameters (hypertension, diabetes, hyperlipidemia, kidney disease), and anthropometric measures (BMI). It is worth mentioning that, to improve precision of confounding control, we specifically included kidney disease as an additional covariate in our analysis. Kidney disease status was determined based on participants’ self-report of having ever been told by a health professional that they had kidney disease, consistent with prior epidemiological studies.

Covariates were selected based on established risk factors for depression identified in previous literature and clinical relevance (18). Age and sex were included as fundamental demographic variables. BMI, smoking, and alcohol consumption represent modifiable lifestyle factors (19). Education and occupation reflect socioeconomic status (20). Hypertension and diabetes were included as they may share inflammatory pathways with depression (21). Data collection adhered to consistent methodologies and quality controls as in our prior work (11).

In our study, analyses were prespecified around a single primary exposure (HRR) and a single primary outcome (SDS-defined depression). Multivariable models were used throughout; no univariate screening was employed to select covariates. Subgroup and ROC analyses were considered exploratory; exact p-values are reported and interpretation emphasizes effect sizes and confidence intervals rather than multiplicity-driven dichotomization. All the reporting follows the STROBE Statement (22). Continuous variables were presented as mean ± standard deviation (SD) and categorical variables as frequencies (percentages). To assess the normality of continuous variables, we use the Shapiro–Wilk test combined with visual inspection of Q-Q plots and histograms. Variables following normal distribution were presented as mean ± standard deviation, while non-normally distributed variables were presented as median (interquartile range). Between-group comparisons (depressed vs. non-depressed) utilized Student’s t-tests for continuous measures and χ² tests for proportions.

Logistic regression models quantified associations between HRR (analyzed both continuously and by quartiles) and risk of elevated depressive symptoms across four sequential adjustments: crude model (unadjusted), Model 1 (age, sex), Model 2 (+BMI, smoking, alcohol use), and Model 3 (+education, occupation, chronic diseases). Covariates were adjusted across sequential models given their established dual associations with both depression pathophysiology and hematological regulation—where age, sex, lifestyle factors, and cardiometabolic conditions demonstrably influence HRR values while concurrently modifying risk of elevated depressive symptoms through shared inflammatory pathways. This hierarchical adjustment strategy specifically addresses the biological interdependence observed in our stratified analyses, ensuring HRR’s independent association is isolated from confounding by these covarying factors. In addition, we computed E-values for the point estimate and the 95% CI lower bound to evaluate the minimum strength of association an unmeasured confounder would need with both exposure and outcome to move the observed OR to the null (23). We emphasize that because exposure and outcome were measured concurrently, the study design is cross-sectional and precludes causal inference. The analyses thus quantify associations and should be treated as hypothesis-generating. We computed E-values for the point estimates and the lower bound of the 95% CI to quantify the minimum strength of association that an unmeasured confounder would need with both exposure and outcome to fully account for the observed association. While informative, E-values do not eliminate the possibility of residual confounding.

To address multicollinearity concerns arising from the mathematical relationship HRR = hemoglobin/RDW, we employed generalized variance inflation factor (GVIF) diagnostics. Crucially, regression models never simultaneously included HRR along with its component variables (hemoglobin or RDW) to avoid inherent mathematical collinearity. For all covariates, GVIF values were computed, with degrees-of-freedom adjusted values [GVIF^(1/(2*Df))] calculated for multi-categorical variables (e.g., education, occupation). Adjusted values exceeding √5 (≈2.24) were considered indicative of problematic collinearity (24).

Subgroup analyses assessed interaction effects through stratified models adjusted for covariates excluding the stratification variable, with heterogeneity tested via multiplicative interaction test. Analyses were conducted using R 4.2.3 and Free Statistics software (version 2.0; Beijing Free Clinical Medical Technology Co., Ltd., Beijing, China), with two-tailed p < 0.05 defining significance. Receiver operating characteristic (ROC) curve analysis was conducted to compare the discriminatory performance of HRR against conventional hematological markers (hemoglobin and RDW) for depression identification. Areas under the curve (AUCs) were calculated with 95% confidence intervals, and statistical comparisons between ROC curves were performed using DeLong’s test.

The study population was carefully selected to balance scientific rigor with real-world relevance. We excluded conditions like heart disease and cancer since they independently affect both blood markers and risk of elevated depressive symptoms, while retaining anemia and kidney issues as these may naturally link to elevated depressive symptoms development. This approach was validated through sensitivity checks and population comparisons, with full documentation in Supplementary Materials. This study included 30,427 Chinese adults, with 6,261 (20.58%) exhibiting depressive symptoms (Table 1). Compared to non-depressed individuals, those with elevated depressive symptoms demonstrated distinct socio-demographic patterns: higher female proportion (53.62% vs. 44.38%), lower educational attainment (college or above: 42.17% vs. 60.50%), and overrepresentation in agricultural/industrial occupations (10.99% vs. 6.46%) (all p < 0.0001). Metabolic profiles revealed elevated diabetes prevalence in the elevated depressive symptoms group (9.25% vs. 7.59%, p < 0.0001), though hypertension and hyperlipidemia showed comparable rates. Hematological parameters exhibited significant divergence, with depressed participants showing lower hemoglobin (14.60 ± 1.71 vs. 14.82 ± 1.66 g/dL), higher RDW (13.23 ± 1.21% vs. 13.16 ± 1.14%), and consequently reduced HRR (1.11 ± 0.17 vs. 1.14 ± 0.17, all p < 0.0001). Lifestyle factors also showed differences, with current smoking slightly more prevalent in elevated depressive symptoms group (22.70% vs. 20.50%, p < 0.001).

To further examine the robustness of the HRR-depression association, we stratified participants into quartiles based on HRR values (Supplementary Table S2). The HRR quartiles were defined as Q1 (≤1.031, n = 7,559), Q2 (1.031–1.142, n = 7,663), Q3 (1.142–1.25, n = 7,705), and Q4 (≥1.25, n = 7,500). Quartile-based analysis revealed a graded inverse association between HRR and elevated depressive symptoms: compared with the lowest quartile (Q1 ≤ 1.031), participants in Q2 (1.031–1.142), Q3 (1.142–1.25), and Q4 (≥1.25) exhibited progressively lower odds of elevated depressive symptoms. The proportion of participants with elevated depressive symptoms decreased progressively across quartiles: 23.77% in Q1, 21.36% in Q2, 19.71% in Q3, and 17.44% in Q4 (p < 0.0001). This represents a reduction in elevated depressive symptoms prevalence from the lowest to highest HRR quartile. The quartile analysis also revealed systematic variations in demographic and clinical characteristics: higher HRR quartiles were associated with younger age, higher BMI, better educational attainment, increased prevalence of government/institutional employment, and higher rates of metabolic conditions including hypertension, diabetes, and hyperlipidemia. Notably, higher HRR quartiles differed systematically in these sociodemographic and clinical features.

Logistic regression analyses revealed a consistent inverse graded statistical association between HRR and risk of elevated depressive symptoms across progressively adjusted models (Table 2). In unadjusted analyses, each unit elevation in HRR as a continuous variable was associated with a 52% lower risk of elevated depressive symptoms (Crude OR = 0.48, 95% CI: 0.41–0.56, p < 0.0001). This protective effect persisted through sequential adjustments: controlling for age and sex attenuated the association slightly (Model 1 OR = 0.49, 95% CI: 0.41–0.57, p < 0.0001), while further inclusion of BMI and lifestyle factors (smoking/drinking) maintained a 49% risk reduction (Model 2 OR = 0.51, 95% CI: 0.43–0.60, p < 0.0001). Even after comprehensive adjustment for socioeconomic (education/occupation) and cardiometabolic confounders (hypertension/diabetes/hyperlipidemia), HRR retained significant predictive capacity, with each unit increase conferring 39% lower odds of elevated depressive symptoms (Model 3 OR = 0.61, 95% CI: 0.51–0.72, p < 0.0001). Quartile-based analysis revealed a dose–response pattern: compared to the lowest quartile (Q1: ≤1.031), participants in higher HRR quartiles demonstrated progressively lower risks, with the most pronounced effect in Q4 (≥1.25) showing 24% risk reduction after full adjustment (OR = 0.76, 95% CI: 0.70–0.83, p < 0.0001). The quartile-stratified analysis demonstrated a progressive risk of elevated depressive symptoms reduction with ascending HRR levels in fully adjusted models (Table 2; Figure 2). Compared to the lowest quartile (Q1 ≤ 1.031), participants in Q2 (1.031–1.142) exhibited 10% lower risk (OR = 0.90, 95% CI: 0.84–0.98, p = 0.01), while those in Q3 (1.142–1.25) and Q4 (≥1.25) showed 13% (OR = 0.87, 95% CI: 0.80–0.94, p < 0.001) and 24% (OR = 0.76, 95% CI: 0.70–0.83, p < 0.0001) reduced odds, respectively. A strong dose–response pattern was confirmed by the significant trend test (p < 0.0001). Restricted cubic spline analyses confirmed a linear rather than threshold relationship, with risk of elevated depressive symptoms progressively decreasing across the entire HRR spectrum in all models (Figure 3). Multicollinearity diagnostics revealed acceptable levels for all covariates, with adjusted variance inflation factors ranging from 1.00 to 1.80, all below the recommended threshold of 2.24 (Supplementary Table S3).

Additionally, ROC curve analysis demonstrated that HRR exhibited superior predictive performance for elevated depressive symptoms compared to individual hematological parameters. HRR achieved an AUC of 0.630 (95% CI: 0.623–0.638), significantly outperforming hemoglobin (AUC = 0.542, 95% CI: 0.534–0.550) and RDW (AUC = 0.518, 95% CI: 0.510–0.527) (both p < 0.001, Supplementary Figure S2). These results confirm that the composite HRR index provides enhanced discriminatory ability over conventional hematological markers for identifying individuals with elevated depressive symptoms. The discriminative ability of HRR for elevated depressive symptoms showed an AUC of 0.63, a value that falls within the range (around 0.60–0.65) generally recognized in psychiatric epidemiology as providing useful, albeit moderate, predictive information for biomarkers. This AUC value thus confirms the meaningful predictive value of HRR in the context of elevated depressive symptoms assessment and to be more clear, this level of discrimination suggests HRR may provide incremental value when combined with demographic and clinical factors, but is insufficient as a stand-alone diagnostic marker.

To evaluate the robustness of the observed association, we conducted a sensitivity analysis excluding participants with anemia (<13 g/dL for male and <12 g/dL for female) (25) and those with kidney disease based on self-report (Supplementary Table S4). A total of 2,618 individuals with anemia and 104 with kidney disease (2,713 in total, accounting for overlapping cases) were removed from the analytic sample. The inverse association between HRR and elevated depressive symptoms remained consistent in both direction and magnitude. In the unadjusted model, each unit increase in HRR was associated with a 63% lower risk of elevated depressive symptoms (Crude OR = 0.37, 95% CI: 0.30–0.46, p < 0.0001), and this protective association persisted after progressive adjustment for demographic, lifestyle, socioeconomic, and cardiometabolic factors (Model 3 OR = 0.56, 95% CI: 0.45–0.70, p < 0.0001). Quartile analysis similarly showed a graded protective effect, with the highest quartile (Q4) having 24% lower odds of elevated depressive symptoms compared with Q1 after full adjustment (OR = 0.76, 95% CI: 0.69–0.83, p < 0.0001; p for trend < 0.0001) (Supplementary Table S4). These results indicate that the association between HRR and risk of elevated depressive symptoms is unlikely to be driven by participants with anemia or kidney disease.

To quantify robustness to unmeasured confounding, we computed the E-value for the association between HRR and elevated depressive symptoms. For the highest versus lowest HRR quartile comparison (OR = 0.76), the calculated E-value was 1.88 (Supplementary Figure S1). Thus, an unmeasured confounder would need to be associated with both HRR and elevated depressive symptoms by a risk ratio of at least 1.88 each, above and beyond the measured covariates, to explain away the observed association. This indicates moderate robustness to potential unmeasured confounding.

Age significantly modified HRR-depression associations (p-interaction = 0.004), with greater protection in older adults (≥45 years: OR = 0.401, 95% CI: 0.318–0.505) versus younger individuals (<45 years: OR = 0.639) (Table 3). No other subgroups showed significant interactions (all p-interaction >0.05), including sex (p = 0.26), BMI (p = 0.127), smoking (p = 0.898), and cardiometabolic conditions (hypertension p = 0.931; diabetes p = 0.851). Protective effects remained consistent across occupational (p = 0.895) and educational strata (p = 0.094), confirming HRR’s robust elevated depressive symptoms prevention capacity across diverse populations.

In this large cross-sectional sample of Chinese adults, higher HRR was associated with lower prevalence of elevated depressive symptoms, particularly among individuals aged ≥45 years. These findings suggest HRR may serve as a readily available biomarker for elevated depressive symptoms screening in clinical practice. However, these results are preliminary and hypothesis-generating; prospective studies measuring additional biological and psychosocial covariates are required to determine causality and clinical utility.