The total cholesterol to high-density lipoprotein cholesterol ratio predicts post-stroke depression: linear association and individualized risk nomogram

Abstract

Background

Post-stroke depression (PSD) is a common, impactful complication of stroke, yet clinically applicable early prediction tools remain limited. The ratio of total cholesterol to high-density lipoprotein cholesterol (TC/HDL) integrates atherogenic burden and neuroprotective reserve, making it a promising candidate for PSD risk stratification. However, its association with 90-day PSD and utility in predictive models remain unestablished.

Methods

The primary endpoint was PSD at 90 days after stroke. Multivariable logistic regression assessed TC/HDL–PSD association, and restricted cubic spline (RCS) analysis tested non-linearity. A nomogram was constructed and internally validated, with its performance assessed using decision curve, calibration curves, and receiver operating characteristic analysis.

Results

Thirty-five point seven percent of patients developed PSD. TC/HDL was an independent predictor of PSD (OR = 1.2726, P = 0.0028), with a linear relationship (no nonlinearity; P = 0.879). The highest TC/HDL quartile correlated with a 136.26% increased PSD risk (OR = 2.3626, P = 0.0190). The nomogram, incorporating TC/HDL, NIHSS, MoCA, sex, and neutrophil count, demonstrated good discrimination and calibration.

Conclusions

Admission TC/HDL is an independent prognostic marker of 90-day PSD, with a linear dose-response relationship. The internally validated nomogram may assist in risk estimation. However, its clinical applicability remains to be established and requires external validation. Given that TC/HDL is a simple, widely available and low-cost biomarker, these findings may support early risk stratification and preventive psychological care in routine stroke management, particularly in resource-limited clinical settings.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12944-026-02977-4.

Introduction

Approximately 33% of individuals experience post-stroke depression (PSD) within three months following acute ischemic stroke (AIS), making PSD a prevalent neuropsychiatric complication [1]. Accumulating evidence has confirmed that PSD not only severely impedes neurological recovery and quality of life in affected patients [2, 3], but correlates strongly with adverse outcomes such as stroke recurrence and increased mortality [4]. Therefore, identifying early accessible risk markers for PSD in AIS patients holds crucial clinical significance for implementing precise interventions and improving prognostic outcomes [5]. However, the availability of clinically applicable early prediction tools remains limited.

In recent years, the lipid profile, a widely acknowledged determinant of cerebrovascular disease risk, has garnered growing attention for its potential association with post-stroke neuropsychiatric complications—particularly post-stroke depression (PSD) [6–8]. Against this backdrop, the search for objective biomarkers capable of enabling early prediction of PSD has gained increasing research attention. Although the pathophysiology of PSD is thought to involve multiple processes, including neuroinflammation, cerebrovascular dysfunction, and impaired neuroplasticity [9, 10], an ideal predictive marker in clinical practice should be able to reflect overall cerebrovascular and metabolic status rather than individual pathways alone. In this context, the systemic ratio of total cholesterol (TC) to high-density lipoprotein cholesterol (HDL)—termed the TC/HDL ratio—has attracted interest due to its potential value as an integrated assessment tool. Previous studies have suggested abnormal TC levels may be linked to depressive symptoms through effects on neurotransmitter-related processes [11]; whereas HDL is generally considered to exert neuroprotective effects after stroke owing to its anti-inflammatory and antioxidant properties [12]. Thus, the TC/HDL ratio is regarded as a composite indicator that simultaneously evaluates atherogenic burden and the body’s protective physiological reserve. It should be noted, however, that these mechanistic explanations are derived primarily from prior studies and are not directly examined in the present analysis.

Beyond the mechanistic support from the aforementioned pathophysiological pathways, evidence-based medical data on the TC/HDL ratio in cardiovascular and cerebrovascular diseases provide more direct validation of its predictive value. TC/HDL has been established as an important marker of atherosclerotic severity and cardiovascular event risk [13–15]. In the field of cardiology, it is widely recognized as a superior risk assessment tool compared to individual lipid components, as it more effectively quantifies the degree of dyslipidemia. Notably, in cerebrovascular disease research, an abnormally high TC/HDL is positively linked to the onset, severity, and long-term prognosis of stroke [16, 17]. Given that cerebrovascular injury constitutes a critical biological substrate for PSD, and that the TC/HDL ratio can sensitively quantify cerebrovascular-related risk and overall health status, extending its application to PSD risk assessment appears clinically reasonable. However, while its value in prognostic assessment for ischemic stroke has been confirmed, the specific relationship between TC/HDL and PSD risk within 90 days of AIS onset remains unclear. Its independent predictive performance has yet to be systematically evaluated.

From a clinical tool perspective, nomograms have demonstrated significant value in stroke prognosis research due to their ability to integrate multiple factors and enable visualized, individualized risk calculation. Although several studies have attempted to develop risk prediction models for PSD in AIS patients, most have focused on clinical and radiological features. Research incorporating simple, readily accessible lipid markers with comprehensive metabolic significance—such as the TC/HDL ratio—into these models remains scarce, and predictive models for PSD specifically based on the TC/HDL ratio are currently nonexistent. Given that the 90-day post-stroke period represents a high-incidence window for PSD, the development of a nomogram prediction tool integrated with the TC/HDL ratio for this time point is particularly urgent.

Against this background, we hypothesized that TC/HDL would be independently linked to PSD risk 90 days after AIS. The novelty of this investigation lies in introducing the TC/HDL ratio, a simple and clinically accessible composite lipid indicator, into the PSD research framework and integrating it into a predictive nomogram model. Specific objectives include: (1) investigating the association between TC/HDL and PSD at 90 days post-AIS; (2) analyzing the potential nonlinear dose-response relationship between these two variables; and (3) developing a PSD risk nomogram based on independent predictive factors, followed by an evaluation of its discriminative ability, calibration, and clinical applicability in an internal validation cohort. By focusing on a simple and clinically accessible lipid indicator, this study seeks to provide a practical tool to facilitate early identification of PSD risk.

Methods

Patients

This study retrospectively enrolled AIS patients at Shanxi Bethune Hospital (October 2023 – December 2024). Stroke diagnoses adhered to WHO criteria and were verified by MRI or brain CT for all hospitalized individuals.

Patients were included if they met the following criteria: (1) a diagnosis of AIS confirmed by CT or MRI within 72 h of symptom onset, according to the 2018 Chinese AIS guidelines; (2) first-time stroke without previous severe neurological impairment; (3) age 18 years or older; (4) complete clinical records available for analysis.

Exclusion criteria comprised: (1) TIA or hemorrhagic stroke; (2) history of psychiatric illness confirmed by prior diagnosis or treatment; (3) severe aphasia, consciousness disturbance, or inability to complete assessments due to sensory or cognitive deficits; (4) severe cardiopulmonary comorbidities or multiple organ failure; (5) chronic inflammatory disease; (6) incomplete data; (7) follow-up loss. 541 participants were analyzed (Fig. 1).

Fig. 1.

Participant selection flowchart

Clinical data collection

For all patients, demographic and clinical baseline characteristics were documented, encompassing body mass index (BMI), Montreal Cognitive Assessment (MoCA) score, National Institutes of Health Stroke Scale (NIHSS) score, sex, age, educational level, infarct location, responsible lesion, and histories of hypertension (HTN), coronary artery disease (CAD), diabetes mellitus (DM), alcohol consumption, and smoking. MoCA was administered at baseline during the acute stroke hospitalization, prior to the 90-day follow-up assessment of depressive symptoms. Assessors were blinded to subsequent depression status.

Laboratory assessments included hematological and biochemical analyses. For hematology, 2 mL of venous blood in EDTA-K₂ tubes was thoroughly mixed and analyzed with a Beckman DXH800 hematology analyzer to obtain monocyte count (Mon), neutrophil count (Neu), white blood cell count (WBC), platelet count (PLT), and lymphocyte count (Lym). For biochemical tests, 5 mL of venous blood in non-anticoagulant tubes was centrifuged to measure TC, triglycerides (TG), low-density lipoprotein cholesterol (LDL), serum uric acid (UA), HDL, and homocysteine. D-dimer levels were measured in citrated plasma (anticoagulant-to-blood ratio 1:9) using a Werfen ACL-TOP 750 analyzer.

The TC/HDL ratio was defined as the ratio of TC to HDL.

Assessment of study outcomes

The primary outcome was depressive status at 90 days after hospital admission. Two trained physicians assessed depressive symptoms at 3 months post-stroke during outpatient visits or follow-up consultations using the 24-item Hamilton Depression Rating Scale (HAMD-24). PSD was defined as HAMD-24 > 7, a threshold capturing clinically meaningful mild depressive symptomatology [18] and used for screening rather than formal clinical diagnosis. The distribution of HAMD-24 scores and sensitivity analyses using HAMD-24 > 10 and > 14 are reported in Supplementary Materials.

Statistical analysis

Patients were stratified according to the presence or absence of PSD, and baseline characteristics were analyzed accordingly. Normality of continuous variables was evaluated using the Shapiro-Wilk test. Normally distributed data are reported as mean ± SD, while data not following a normal distribution are presented as median (25th, 75th percentiles). For group comparisons, the independent samples t-test was applied to normally distributed variables, and the Mann-Whitney U test to non-normally distributed variables. Categorical variables were expressed as counts and percentages, and differences between groups were assessed with the chi-square test.

Multivariable logistic regression analysis was performed to assess the association between TC/HDL and PSD. Confounding variables were adjusted for based on the results of univariate analyses (P < 0.05) and clinical relevance. Three models were established: Model 1 was unadjusted; Model 2 adjusted for sex, smoking, and alcohol; and Model 3 (fully adjusted model) included further adjustments for MoCA, NIHSS, D-dimer, and neutrophil count.

In sensitivity analyses, a test for linear trend was conducted using quartiles of TC/HDL as the independent variable to evaluate the robustness of the association. A restricted cubic spline (RCS) model was applied to investigate potential nonlinear relationships between TC/HDL and PSD. Additionally, subgroup analyses and interaction tests were performed to investigate the relationship between TC/HDL and PSD across subgroups stratified by sex, drinking, smoking status, HTN, and DM.

A nomogram model was developed following stepwise backward regression for variable selection based on Model 3. Data from the entire cohort of 541 patients were randomly partitioned into a training set (70%) and a validation set (30%) using stratified randomization based on PSD occurrence, with a fixed random seed (set.seed = 123) to ensure reproducibility. Baseline characteristics of the two datasets are shown in Supplementary Table S2. Discriminative ability was assessed using the area under the receiver operating characteristic curve (ROC-AUC), calibration using calibration plots, and clinical utility using decision curve analysis (DCA). Bootstrap internal validation (500 resamples) was conducted to quantify optimism, with correction for AUC, calibration slope, intercept, Brier score, and mean absolute error. Sensitivity analysis was performed by comparing the full model (with MoCA) and a model excluding MoCA to evaluate robustness and incremental predictive value. The incremental predictive value of TC/HDL was assessed in the validation set by comparing Model A (with TC/HDL) versus Model B (without TC/HDL) using ΔAUC, NRI, IDI, and DCA.

Analyses were conducted in R 4.4.2, and statistical significance was set at P < 0.05 (two-tailed).

Results

Baseline characteristics

Table 1 summarizes the final analysis cohort, which included 541 patients—348 in non-PSD group and 193 in PSD group. Significant differences between groups were observed for sex, alcohol consumption, smoking status, MoCA score, NIHSS score, neutrophil count, D-dimer, and the TC/HDL ratio (all P < 0.05).Specifically, the PSD group exhibited a greater proportion of males, higher rates of alcohol use and smoking, significantly elevated D-dimer levels, NIHSS scores, and TC/HDL ratios, whereas MoCA scores were markedly lower.

Table 1.

Participant baseline characteristics

Characteristic	Overall N = 5411	Non-PSD N = 3481	PSD N = 1931	P-value2
Sex				< 0.001
Female	162 (30%)	131 (38%)	31 (16%)
Male	379 (70%)	217 (62%)	162 (84%)
education				0.708
illiteracy	88 (16%)	59 (17%)	29 (15%)
Primary schhool	193 (36%)	118 (34%)	75 (39%)
junior high school	218 (40%)	143 (41%)	75 (39%)
senior high school and above	42 (7.8%)	28 (8.0%)	14 (7.3%)
Smoking				< 0.001
No	220 (41%)	161 (46%)	59 (31%)
Yes	321 (59%)	187 (54%)	134 (69%)
Drinking				< 0.001
No	269 (50%)	192 (55%)	77 (40%)
Yes	272 (50%)	156 (45%)	116 (60%)
HTN				0.636
No	217 (40%)	137 (39%)	80 (41%)
Yes	324 (60%)	211 (61%)	113 (59%)
DM				0.541
No	376 (70%)	245 (70%)	131 (68%)
Yes	165 (30%)	103 (30%)	62 (32%)
CAD				0.438
No	495 (91%)	316 (91%)	179 (93%)
Yes	46 (8.5%)	32 (9.2%)	14 (7.3%)
Lesion Location				0.062
Basal ganglia or lateral ventricles	237 (44%)	142 (41%)	95 (49%)
Brain stem or Cerebellum	152 (28%)	106 (30%)	46 (24%)
Thalamus	15 (2.8%)	12 (3.4%)	3 (1.6%)
cerebral lobe	35 (6.5%)	27 (7.8%)	8 (4.1%)
Multiple infarction	102 (19%)	61 (18%)	41 (21%)
Stroke Location				0.358
Left	339 (63%)	213 (61%)	126 (65%)
Right	180 (33%)	118 (34%)	62 (32%)
Both	22 (4.1%)	17 (4.9%)	5 (2.6%)
NIHSS	3.00 (1.00, 4.00)	2.00 (1.00, 3.00)	4.00 (3.00, 5.00)	< 0.001
Age(years)	63.00 (55.00, 70.00)	62.00 (55.00, 70.00)	63.00 (56.00, 70.00)	0.916
MoCA	25.00 (21.00, 26.00)	25.00 (24.00, 26.00)	21.00 (20.00, 24.00)	< 0.001
BMI	22.44 (18.63, 27.09)	22.16 (18.34, 26.72)	23.15 (19.64, 27.74)	0.055
WBC(×10^9/L)	6.40 (5.30, 7.70)	6.40 (5.20, 7.70)	6.40 (5.53, 7.90)	0.329
Lym(×10^9/L)	2.53 (1.97, 3.36)	2.54 (1.98, 3.38)	2.52 (1.91, 3.32)	0.473
Mon(×10^9/L)	0.53 (0.44, 0.65)	0.52 (0.43, 0.65)	0.55 (0.46, 0.65)	0.302
Neu(×10^9/L)	5.94 (4.77, 6.89)	6.02 (5.16, 6.89)	5.70 (3.42, 6.86)	0.029
PLT(×10^9/L)	238.76 (194.28, 284.48)	235.97 (194.59, 284.20)	241.47 (192.64, 285.27)	0.462
D-dimer(ng/mL)	125.00 (77.00, 253.00)	112.00 (70.00, 193.50)	178.00 (94.00, 354.00)	< 0.001
TG(mmol/L)	1.48 (1.22, 1.75)	1.47 (1.17, 1.82)	1.49 (1.26, 1.72)	0.984
TC(mmol/L)	4.16 (3.47, 4.88)	4.16 (3.51, 4.98)	4.15 (3.44, 4.77)	0.365
LDL(mmol/L)	2.66 (2.22, 3.18)	2.61 (2.19, 3.27)	2.68 (2.27, 3.10)	0.440
Hcy(umol/L)	16.60 (12.90, 26.50)	16.20 (12.90, 25.80)	17.60 (12.90, 27.10)	0.413
UA(mmol/L)	299.00 (244.40, 386.20)	294.90 (240.15, 378.30)	305.60 (255.30, 393.60)	0.073
HDL(mmol/L)	0.93 (0.80, 1.10)	0.99 (0.87, 1.16)	0.83 (0.74, 0.96)	< 0.001
TC/HDL	4.30 (3.58, 5.25)	4.12 (3.42, 4.97)	4.86 (3.91, 5.64)	< 0.001

HTN hypertension, PLT platelet counts, HDL high-density lipoprotein, WBC white blood cell, LDL low-density lipoprotein, Mon monocyte, TC total cholesterol, Neu neutrophil, CAD coronary artery disease, TG triglycerides, Lym lymphocyte, UA uric acid, DM diabetes mellitus, Hcy homocysteine

¹n (%); Median (Q1, Q3)

² Wilcoxon rank sum test; Pearson’s Chi-squared test

Association between TC/HDL and post-stroke depression

Univariate regression analysis (see Supplementary Table S1) demonstrated positive associations between PSD in AIS patients and smoking status, drinking, sex, MoCA score, NIHSS, neutrophil count, D-dimer levels, HDL, and the TC/HDL ratio. Multivariable logistic regression analysis (see Table 2) further confirmed a significant positive correlation between the TC/HDL ratio and PSD. In Model 1 (unadjusted), each 1-unit increase in TC/HDL was associated with a 43.43% increase in PSD risk (OR = 1.4343, P < 0.0001). The association persisted in Model 2 (OR = 1.4109, P < 0.0001). Even after controlling for additional confounders in Model 3, the association continued to demonstrate statistical significance(OR = 1.2726, P = 0.0028).

Table 2.

Multivariable logistic regression of TC/HDL and PSD

Exposures	Model1 OR(95%CI),P-value	Model2 OR(95%CI),P-value	Model3 OR(95%CI),P-value
TC/HDL
Per1-unit increase	1.4343(1.2627, 1.6414),<0.0001	1.4109(1.2396, 1.6190), < 0.0001	1.2726(1.0889, 1.5031),0.0028
Per 1-SD increase	1.7121(1.4159, 2.0935),<0.0001	1.6706(1.3774, 2.0511),<0.0001	1.4392(1.1354, 1.8361),0.0028
Quartiles
Q1	Reference	Reference	Reference
Q2	1.2073(0.6953, 2.1058),0.5040	1.1488(0.6538, 2.0264),0.6298	1.3281(0.6227, 2.8569),0.4639
Q3	2.3883(1.4156, 4.0857),0.0013	2.2784(1.3325, 3.9474),0.0029	2.1835(1.0512, 4.6069),0.0377
Q4	3.7576(2.2426, 6.4099), < 0.0001	3.5975(2.1179, 6.2182), 0.0001	2.3626(1.1588, 4.8967),0.0190
P for trend	< 0.0001	< 0.0001	0.0087

*Model 1 = no covariates were adjusted

Model 2 = Model 1 + sex, smoking, drinking, were adjusted

Model 3 = Model 2 + NIHSS, MoCA, Neu, D-dimer, were adjusted

Additionally, TC/HDL was independently associated with 90-day PSD (Model 3, Table 2): per 1-SD increase increased PSD risk by 43.9% (OR = 1.4392, 95% CI:1.1354–1.8361, P = 0.0028); highest vs. lowest quartile had 2.36-fold higher risk (OR = 2.3626, P = 0.0190) with significant dose–response (P = 0.0087). The test for linear trend yielded a P-value of 0.0087, indicating a significant increasing trend in the risk of PSD as the quartile of TC/HDL rose.

Assessing nonlinear relationships

Results of the RCS analysis (Fig. 2) revealed that the 90-day PSD risk in individuals with AIS increased continuously with rising TC/HDL ratios, revealing a significant positive relationship (P = 0.028). Further nonlinearity testing yielded a P-value of 0.879, indicating no nonlinear relationship.

Fig. 2.

RCS analysis of TC/HDL and PSD

Subgroup analyses

No significant interaction between TC/HDL and the subgroups was found, as shown in Fig. 3 (P for interaction > 0.05).

Fig. 3.

Subgroup analysis of TC/HDL and PSD

Nomogram development and validation

Based on Model 3, stepwise backward regression was used to identify five significant predictive variables: NIHSS score, MoCA score, sex, neutrophil count, and the TC/HDL ratio. These variables were incorporated to develop a nomogram for predicting PSD risk (Fig. 4).

Fig. 4.

Nomogram predicting 90-day PSD risk

The nomogram demonstrated excellent discriminative ability, with AUCs of 0.919 (training) and 0.889 (validation) (Fig. 5). Bootstrap internal validation (500 resamples) confirmed minimal optimism, yielding bias-corrected AUCs of 0.91 (training) and 0.88 (validation) (Supplementary Figure S4, Table S6). Calibration was excellent in the training set (slope 0.896, intercept − 0.005) and acceptable in the validation set (slope 0.705, intercept 0.110). Brier scores and bootstrap-corrected mean absolute error (MAE) further supported good predictive accuracy and stability (Supplementary Table S6). DCA in the validation cohort indicated superior net clinical benefit of the nomogram compared with treat-all and treat-none strategies across threshold probabilities (5.2%–82.5%) (Fig. 5C).

Fig. 5.

Nomogram performance and clinical utility: (A) training set ROC; (B) internal validation ROC; (C) internal validation DCA; (D) internal validation calibration curve

Sensitivity analysis

The 90-day HAMD-24 scores were right-skewed (Suppl. Fig S1, Suppl. Table S3), with most participants having scores of 0–5. Sensitivity analyses using HAMD-24 > 10 and > 14 showed stable model performance: AUCs ranged from 0.852 to 0.912 across training and validation cohorts (Suppl. Fig S2, Suppl. Table S4), confirming the robustness of the TC/HDL ratio’s predictive value for PSD. The model including MoCA showed better discrimination (AUC training: 0.919 vs. 0.886; validation: 0.889 vs. 0.821), accuracy, calibration, and net clinical benefit than the model without MoCA across training and validation cohorts (Supplementary Figure S3, Table S5). MoCA provided independent and incremental predictive information for PSD. Comparing Model A and B in the validation set: adding TC/HDL improved discrimination (ΔAUC = 0.008, DeLong P = 0.0012, Table S7), with significant continuous NRI = 0.625 (P < 0.05) and IDI = 0.071 (P = 0.00132, Table S8); categorical NRI was not significant (lack of PSD risk thresholds). DCA showed Model A had comparable or better net benefit (Figure S5).

Discussion

This investigation assessed the association of admission TC/HDL with PSD risk. Furthermore, a nomogram prediction model integrating metabolic, inflammatory, and neurological function indicators was developed and internally validated. The key findings of this study are as follows: (1) The admission TC/HDL was significantly positively correlated with 90-day PSD risk, and it retained independent predictive value even after full adjustment for multiple confounding factors; (2) A linear relationship was observed between PSD and TC/HDL, with no obvious threshold or nonlinear effects; (3) Incorporation of TC/HDL into the nomogram model, which included the NIHSS score, MoCA score, sex, and neutrophil count as core variables, enabled good discrimination and calibration of 90-day PSD risk, indicating potential clinical applicability.

A growing body of evidence indicates a close association between the development of depression and dyslipidemia [19]. Existing evidence suggests that patients with depression often exhibit abnormal lipid profiles [20–22], but most traditional research has focused on individual lipid indices (e.g., TC or HDL), yielding highly inconsistent conclusions. Previous research indicates that reduced TC levels may be linked to higher depression risk, speculating that insufficient cholesterol may impair neuronal membrane stability and serotonin receptor function [23, 24];however, other studies have observed opposite or null results [21, 25, 26]. These discrepancies suggest that individual lipid indices may be insufficient to capture the complexity of systemic lipid metabolic homeostasis.

To overcome these limitations, researchers have recently turned their attention to the potential value of lipid ratios in psychiatric disease research. However, a previous study investigating the relationship between lipid ratios and depression reported negative findings. Specifically, a cross-sectional study from South Korea found no significant association between lipid ratios and depression [27]. Notably, this study was based on general populations and cohorts of patients with uncomplicated depression, without definite cerebrovascular structural damage. The development of depression in these populations is more likely dominated by psychosocial factors, where abnormal lipid metabolism plays a relatively limited role. In contrast, patients with AIS exhibit a pathological state characterized by cerebrovascular injury, neuroinflammatory activation, and systemic metabolic disorders. In this context, the imbalance between atherogenic burden and protective lipid reserve reflected by TC/HDL may be more relevant to post-stroke neurovascular vulnerability and impaired emotional regulation, potentially strengthening its association with PSD.

In the present study, an independent positive correlation between TC/HDL and PSD risk was clearly confirmed. After full adjustment for confounding factors—including stroke severity (NIHSS score), global cognitive function (MoCA score), inflammatory markers (neutrophil count), and coagulation activation status (D-dimer levels)—each 1-unit increase in the TC/HDL ratio was linked to a 27.26% elevated risk of PSD. Further quartile analysis revealed a 136.26% increase in PSD risk in the highest quartile versus the lowest, presenting a significant increasing trend. These results provide clinical epidemiological evidence for the involvement of dyslipidemia in PSD. It also suggests that TC/HDL is not merely a surrogate marker for disease severity or inflammatory response, but may reflect a broader metabolic–vascular milieu rather than serving solely as a surrogate for stroke severity or inflammation.

RCS analysis further revealed a significant linear positive correlation, with no evidence of a nonlinear association. This finding holds important clinical implications, as it suggests that the impact of the TC/HDL ratio on PSD risk does not exhibit an obvious “threshold effect”—instead, the risk increases continuously with rising TC/HDL ratios. Additionally, extensive subgroup analyses and interaction tests demonstrated that the association between TC/HDL and PSD remained consistent across patients stratified by sex, alcohol consumption, smoking status, HTN, and DM, with no significant interactions detected. This strongly confirms the generalizability and robustness of TC/HDL as a predictive marker for PSD, whose predictive value is not influenced by these common baseline characteristics.

Another core contribution is the development and internal validation of a PSD risk prediction nomogram integrated with the TC/HDL ratio. The model incorporates five readily accessible clinical indicators: the NIHSS score (reflecting stroke severity), MoCA score (assessing cognitive function), sex, neutrophil count (representing systemic inflammatory status), and the TC/HDL ratio (comprehensively reflecting lipid metabolic status). With an AUC of 0.889, this model exhibited excellent discriminative ability. Calibration curves showed a high degree of agreement between predicted and observed probabilities, and DCA confirmed the model could provide significant clinical net benefit. Although statistically significant, the absolute improvement in AUC was modest, indicating that TC/HDL offers limited incremental enhancement in predictive performance and should be interpreted as a complementary rather than primary determinant of model discrimination. Given that TC/HDL is routinely available and inexpensive in standard clinical practice, even modest incremental predictive value may still carry practical relevance in individualized risk stratification. Compared with models relying on imaging or complex indicators, the prediction tool developed in this study is more convenient, cost-effective, and applicable in the early post-admission period. Clinicians can use routine blood test results obtained during early admission to rapidly and intuitively stratify patients’ PSD risk via the nomogram. As an observational prediction model, the nomogram serves as a risk-stratification tool rather than a guide to specific therapeutic or lipid-modifying interventions; its findings should not be interpreted as supporting causal or treatment implications.

Finally, potential biological mechanisms may help contextualize the observed association between PSD and TC/HDL, although these pathways were not directly examined in the present study. An increased TC/HDL ratio reflects relative HDL deficiency and atherogenic lipid predominance, which have been linked in prior studies to oxidative stress, neuroinflammation, endothelial dysfunction, and neuroendocrine imbalance. HDL exhibits well-documented antioxidant and anti-inflammatory properties, capable of inhibiting reactive oxygen species production and alleviating lipid peroxidation damage [12, 28, 29]. Conversely, increased oxidative stress, a common feature in patients with depression, is recognized as one of its key pathophysiological mechanisms [30, 31]. Meanwhile, TC is critical for maintaining membrane fluidity and the function of serotonin receptors and transporters in neuronal membranes. However, its abnormal elevation may disrupt this delicate balance and impair synaptic plasticity [32]; Furthermore, excess cholesterol can promote the formation of cholesterol oxidation products (oxysterols), which can cross the blood-brain barrier and exert distinct neurotoxic and pro-inflammatory effects, further exacerbating intracerebral oxidative stress [33]. At the vascular level, previous studies have demonstrated TC/HDL is associated with endothelial dysfunction and subclinical atherosclerosis. Its elevation may lead to impaired cerebral microcirculation and chronic hypoperfusion, thereby compromising the functional integrity of key brain regions involved in emotional regulation [34, 35]. In addition, patients with depression often exhibit hyperactivation of the hypothalamic-pituitary-adrenal axis, and dysfunction of this axis can further promote cholesterol elevation, forming a vicious cycle between metabolic abnormalities and mood disorder [36]. Against the backdrop of stroke—a severe stressor—the imbalance in metabolic-vascular-neuroendocrine crosstalk reflected by an elevated TC/HDL ratio may be further amplified, thereby increasing the risk of PSD. Collectively, these mechanisms, derived from previous experimental and observational studies, provide plausible biological context for present findings but should be interpreted cautiously.

Strengths and limitations

This study has several strengths. First, it is among the first to investigate the association between the TC/HDL ratio and post-stroke depressive symptoms in patients with acute ischemic stroke. Second, restricted cubic spline analysis and nomogram modeling were applied to assess the dose–response relationship and individualized risk. Third, TC/HDL is derived from routinely available laboratory parameters, supporting the potential clinical applicability of the findings.

Certain limitations warrant consideration. First, given its single-center retrospective design, although internal validation was performed, the external generalizability still requires further verification through multicenter, prospective large-sample cohorts. The model should therefore be regarded as an internally validated risk stratification tool that requires independent external validation prior to clinical implementation. Second, the study only assessed the baseline TC/HDL ratio at admission and failed to dynamically monitor changes in this ratio during the acute and recovery phases and their association with PSD risk. Third, detailed data on baseline lipid-lowering medications, antidepressants, and specific vascular risk-modifying therapies were unavailable, precluding adjustment for these important potential confounders. Fourth, although known confounding factors were adjusted for via statistical methods, residual confounding, reverse causation, and unmeasured confounding (including acute-phase responses, nutritional status, and comorbidity burden) cannot be completely excluded, and present findings reflect observational associations rather than causal relationships. Fifth, potential selection bias existed: excluding patients with severe aphasia/consciousness disturbance may have missed severe stroke cases, possibly underestimating PSD incidence. Furthermore, as the most severely affected patients were excluded, the generalizability of the nomogram to this subgroup remains uncertain.

Conclusion

This study demonstrates admission TC/HDL is independently associated with PSD risk, as assessed using a screening-based definition, with a linear dose–response relationship. Although TC/HDL provided incremental predictive value, the magnitude of improvement was modest.The internally developed nomogram showed good discrimination and calibration; however, its findings should be interpreted cautiously given the lack of external validation, and its clinical applicability remains to be established. Given that TC/HDL is a simple, widely available and low-cost biomarker, these findings may support early risk stratification and preventive psychological care in routine stroke management, particularly in resource-limited clinical settings. These results provide additional evidence linking lipid metabolism to post-stroke mood disturbances, but should be considered exploratory. Future studies with external validation and standardized diagnostic criteria are warranted to confirm these findings.

Authors’ contributions

Xiangqi Kong was responsible for the primary data analysis and drafting of the original manuscript. Haobo Wang, Xinyue Yuan, Mina Zhao, and Wei Jing contributed to the study design and data collection. Wei Jing supervised the study and critically reviewed the manuscript. All authors read and approved the final version of the manuscript.

Funding

This work was supported by Basic Research Programs of Science and Technology Commission Foundation of Shanxi Province (grant number 202503021211244), Cultivation Project for National Natural Science Foundation at Shanxi Bethune Hospital (2024GZRZ01) and Shanxi Medical University Education and Teaching Research Project (XJ2024025).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

The study protocol was approved by the Ethics Committee of Shanxi Bethune Hospital (ethics approval number: YXLL-2026-031). Written informed consent was not required for this study.

Competing interests

The authors declare no competing interests.