Predictive factors for post-cerebral infarction depression: A retrospective study of 156 patients

Abstract

BACKGROUND

Post-stroke depression (PSD) is one of the most common neuropsychiatric complications after cerebral infarction, and reported to affect 25%-55% of stroke survivors. PSD profoundly affects rehabilitation, increases mortality and limits quality of life. Notwithstanding its clinical significance, prediction and identification of PSD in advance is a challenge as the pathogenesis of the disease is multifactorial. This study aimed to identify clinical, neuroimaging and biochemical predictive factors for PSD, and a comprehensive predictive model for the early screening of clinical patients was established.

AIM

To identify clinical, neuroimaging, and biochemical predictors of PSD and to develop a comprehensive predictive model for its early screening in patients with acute cerebral infarction.

METHODS

This was a single-center, retrospective cohort study that included 156 patients who were admitted to Department of Neurology with acute cerebral infarction between March 2021 and June 2024. At 2 weeks and 3 months after stroke, depression was evaluated using the 17-item Hamilton Depression Rating Scale and the Patient Health Questionnaire-9. According to 17-item Hamilton Depression Rating Scale scores at 3 months (≥ 7 PSD group; < 7 non-PSD group), patients were classified into PSD (n = 68, 43.6%) and non-PSD (n = 88, 56.4%) groups. Extensive data were collected: Demographic and comorbidities, stroke-related characteristics {including neuroimaging features (infarct volume, location, white matter hyperintensity, brain atrophy) and serum biomarkers [cytokines/inflammatory markers, brain-derived neurotrophic factor (BDNF)], cortisol/homocysteine}. Independent predictors were derived from multivariate logistic regression and a combined predictive model was established and validated through receiver operating characteristic analysis.

RESULTS

A total of 68 of the 156 patients (43.6%) developed PSD. Eight independent predictors were revealed by multivariate logistic regression: Psychiatric history [odds ratio (OR) = 3.85, 95% confidence interval (CI): 2.15-6.89, P < 0.001], infarction in frontal lobe (OR = 3.21, 95%CI: 1.88-5.48, P < 0.001), serum BDNF < 15.8 ng/mL (OR = 2.97, 95%CI: 1.72-5.13, P < 0.001), National Institutes of Health Stroke Scale ≥ 8 (OR = 2.68, 95%CI: 1.58-4.54, P < 0001), female gender(OR = 2.45, 95%CI: 2.52-3.73, P = 0.001), low social support (OR = 2.31, 95%CI: 1.38-3.87, P = 0.002), elevated interleukin-6 > 6.5 pg/mL (OR = 2.18, 95%CI: 1.25-3.81, P = 0.006), and left hemisphere lesion (OR = 1.95, 95%CI: 1.12-3.39, P = 0.018). The area under the curve of the multi-model was 0.89, (95%CI: 0.84-0.94), with a sensitivity of 82.4%, and specificity of 85.2%. The highest individual diagnostic value (area under the curve = 0.83) was held by serum BDNF alone. The PSD cluster had markedly increased interleukin-6 (8.6 ± 2.1 pg/mL vs 4.2 ± 1.3 pg/mL; P < 0.001), tumor necrosis factor-α (16.8 ± 3.5 pg/mL vs 9.5 ± 2.2 pg/mL; P < 0.001) and morning cortisol levels (395.2 ± 98.7 nmol/L vs 318.4 ± 722.3 nmol/L; P < 0.001), alongside lower overall BDNF concentrations (12.5 ± 36 nmol/L vs 22.7 ± 56 ng/mL; P < 0.001).

CONCLUSION

It is a multifactorial neuropsychiatric disorder with predictors identifiable across demographic, neuroanatomical, inflammatory and neurotrophic domains. The proposed predictive model based on both clinical features and serum biomarkers is sufficiently sensitive for diagnosis of early PSD. BDNF may be a particularly good biomarker to help with this. This study lays the scientific foundation for focusing early intervention plans on patients at high risk of cerebral infarction.

Key Words: Cerebral infarction; Post-stroke depression; Predictive factors; Brain-derived neurotrophic factor; Neuroinflammation; Risk model

Core Tip: Post-stroke depression (PSD) is a common and disabling complication after cerebral infarction, but early prediction of PSD remains challenging. In this cohort of 156 patients, we demonstrated clinical, neuroimaging and biological predictors most notably low serum brain-derived neurotrophic factor and high interleukin-6 levels which independently predict PSD at three months. A combination predictive model showed high diagnostic accuracy. These findings support early multidimensional risk stratification and indicate that easily measurable biomarkers can aid the screening of PSD and improve targeted early treatment in high-risk stroke survivors.

INTRODUCTION

Graeco-Roman 70%-80% of all stroke cases are cerebral infarction, the most common form of stroke, which is still one of the leading causes of long-term disability in the world. In recent two decades, survival has improved dramatically with advances in thrombolytic therapy, endovascular interventions and structured rehab programs. However, the management of neuropsychiatric complications post-stroke remains an enormous challenge for comprehensive stroke care, in which post-stroke depression (PSD) has become the most prevalent and functionally relevant psychiatric decompensation[1-5].

Epidemiological studies have consistently shown that PSD is present in 25%-55% of stroke survivors within the first year after a cerebral vascular event, with the highest prevalence reported during the first three months[6-8]. PSD has consequences much more profound than emotional distress; it is an independent predictor of worse functional recovery, lower rehabilitation adherence, higher stroke recurrence rates, longer hospital stays and healthcare costs as well as increased mortality[9-11]. Despite these well-established negative outcomes, PSD continues to be underdiagnosed and undertreated in everyday clinical practice, with estimates suggesting that only 20%-40% of cases are adequately treated[12].

The neurobiological basis of PSD is complex and presumably involves diverse neurobiological, neuroanatomical, and psychosocial mechanisms. On the neurobiological level, ischemic insults of monoaminergic pathways (serotonergic and noradrenergic) projecting from brainstem nuclei to frontal cortical and limbic structures affects the neuromodulatory balance necessary for mood maintenance[13-16]. Cerebral ischemia activates a neuroinflammatory cascade characterized by elevated levels of pro-inflammatory cytokines, including interleukin (IL)-6 and tumor necrosis factor (TNF)-α[17-19], exacerbating neurotransmitter dysregulation and neuronal mortality in circuits regulating mood. Additionally, increased serum cortisol levels along with disruptions of the diurnal rhythm of serum cortisol concentration which clearly indicates dysregulation in hypothalamic-pituitary-adrenal (HPA) axis dysfunction have been consistently observed in PSD patients[20,21].

Brain-derived neurotrophic factor (BDNF) is a prevalent neurotrophins, which plays a key role in the maintenance of synaptic plasticity, neuronal survival and hippocampal neurogenesis, has been implicated as one of the most promising biomarkers for PSD[22,23]. The relationship between low BDNF, which is reported in primary depression and PSD, mediated impaired neuroplasticity may also leads stroke survivors to depressive recurrences[22-25]. However, little is known about a direct comparison and combination of early PSD-relevance biomarkers like BDNF.

While previous studies have determined several individual risk factors for PSD, such as female gender, stroke severity, lesion location, past psychiatric history and low social support[26-29], most research has been directed at isolated predictors without developing integrated predictive models that merge clinical data with neuroimaging and biochemical databases. The absence of these types of holistic models impedes the early identification of high-risk patients, as well as their targeted preventive interventions.

This study was designed to: (1) Systematically evaluate demographic, clinical, neuroimaging, and biochemical predictive factors for PSD in one cohort of 156 patients with cerebral infarction; (2) Identify independent predictors using multivariate analysis; (3) Evaluate the diagnostic performance of serum biomarkers in particular to BDNF; and (4) Then finally build and validate a combined predictive model for early screening of PSD. The results aim to provide an evidence-based groundwork for risk stratification and preventive early intervention in clinical stroke care.

MATERIALS AND METHODS

Study design and participants

We performed a retrospective cohort study to evaluate clinical data of consecutive patients with acute cerebral infarction who were hospitalized at the Department of Neurology in our hospital from March 2021 to June 2024. The study protocol was approved by the Institutional Ethics Committee of Changde Hospital, Xiangya School of Medicine, Central South University (The First People’s Hospital of Changde City), Approval No. 2025-249-01. The requirement for written informed consent was waived by the Institutional Ethics Committee, due to the retrospective design of the study and the use of anonymized clinical data.

Inclusion criteria were: (1) Age ≥ 18 years; (2) First-ever or recurrent ischemic stroke confirmed by cranial computed tomography/magnetic resonance imaging (MRI) within 72 hours of admission; (3) Time from symptom onset to admission ≤ 7 days; (4) Could cooperate with neuropsychological assessment at the follow-up after 2 weeks of index stroke; and (5) Availability of complete clinical, imaging, and laboratory data.

Exclusion criteria: (1) Hemorrhagic stroke or transient ischemic attack; (2) Severe aphasia (Boston Diagnostic Aphasia Examination score ≤ 2), or consciousness disturbance - rendering reliable assessment impossible; and (3) Pre-existing major depressive disorder, bipolar disorder, use of any psychotic medication requiring continuous administration.

Initially, 203 individuals were screened. Based on exclusion criteria 47 subjects were excluded (including severe aphasia n = 12; preexisting psychiatric disorders, n = 9; Mini-Mental State Examination < 18, n = 8; incomplete data collection, n = 7; concurrent severe illness and/or psychotropic medication in the last months and/or substance abuse, n = 6; and others conditions, n = 5) leading to a final cohort of 156 patients. Data imputation was not performed as data completeness was an explicit inclusion criterion; no variable in the final analytic cohort had missing values (among 156 included patients).

Assessment and grouping

Standardized neuropsychological assessment in all patients was performed at 2 weeks (± 3 days) and 3 months (± 7 days) after the event. The main outcome measure was the 17-item Hamilton Depression Rating Scale (HAMD-17) administered by trained psychiatrists blinded to patients’ clinical data. The HAMD-17 assesses degree of severity for depressive symptoms such as depressed mood, guilt, suicidal ideation and insomnia, psychomotor agitation or retardation anxiety, somatic symptoms and weight loss with a total score ranging from 0 to 52. Furthermore, Patient Health Questionnaire-9 was used as an additional self-report measure.

According to 3-month HAMD-17 scores, patients were assigned to the PSD group (HAMD-17 ≥ 7; n = 68) and the non-PSD group (HAMD-17 < 7; n = 88). Mild depression (HAMD-17: 7-16; n = 31, 45.6%), moderate depression (HAMD-17: 17-23; n = 25, 36.8%); severe depression (HAMD-17 ≥ 24; n = 12, 17.6%) within the PSD group. This cutoff was selected in accordance with its established use across large-scale PSD studies[5,9]. We acknowledge that somatic symptoms shared with stroke sequelae (e.g., sleep disturbance, fatigue, appetite changes) may inflate HAMD-17 scores; alternative approaches such as somatic-exclusive subscales or complementary instruments should be considered in future research.

Data collection

Socio-demographics, medical history and lifestyle factors documented during admission were age, gender, body mass index, level of education completed, marital status of patients, employment status at time of diagnosis (employed/unemployed), residential condition (rural/urban), and medical comorbidities including hypertension, type 2 diabetes mellitus, hyperlipidemia coronary heart disease and atrial fibrillation or personal history of stroke was extracted from electronic records. A family history of psychiatric disorders and previous history of a psychiatric disorder in the patient was evaluated; also, corrections used as behavioral measures cigarette smoking during these years alcohol intake levels: Social, occasionally or regularly) with physical activity level assessed via International Physical Activity Questionnaire. Social support was measured by the Social Support Rating Scale, defined as low social support when scores < 35. The modified Rankin Scale and Barthel Index scores were assessed at 2 weeks and 3 months after stroke to evaluate functional status.

The National Institutes of Health Stroke Scale (NIHSS) was assessed both at admission and 2 weeks. Stroke etiology was defined by the classification of the Trial of Org 10172 in Acute Stroke Treatment. Stroke details including stroke location, laterality and whether or not the frontal lobe vs temporal lobe vs parietal lobe vs basal ganglia vs thalamus vs internal capsule vs brainstem vs cerebellum was involved were made by senior neuroradiologists dependent on imaging findings.

Neuroimaging assessment

All patient underwent brain MRI (1.5T or 3.0T) within 72 hours of admission including diffusion-weighted imaging, fluid-attenuated inversion recovery, T1-weighted, T2-weighted and susceptibility-weighted imaging sequences and at two weeks post-stroke. Methods in diffusion-weighted imaging sequences, infarct volume was measured through semi-automated segmentation. White matter hyperintensities were graded according to the Fazekas scale (distinguishing periventricular and deep white matter, scoring between 0-3). Global cortical atrophy scale and medial temporal atrophy visual rating scale were used to evaluate brain atrophy. On susceptibility-weighted imaging, cerebral microbleeds were detected. Specific attention on the dorsolateral prefrontal cortex, anterior cingulate cortex and basal ganglia-thalamic functions.

Laboratory biomarkers

Venous blood samples were taken at just before 24 hours overnight fast (between 07:00 and 08:00) of two weeks after stroke. Serum was separated at 30 minutes and frozen at -80 °C until batch analysis. Biomarkers We assessed inflammatory markers including IL-6, TNF-α, high sensitivity C-reactive protein, and monocyte chemoattractant protein-1 by enzyme-linked immunosorbent assay; neurotrophic factors encompassing BDNF and insulin like growth factor-1 via enzyme-linked immunosorbent assay; neuroendocrine parameters comprised 8:00 serum cortisol, adrenocorticotropic hormone and 24-hour urinary free cortisol; metabolic markers measured included homocysteine, folic acid and vitamin B12 veterans adopted chemiluminescence immunoassay routine biochemistry listed with complete blood count, hepatic function tests (alanine transaminase/aspartate transaminase) and renal function tests (blood urea nitrogen/creatinine), lipid panel (triglycerides, total cholesterol), fasting glucose levels glycated hemoglobin.

Statistical analysis

Statistical analyses were performed with R 4.2.1 software package for Windows. Normally distributed continuous variables were presented as mean ± SD and compared with independent-samples t-test; while nonnormally distributed variables were expressed as median (interquartile range), which was analyzed using the Mann-Whitney U test. Categorical variables were expressed as n (%) and compared by χ² test or Fisher’s exact test. Variables with P < 0.10 from univariate analysis entered into multivariate logistic regression via forward stepwise selection. Results were given as odds ratios (ORs) with 95% confidence intervals (CIs). Using the model, we measured receiver operating characteristic curves to assess the diagnostic performance of each biomarker and also of the combined predictive model; area under the curve (AUC), sensitivity and specificity were defined with optimal cutoff values by the Youden index. The Hosmer-Lemeshow goodness-of-fit test was used to evaluate the calibration of the model. All tests were two-sided; a P-value less than 0.05 was considered statistically significant.

RESULTS

Demographic and clinical characteristics

Out of 156 enrolled patients, 68 (43.6%) were classified as the PSD group and the other 88 (56.4%), as non-PSD according to 3-month follow-up data. Table 1 presents the demographic and clinical characteristics of both groups. Community patients in the PSD group included a higher proportion of females compared to the control group (57.4% vs 37.5%, P = 0.012), had more prior psychiatric diagnosis (19.1% vs 5.7%, P = 0.007) and less social support score (30.8 ± 7.5 vs 39.2 ± 8.8, P < 0.001). There were no significant differences in age, body mass index, education level, marital status, residential setting or major comorbidities between the two groups.

Table 1 Demographic and clinical characteristics of study participants, mean ± SD/n (%).

Characteristic	PSD group (n = 68)	Non-PSD group (n = 88)	P value
Age (years)	63.8 ± 11.5	65.7 ± 12.3	0.328
Female	39 (57.4)	33 (37.5)	0.012
BMI (kg/m2)	24.6 ± 3.2	24.1 ± 3.5	0.371
Education level			0.358
Primary or below	30 (44.1)	33 (37.5)
Secondary	26 (38.2)	36 (40.9)
College or above	12 (17.6)	19 (21.6)
Married	55 (80.9)	76 (86.4)	0.348
Urban residence	38 (55.9)	50 (56.8)	0.905
Employed	28 (41.2)	42 (47.7)	0.408
Comorbidities
Hypertension	46 (67.6)	54 (61.4)	0.414
Diabetes mellitus	20 (29.4)	22 (25.0)	0.539
Hyperlipidemia	22 (32.4)	25 (28.4)	0.588
Coronary heart disease	12 (17.6)	13 (14.8)	0.626
Atrial fibrillation	9 (13.2)	10 (11.4)	0.719
Previous stroke	11 (16.2)	9 (10.2)	0.265
Psychiatric history	13 (19.1)	5 (5.7)	0.007
Family psychiatric history	8 (11.8)	4 (4.5)	0.085
Smoking	20 (29.4)	24 (27.3)	0.766
Alcohol use	14 (20.6)	15 (17.0)	0.570
SSRS score	30.8 ± 7.5	39.2 ± 8.8	< 0.001

BMI: Body mass index; SSRS: Social Support Rating Scale; PSD: Post-stroke depression.

Stroke-related clinical features

Clinical indicators related to stroke are outlined in Table 2. The level of neurological deficits for PSD group was at admission with higher NIHSS score than that of non-PSD group (8.5 ± 2.6 vs 4.8 ± 1.9, P < 0.001). Infarcts that affected the frontal lobe (33.8% vs 15.9%, P = 0.008) and limbic system structures such as anterior cingulate cortex and basal ganglia-thalamus region (30.9% vs 13.6%, P = 0.005) were significantly more common in the PSD group Left hemisphere involvement was more common among PSD patients (63.2% vs 44.3%, P = 0.017). Functional assessments performed at 2 weeks showed significantly worse modified Rankin Scale (3.2 ± 0.9 vs 1.9 ± 0.8, P < 0.001) and Barthel Index measurements (62.5 ± 14.8 vs 81.2 ± 11.3, P < 0.001) in the PSD group compared with others during these time periods PSD group also had longer duration of onset-admission (4.8 ± 1.5 hours vs 3.2 ± 1.0 hours, P = 0.008).

Table 2 Stroke-related clinical features, mean ± SD/n (%).

Stroke feature	PSD group (n = 68)	Non-PSD group (n = 88)	P value
NIHSS at admission	8.5 ± 2.6	4.8 ± 1.9	< 0.001
NIHSS at 2 weeks	5.8 ± 2.1	3.2 ± 1.5	< 0.001
Infarct location
Frontal lobe	23 (33.8)	14 (15.9)	0.008
Temporal lobe	12 (17.6)	18 (20.5)	0.656
Parietal lobe	10 (14.7)	15 (17.0)	0.687
Basal ganglia	18 (26.5)	20 (22.7)	0.583
Thalamus	8 (11.8)	7 (8.0)	0.420
Limbic system structures	21 (30.9)	12 (13.6)	0.005
Internal capsule	15 (22.1)	16 (18.2)	0.544
Brainstem	5 (7.4)	8 (9.1)	0.700
Cerebellum	3 (4.4)	5 (5.7)	0.723
Left hemisphere	43 (63.2)	39 (44.3)	0.017
TOAST classification			0.412
Large-artery atherosclerosis	25 (36.8)	28 (31.8)
Cardioembolism	12 (17.6)	14 (15.9)
Small-vessel occlusion	22 (32.4)	35 (39.8)
Other/undetermined	9 (13.2)	11 (12.5)
Onset to admission (hours)	4.8 ± 1.5	3.2 ± 1.0	0.008
SBP at admission (mmHg)	162.3 ± 21.5	151.8 ± 16.2	0.038
Glucose at admission (mmol/L)	7.8 ± 1.6	6.9 ± 1.2	0.025
mRS at 2 weeks	3.2 ± 0.9	1.9 ± 0.8	< 0.001
Barthel Index at 2 weeks	62.5 ± 14.8	81.2 ± 11.3	< 0.001
mRS at 3 months	2.5 ± 1.0	1.3 ± 0.7	< 0.001
Barthel Index at 3 months	72.8 ± 12.5	88.6 ± 9.2	< 0.001

NIHSS: National Institutes of Health Stroke Scale; TOAST: Trial of Org 10172 in Acute Stroke Treatment; SBP: Systolic blood pressure; mRS: Modified Rankin Scale; PSD: Post-stroke depression.

Neuroimaging findings

Neuroimaging group comparisons are shown in Table 3. The infarct volume was significantly larger in the PSD group (9.2 ± 3.1 cm³ vs 5.1 ± 2.0 cm³, P < 0.001). Compared to non-PSD participants, PSD patients had worse Fazekas scores periventricular (2.1 ± 0.6 vs 1.2 ± 0.5, P < 0.001) and deeper white matter lesions (1.9 ± 0.5 vs 1.0 ± 0.4, P < 00001), indicating more severe white matter hyperintensities on MRI. A greater extent of frontal (global cortical atrophy frontal: 1.8 ± 0.6 vs 1.1 ± 0.5, P < 0.001) and medial temporal atrophy (medial temporal atrophy: 1.6 ± 0.7 vs 1.0 ± 0.5, P = 0.003) was responsible for the worse brain atrophy indices for the PSD group compared to controls. A higher number of cerebral microbleeds (mean count 2.8 ± 1.9 vs 1.5 ± 1.2, P = 0.009) was seen in the PSD group compared to no-PSD patients. In conclusion, these findings support the involvement of both acute lesion pathology and pre-existing chronic cerebrovascular pathology in susceptibility to PSD.

Table 3 Neuroimaging findings, mean ± SD/n (%).

Imaging feature	PSD group (n = 68)	Non-PSD group (n = 88)	P value
Infarct volume (cm3)	9.2 ± 3.1	5.1 ± 2.0	< 0.001
Multiple infarcts	28 (41.2)	25 (28.4)	0.089
White matter hyperintensities
Fazekas periventricular	2.1 ± 0.6	1.2 ± 0.5	< 0.001
Fazekas deep WM	1.9 ± 0.5	1.0 ± 0.4	< 0.001
Fazekas total score	4.0 ± 1.0	2.2 ± 0.8	< 0.001
Brain atrophy
GCA frontal	1.8 ± 0.6	1.1 ± 0.5	< 0.001
GCA parietal	1.3 ± 0.5	1.0 ± 0.4	0.065
MTA score	1.6 ± 0.7	1.0 ± 0.5	0.003
Global atrophy (GCA total)	1.5 ± 0.5	1.1 ± 0.4	0.012
Microbleed count	2.8 ± 1.9	1.5 ± 1.2	0.009
Perfusion parameters
ADC (× 10-3 mm2/second)	0.80 ± 0.11	0.90 ± 0.12	0.032
TTP (seconds)	122 ± 12	108 ± 9	0.018
CBF (mL/100 g/minutes)	32.5 ± 6.8	38.2 ± 7.5	0.028

WM: White matter; GCA: Global cortical atrophy; MTA: Medial temporal atrophy; ADC: Apparent diffusion coefficient; TTP: Time to peak; CBF: Cerebral blood flow; PSD: Post-stroke depression.

Laboratory biomarkers

Serum biomarker profiles differed significantly between the PSD vs non-PSD groups (Table 4). The study involved data recorded from 105 patients dialyzed in a large Guatemala Hospital, aged between 18 years and 92 years; the cohort consisted of patients with a diagnosis of PSD or those without (non-PSD), hereafter referred to as neurotic state. Based on the assessment of neuroendocrine changes, people with suffering had elevated morning levels of cortisol (395.2 ± 98.7 nmol/L vs 318.4 ± 72.3 nmol/L, P < 0.001), heightened adrenocorticotropic hormone (42.8 ± 8.5 pg/mL vs 34.6 ± 6.8 pg/mL, P < 0.001) and a blunted diurnal rhythm of cortisol (22.5% ± 5.3% vs 36.8% ± 6.5%, P < 0.001). Compared to non-PSD participants, PSD group had also higher homocysteine (18.5 ± 5.2 μmol/L vs 13.8 ± 4.1 μmol/L, P < 0.001) and lower folic acid and vitamin B12 levels per caput.

Table 4 Laboratory biomarkers in post-stroke depression and post-stroke depression groups, mean ± SD/n (%).

Biomarker	PSD group (n = 68)	Non-PSD group (n = 88)	P value
Inflammatory markers
IL-6 (pg/mL)	8.6 ± 2.1	4.2 ± 1.3	< 0.001
TNF-α (pg/mL)	16.8 ± 3.5	9.5 ± 2.2	< 0.001
hs-CRP (mg/L)	13.2 ± 4.1	7.8 ± 2.5	< 0.001
MCP-1 (pg/mL)	468 ± 72	372 ± 55	< 0.001
Neurotrophic factors
BDNF (ng/mL)	12.5 ± 3.8	22.7 ± 5.6	< 0.001
IGF-1 (ng/mL)	85.3 ± 20.5	118.6 ± 25.8	< 0.001
Neuroendocrine markers
Morning cortisol (nmol/L)	395.2 ± 98.7	318.4 ± 72.3	< 0.001
ACTH (pg/mL)	42.8 ± 8.5	34.6 ± 6.8	< 0.001
24 hours urinary cortisol (μg/24 hours)	125.3 ± 32.5	98.6 ± 24.8	< 0.001
Diurnal cortisol variation (%)	22.5 ± 5.3	36.8 ± 6.5	< 0.001
Cortisol awakening response	145.8 ± 35.2	182.5 ± 42.6	0.006
Metabolic markers
Homocysteine (μmol/L)	18.5 ± 5.2	13.8 ± 4.1	< 0.001
Folic acid (ng/mL)	6.2 ± 2.1	8.5 ± 2.8	< 0.001
Vitamin B12 (pg/mL)	285 ± 75	352 ± 90	0.003
HbA1c (%)	6.8 ± 1.2	6.3 ± 1.0	0.048
LDL-C (mmol/L)	3.2 ± 0.9	2.9 ± 0.8	0.102
Albumin (g/L)	36.5 ± 4.2	39.8 ± 3.8	0.015

IL: Interleukin; TNF: Tumor necrosis factor; hs-CRP: High sensitivity C-reactive protein; MCP-1: Monocyte chemoattractant protein-1; BDNF: Brain-derived neurotrophic factor; IGF-1: Insulin like growth factor-1; ACTH: Adrenocorticotropic hormone; HbA1c: Hlycated hemoglobin; LDL-C: Low-density lipoprotein cholesterol; PSD: Post-stroke depression.

Multivariate logistic regression analysis

Univariate analysis was done and P < 0.10 variables were included in the multivariable logistic regression analysis. As a result, eight independent predictors of PSD were identified (Table 5). History of psychiatric disorders (OR = 3.85, 95%CI: 2.15-6.89, P < 0.001) was the strongest predictor followed by frontal lobe infarction (OR = 3.21, 95%CI: 1.88-5.48; P < 0.001) and serum BDNF < 15.8 ng/mL (OR = 2.97, 95%CI: 1.72-5.13; P < 0.001). Other strong predictors included NIHSS > 8, female sex, low social support, high levels of IL-6 and left hemisphere lesion. The Hosmer-Lemeshow statistic was 6.82 (P = 0.556), indicating a well-fitting model.

Table 5 Independent predictors of post-stroke depression: Multivariate logistic regression.

Variable	β	SE	Wald	OR (95%CI)	P value
Psychiatric history	1.348	0.298	20.45	3.85 (2.15-6.89)	< 0.001
Frontal lobe infarction	1.166	0.273	18.23	3.21 (1.88-5.48)	< 0.001
BDNF < 15.8 ng/mL	1.088	0.280	15.10	2.97 (1.72-5.13)	< 0.001
NIHSS ≥ 8	0.986	0.269	13.42	2.68 (1.58-4.54)	< 0.001
Female gender	0.896	0.244	13.48	2.45 (1.52-3.95)	0.001
Low social support	0.838	0.263	10.15	2.31 (1.38-3.87)	0.002
IL-6 > 6.5 pg/mL	0.779	0.285	7.48	2.18 (1.25-3.81)	0.006
Left hemisphere lesion	0.668	0.282	5.61	1.95 (1.12-3.39)	0.018
Non-significant variables
Homocysteine > 15 μmol/L	0.512	0.295	3.01	1.67 (0.94-2.97)	0.083
Morning cortisol > 360 nmol/L	0.598	0.312	3.67	1.82 (0.99-3.35)	0.055
WMH Fazekas ≥ 4	0.425	0.278	2.34	1.53 (0.89-2.63)	0.126
Diabetes mellitus	0.382	0.265	2.08	1.47 (0.87-2.47)	0.149
Previous stroke	0.356	0.312	1.30	1.43 (0.78-2.63)	0.254
Admission glucose > 7.0	0.298	0.258	1.33	1.35 (0.81-2.24)	0.248

Hosmer-Lemeshow test: χ² = 6.82, P = 0.556. BDNF: Brain-derived neurotrophic factor; NIHSS: National Institutes of Health Stroke Scale; IL: Interleukin; WMH: White matter hyperintensities; OR: Odds ratio; CI: Confidence interval.

Evaluation of the predictive model and biomarkers

The AUC of the dual predictive model by these eight independent risk factors was 0.89 (95%CI: 0.84-0.94), and sensitivity and specificity at optimal cutoff points were 82.4% and 85.2%, respectively. BDNF exhibited the best diagnostic performance among the individual biomarkers with an AUC of 0.83 (95%CI: 0.77-0.89), followed by IL-6 (AUC = 0.78), cortisol (AUC = 0.75) and NIHSS score (AUC = 0.72). Receiver operating characteristic Curve of combined model and individual predictors (Figure 1). A forest plot illustrating the ORs and 95%CIs for all variables entered into the multivariate model is shown in Figure 2. Variables with CIs entirely to the right of 1.0 represent statistically significant predictors.

Figure 1 Receiver operating characteristic curves comparing the diagnostic performance of the combined predictive model and individual biomarkers for post-stroke depression. The combined model integrating clinical, neuroimaging, and biochemical parameters achieved the highest area under the curve = 0.89, followed by brain-derived neurotrophic factor alone (area under the curve = 0.83). All curves were significantly superior to the reference line (all P < 0.001). AUC: Area under the curve; BDNF: Brain-derived neurotrophic factor; IL: Interleukin; NIHSS: National Institutes of Health Stroke Scale.

Figure 2 Forest plot of odds ratios (95% confidence interval) for predictive factors of post-stroke depression from multivariate logistic regression. Blue markers indicate statistically significant independent predictors (P < 0.05); gray markers indicate non-significant variables. The dashed vertical line represents odds ratio = 1.0 (no effect). BDNF: Brain-derived neurotrophic factor; NIHSS: National Institutes of Health Stroke Scale; SSRS: Social Support Rating Scale; IL: Interleukin; CI: Confidence interval.

Depression severity subgroup analysis

When stratifying by severity of depression, a dose-response effect emerged for several parameters (Table 6). Patients presenting with more severe depression (HAMD ≥ 24) had higher NIHSS scores (10.2 ± 2.8), larger infarct volumes (11.5 ± 3.8 cm³), lower levels of BDNF (8.2 ± 2.5 ng/mL), upper level of IL-6 (11.3 ± 2.8 pg/mL), and lower social support score on average (26.5 ± 6.2). Frontal lobe involvement was most frequent in the severe depression subgroup (50.0%), and lesions of the left hemisphere showed a graded increase from 54.8% in mild, to 75.0% in severe depression. These results are in line with a biological gradient in PSD that is dependent on severity.

Table 6 Clinical and biomarker characteristics stratified by depression severity, mean ± SD/n (%).

Characteristic	Mild (n = 31)	Moderate (n = 25)	Severe (n = 12)	P value
HAMD-17 score	11.2 ± 2.8	19.5 ± 1.8	28.3 ± 3.5	< 0.001
PHQ-9 score	8.5 ± 2.2	14.8 ± 2.5	21.2 ± 3.8	< 0.001
NIHSS at admission	7.2 ± 2.0	9.0 ± 2.3	10.2 ± 2.8	0.003
Infarct volume (cm3)	7.5 ± 2.5	9.8 ± 3.0	11.5 ± 3.8	0.001
Frontal lobe infarct, %	25.8	36.0	50.0	0.018
Left hemisphere, %	54.8	68.0	75.0	0.028
mRS at 3 months	2.0 ± 0.8	2.8 ± 0.9	3.5 ± 1.1	< 0.001
Barthel Index at 3 months	78.5 ± 10.2	70.2 ± 12.5	60.5 ± 15.8	< 0.001
Biomarkers
BDNF (ng/mL)	15.2 ± 3.5	11.0 ± 3.2	8.2 ± 2.5	< 0.001
IL-6 (pg/mL)	7.0 ± 1.8	9.2 ± 2.0	11.3 ± 2.8	< 0.001
TNF-α (pg/mL)	14.2 ± 3.0	17.5 ± 3.2	21.8 ± 4.5	< 0.001
hs-CRP (mg/L)	10.8 ± 3.5	14.2 ± 4.0	17.5 ± 5.2	< 0.001
Morning cortisol (nmol/L)	365 ± 85	405 ± 95	445 ± 110	0.003
Homocysteine (μmol/L)	16.5 ± 4.5	19.2 ± 5.0	22.8 ± 6.5	0.002
SSRS score	33.5 ± 6.8	29.2 ± 7.5	26.5 ± 6.2	0.005
Fazekas total score	3.5 ± 0.9	4.2 ± 1.0	4.8 ± 1.2	0.006

P values were calculated using the Jonckheere-Terpstra test for ordered alternatives. HAMD-17: 17-item Hamilton Depression Rating Scale; PHQ-9: Patient Health Questionnaire-9; NIHSS: National Institutes of Health Stroke Scale; mRS: Modified Rankin Scale; BDNF: Brain-derived neurotrophic factor; IL: Interleukin; TNF: Tumor necrosis factor; hs-CRP: High sensitivity C-reactive protein; SSRS: Social Support Rating Scale.

DISCUSSION

In total 156 cerebral infarction patients were investigated to systematically assess demographic, clinical, neuroimaging and biochemical predictors for PSD, leading to establishment of a combined predictive model with high diagnostic accuracy. The results add to the current knowledge of PSD pathogenesis and constitute a practical framework for early identification and risk stratification.

The previously established meta-analytic PS incidence trajectory indicates that approximate 25%-55% of cerebral infarction survivors develop clinically significant depressive symptoms[6-8], with the observed PSD incidence (43.6%) at 3 months confirming this finding. Altogether, the eight independent predictors identified in this study across multiple pathophysiological domains highlights the multifactorial nature of PSD and suggests that comprehensive assessment approaches are necessary rather than reliance on a single risk factor.

Of the factors we identified as predictors, prior psychiatric history showed the strongest association (OR = 3.85), which is consistent with robust evidence from longitudinal studies of stroke[8,24]. A priori vulnerability in mood regulatory neural circuits may lower the threshold for eliciting a major depressive episode when challenged by the somatic and psychosocial stressors of stroke. Clinically, this finding highlights the need for comprehensive psychiatric history-taking at the time of stroke admission, which is often not done in many acute stroke units.

The independent contribution of frontal lobe infarction (OR = 3.21) is consistent with the previously established role of the prefrontal cortex in executive function, emotion regulation, and reward processing. Damage to dorsolateral prefrontal connections, particularly with respect to their neuronal gating of the anterior cingulate cortex and subcortical structures, upends neural networks crucial for mood homeostasis[13-15]. The additional discovery of limbic system involvement in 30.9% of PSD cases further implicates disruption of the cortico-limbic circuit as a key mechanism underlying PSD[8,9,30].

This study is, most importantly, the first to determine serum BDNF as the strongest independent predictor (OR = 2.97) and single biomarker with highest discriminative power for PSD (AUC = 0.83). BDNF is essential for synaptic plasticity, neuronal survival and both progenitor cell differentiation and monoaminergic neurotransmitter signaling in the hippocampus[22-25]. The much lower BDNF levels in PSD as compared with controls (12.5 ± 3.8 ng/mL vs 22.7 ± 5.6 ng/mL), and a severity-dependent gradient (from 15.2 ng/mL for mild to 8.2 ng/mL for severe depression), show that reduced neurotrophic support is an urgent biological vulnerability candidate for PSD[9]. These results are hypothesis-generating: Because this is observational data, we cannot imply causation and the potentially therapeutic effects for BDNF-enhancing interventions such as exercise, selective serotonin reuptake inhibitors and repetitive transcranial magnetic stimulation need to be prospectively evaluated[31-34].

In PSD patients, significantly higher IL-6 and TNF-α status levels also support the neuroinflammation hypothesis of PSD. A durable enhancement of pro-inflammatory cytokines due to a stimulation of the kynurenine pathway in conjunction with direct effects on monoamine synthesis and reuptake[17-19,35] after cerebral ischemia drives a potent inflammatory cascade that extends into later stages beyond the acute setting.

That IL-6 > 6.5 pg/mL predicted PSD independent (OR = 2.18) indicate that there may exist a unique, we mean threshold for risk stratification with practical clinical utility in our study.

Dysregulation of the HPA axis, as manifested by hypercortisolemia in the morning and blunted diurnal variation, is another important pathophysiologic pathway. Though cortisol approached significance as an independent predictor on multivariate analysis (P = 0.055), significant univariate differences and the previously described neurotoxic effects of chronic hypercortisolism on hippocampal neurites suggest that dysfunction in the HPA axis contributes to PSD by impairing neurogenesis, reducing synaptic plasticity and potentiating neuroinflammatory mechanisms[20,21,36].

The multivariable prediction model that combined all eight independent predictors demonstrated excellent diagnostic performance (AUC = 0.89, 82.4% sensitivity, 85.2% specificity), with substantial improvement over any individual predictor. These findings show the clinical benefit of multidimensional evaluation and suggest that utilization of an organized screening protocol incorporating these factors could substantially enhance early detection of PSD. From a clinical point of view, this leads to the suggested double screening strategy: First, immediate identification of patients with some of these clinical risk factors [female gender, psychiatric history, high-grade stroke severity (Barthel > 70), frontal lobe involvement on imaging studies, left hemisphere lesion and low levels of social support] at admission that will be evaluated with biomarkers specific for those in the subset at 2 weeks after stroke onset (BDNF and IL-6).

This study has several limitations. Although sufficient for the primary analysis, the sample size limited many subgroup analyses, and the retrospective single-center design reduces generalizability. The 3-month follow-up identified early-onset PSD, but may have missed late-onset cases. Serum BDNF should be interpreted with caution as a peripheral proxy rather than a direct measure of central neurotrophic activity: Platelets and endothelial cells contribute substantially to circulating BDNF levels, and the correlation between serum and cerebrospinal fluid BDNF in stroke patients is modest; future studies should incorporate paired cerebrospinal fluid sampling. Furthermore, the exclusion of patients with severe aphasia and significant cognitive impairment, while methodologically necessary for reliable depression assessment, limits generalizability to the broader stroke population, where such patients are prevalent and arguably carry an even greater depression risk. Subsequent studies ought to comprise potential multicenter validation in more substantial cohorts, lengthier follow-up intervals, incorporation of functional neuroimaging data (resting-state functional MRI, diffusion tensor imaging), and exploration of genetic polymorphisms influencing BDNF expression and inflammatory response.

CONCLUSION

PSD following cerebral infarction is a common, multifactorial neuropsychiatric complication with clear predictive factors that have been identified from demographic, neuroanatomical, inflammatory, neurotrophic and psychosocial domains. Eight independent predictors were determined, and a combined model with high diagnostic accuracy (AUC = 0.89) was built for early screening of PSD. Among plasma biomarkers, serum BDNF was the most promising one (AUC = 0.83), and demonstrated a severity-dependent gradient consistent with its role in PSD pathogenesis. This evidence-based framework not only contributes precise and accessible evidence to the early risk stratification for cerebral infarction patients, but also helps carry out targeted intervention measures at an early stage to further enhance psychological outcome and health-related quality of life among stroke survivors.