Comprehensive psychological and neurological rehabilitation for post-stroke depression: A retrospective comparative study with biomarker exploration

Abstract

BACKGROUND

Post-stroke depression (PSD) is one of the most common and debilitating neuropsychiatric sequelae after ischemic stroke, with a prevalence of 30%-50%, which significantly hinders neurological recovery, participation in rehabilitation, and long-term quality of life. Although the clinical burden of stroke is high, integrated rehabilitation strategies that effectively combine structured psychological rehabilitation with its neurological counterpart are under-defined. The best combination of interventions and predictors of response to therapies have not yet been systematically characterized.

AIM

To assess clinical effectiveness of a psychological rehabilitation oriented comprehensive intervention program aimed for PSD patients, find out independent predictors effecting treatment response and novel serum biomarkers including neuropeptide Y (NPY) as indicators for rehabilitation outcome of PSD patients, also use validated nomogram prediction model in order to guide personalized clinical decision-making.

METHODS

A retrospective study was performed comprised of 178 patients with confirmed acute cerebral infarction and PSD, that were admitted to Departments of Neurology and Rehabilitation Medicine between September 2020 and March 2024. Patients were assigned to a comprehensive rehabilitation group (CRG) (n = 96) receiving cognitive behavioral therapy (CBT)-centered psychological rehabilitation combined with systematic neurological rehabilitation, or a control group (CG) (n = 82) receiving standard pharmacological treatment and basic physiotherapy. The primary outcome was change in Hamilton Depression Rating Scale-17 (HAMD-17) and remission rate (HAMD-17 < 7) at 12 weeks. Secondary outcomes were the Fugl-Meyer Assessment (FMA), Modified Barthel Index (MBI), Montreal Cognitive Assessment and Stroke-Specific Quality of Life Scale. Single serum biomarkers included NPY, brain-derived neurotrophic factor (BDNF), pro-inflammatory cytokines [interleukin (IL)-1β, IL-6, tumor necrosis factor-α (TNF-α) and IL-10], neuroendocrine markers and indices of oxidative stress collected at baseline and 12 weeks. Independent predictors of response to treatment were identified through multivariable logistic regression analysis and a nomogram model was built, internally validated using Bootstrap resampling, and assessed by calibration curves and decision curve analysis.

RESULTS

The CRG showed significantly lower HAMD-17 scores (7.2 ± 3.4 vs 13.8 ± 4.6; P < 0.001) and higher rates of depression remission (67.7% vs 29.3%; P < 0.001) from those in the CG at 12 weeks follow-up visit, respectively. The CRG had significantly higher FMA scores (78.5 ± 14.2 vs 58.3 ± 16.8; P < 0.001) and MBI scores (76.8 ± 13.5 vs 58.6 ± 15.2; P < 0.001). In the multivariable logistic regression analyses, seven independent predictors of therapeutic response were identified: Early initiation of psychological intervention [≤ 2 weeks; odds ratio (OR) = 4.12, 95% confidence interval (CI): 2.28-7.44, P < 0.001]; Baseline serum NPY level ≥ 82.5 pg/mL (OR = 3.38, 95%CI: 1.95-5.86, P < 0.001); Mild-to-moderate depression severity (HAMD-17 < 17; OR = 2.94, 95%CI: 1.68-5.14, P < 0.001); High family involvement (family support questionnaire ≥ 28; OR = 2.65, 95%CI: 1.58-4.44, P < 0.001); Intact cognition (Mini-Mental State Examination ≥ 24; OR = 2.42, 95%CI: 1.45-4.04, P = 0.001); Non-dominant hemisphere lesion (OR = 2.08, 95%CI: 1.22-3.55, P = 0.008); and baseline IL-1β < 8.2 pg/mL (OR = 1.86, 95%CI: 1.10-3.15, P = 0.021). The Bootstrap-validated C-index of the nomogram model was 0.85 (area under the curve = 0.87, 95%CI: 0.81-0.92), sensitivity 80.2%, specificity 83.6%. Post-intervention, CRG displayed significantly higher serum NPY levels (98.6 ± 18.4 pg/mL vs 76.2 ± 15.8 pg/mL), BDNF (22.5 ± 5.2 ng/mL vs 15.2 ± 4.8 ng/mL) and markedly lower IL-1β, IL-6 and TNF-α levels compared with CG (all P < 0.001).

CONCLUSION

CBT-based comprehensive rehabilitation program effectively improves depression and motor functions in patients with PSD. The timing of the intervention, neuroinflammatory state, cognitive reserve and family support, as well as lesion laterality shape treatment response. The serum NPY is a candidate biomarker and the Nomogram, a potential decision-support tool pending external validation before clinical use.

Key Words: Post-stroke depression; Psychological rehabilitation; Cognitive behavioral therapy; Neuropeptide Y; Neurological rehabilitation; Neuroinflammation; Nomogram prediction model

Core Tip: Ischemic stroke can lead to post-stroke depression (PSD), which is an important protective factor for neurological recovery, rehabilitation compliance and patient quality of life. In this retrospective study, we found that a cognitive behavioral therapy-centered comprehensive rehabilitation program integrating psychological with neurological rehabilitation was superior to standard care in terms of depressive symptoms, motor recovery, cognitive function and daily living ability. Baseline serum neuropeptide Y identified as a prospective treatment response biomarker. Earlier intervention, enhanced family engagement, preserved cognition, reduced inflammation and non-dominant hemisphere lesions also predicted better outcomes. An internally validated nomogram might facilitate individualized rehabilitation planning in PSD patients.

INTRODUCTION

Cerebral infarction continues to be among the leading causes of adult disability and death globally, comprising 62.4% of all strokes in the latest Global Burden of Disease analysis[1]. Recanalization therapies, including intravenous thrombolysis and mechanical thrombectomy, have made considerable progress in recent decades[2,3]. However, the neurologic impact of cerebral infarction remains a formidable challenge in survivors, caregivers and health systems over the long term. Post-stroke depression (PSD) is the most common and clinically significant neuropsychiatric complication post-stroke, with incidences of 30%-50% among ischemic stroke survivors during the first year; with peak incidence in the first three months[4,5].

PSD has clinical implications that go well beyond the domain of emotional disturbance. The existence of a substantial body of evidence showing that untreated PSD independently predicts worse functional recovery, less adherence to rehabilitation programs, faster cognitive decline, higher risk of recurrent stroke, prolonged length of stay and increased all-cause mortality have been established[6,7]. The bidirectional link between depression and neurological disability generates a vicious cycle, where unaddressed affective symptoms compromise the neuroplastic mechanisms involved in motor and cognitive restitution and sustained functional impairment deepens hopelessness or psychological distress[8]. As a result, appropriate management of PSD is more frequently considered a necessary condition for maximizing post-stroke recovery instead of additional consideration.

The neurobiology of PSD is complex and multi-faceted, with the dysregulation of monoaminergic systems, activation of neuroinflammatory pathways, hypothalamic-pituitary-adrenal (HPA) axis activation and attenuation of trophic support systems all playing a contributory role[5,9]. Release of pro-inflammatory cytokines induced by ischemia such as interleukin (IL)-1β, IL-6 and tumor necrosis factor-α (TNF-α) results in impaired serotonergic and dopaminergic neurotransmission via activation of the kynurenine pathway and direct effects on monoamine synthesis[10,11]. Meanwhile, decreased levels of brain-derived neurotrophic factor (BDNF) alter hippocampal neurogenesis and synaptic plasticity, enabling a blunted neural restore ability environment[12]. More recently, neuropeptide Y (NPY) has gained attention as a candidate biomarker that is particularly relevant to depression after neurological injury due to its widespread expression throughout the brain and involvement in stress resilience, emotional regulation, and hippocampal neuroprotection[13,14].

As for intervention, cognitive behavioral therapy (CBT) has proven the most robust evidence base of psychological treatments for PSD, addressing maladaptive cognitive schemas and behavioral withdrawal patterns as well as dysfunctional coping strategies using structured, manualized protocols[15,16]. Mindfulness-based cognitive therapy (MBCT) has additional additive effects such as promoting metacognitive insight and impairing depressive rumination[17]. Motivational enhancement therapy (MET) overcomes low intrinsic motivation for engagement in rehabilitation, which is a widespread barrier to recovery in PSD[18]. Family-centered psychoeducation harnesses the social support infrastructure that is critical to sustain therapeutic gains in community settings[19]. Then, both task-oriented motor training, occupational therapy and neurostimulation techniques at the level of neurological rehabilitation promote reorganization of the motor cortex and potentially have independent antidepressant effects by increasing NPY, BDNF and anti-inflammatory cytokines via exercise[20,21]. Despite this theoretical underpinnings for integrated biopsychosocial rehabilitation, prospective clinical research has as yet insufficiently defined the synergistic effects of combining psychological and neurological rehabilitation modalities along with the determinants of individual treatment responsiveness[22,23].

Another major gap in the PSD rehabilitation literature is prediction modeling. Current risk stratification tools have primarily concentrated on predicting depression occurrence[23], but not treatment response, and none has added new biomarkers such as NPY to clinical and neuroimaging variables by means of a validated Nomogram approach for bed-side use age[24,25]. Individualized predictive tools are crucial in improving resource allocation, guiding the intensity of intervention according to patient-specific risk profiles, and setting targets for rehabilitation monitoring.

We therefore set out to conduct the present study to achieve four main aims: (1) Compare the effectiveness of a multi-modular rehabilitation intervention combining CBT-based psychological rehabilitation and systematic neurological rehabilitation with usual care on PSD depressive symptoms and functional outcomes; (2) Generally identify independent clinical, imaging and biochemical predictors of treatment response using multivariable regression analysis; and (3) Explore the profile and predictive potential of serum NPY and associated biomarkers throughout the course of rehabilitation; and for aim 4, create an internally validated nomogram-based predictive model that allows individual estimation of probability for successful interim recovery after completion of rehabilitation, which could assist in routine clinical decisions.

MATERIALS AND METHODS

Study design and ethical approval

This retrospective study included consecutive patients admitted to the Departments of Neurology and Rehabilitation Medicine between September 2020 and March 2024. The Institutional Ethics Committee approved the protocol (No. 2026SQ461), which adhered to the Declaration of Helsinki. Patients eligible for the exercise prescription plus adequate family support were assigned to comprehensive rehabilitation group (CRG); those preferring not to undergo comprehensive rehabilitation or who were admitted during resource-poor periods of the study period including weekends and holidays[25] were assigned to control group (CG).

Inclusion and exclusion criteria

Inclusion criteria were: 18-80 years, computed tomography/magnetic resonance imaging (MRI)-diagnosed ischemic stroke, PSD at 2 weeks post-stroke [Hamilton Depression Rating Scale-17 (HAMD-17) ≥ 8 with a mini-confirmed depressive episode], Karnofsky performance status ≥ 50, expected completion of the 12-week protocol and complete clinical data. Patients were excluded if they previously had a hemorrhagic stroke or transient ischemic attack (TIA), psychiatric disorders (past or on medication), severe aphasia [Western Aphasia Battery-Aphasia Quotient (WAB-AQ) < 50] and/or decreased level of consciousness, Mini-Mental State Examination (MMSE) < 15, malignant disease or severe failure of an organ system; contra-indications to physical rehabilitation. Among 234 screened patients, 56 were excluded and a final cohort of 178 was obtained (CRG: n = 96; CG: n = 82). Reasons for exclusion were: Hemorrhagic stroke or TIA (n = 12), prior psychiatric diagnosis or psychotropic medication (n = 14), severe aphasia (WAB-AQ < 50) or impaired consciousness (n = 10), MMSE < 15 (n = 8), malignant disease or severe organ failure (n = 7), and contraindications to physical rehabilitation (n = 5). There was no loss to follow-up and no missing outcome data at 12 weeks; all 178 patients were included in the primary analysis and no imputation was required.

Comprehensive rehabilitation intervention

The program was a 12-week multidisciplinary intervention that began within 2 weeks of PSD confirmation, which included the following core components: CBT: Individual sessions twice weekly (50 minute/session, total of r24 sessions) were structured in four phases: (1) Psychoeducation; (2) Behavioral activation; (3) Cognitive restructuring; and (4) Relapse prevention. Fidelity was assessed through audio review of 20% of sessions (Cognitive Therapy Rating Scale score consistently > 40). MBCT: Group-based (5-8 participants), weekly (1/week) for 8 weeks (60 minute/session), in addition to daily home practice (+ 20 minutes). MET: Individual 5 sessions biweekly. Family psychoeducation: 1 monthly group session for caregivers (3 sessions in total). Arrival and admission for inpatient rehabilitation (daily motor practice; occupational therapy; speech-language therapy = as indicated; neuromuscular electrical stimulation = if indicated. Venlafaxine XR (75-150 mg/day) adjunctive therapy was initiated by the monitoring psychiatrist for subjects with HAMD-17 ≥ 20 and insufficient response at week 4.

CG

Patients were treated with standard stroke pharmacotherapy and basic physiotherapy at a frequency equal to the neurological rehabilitation module. There was no structured psychologic intervention.

Outcome measures

Anonymous clinicians performed assessments at baseline (2 weeks post-stroke), week 6, and week 12.

Main outcomes: HAMD-17 score difference from baseline to week 12; Depression remission rate (HAMD-17 < 7 at week 12), complemented by the Zung Self-Rating Depression Scale.

Secondary outcomes: Motor function [Fugl-Meyer Assessment (FMA)], functional independence [Modified Barthel Index (MBI)], Montreal Cognitive Assessment (MoCA), Stroke-Specific Quality of Life (SS-QOL), Fatigue Severity Scale and safety events.

Baseline data collection

Demographics, vascular risk factors, characteristics of the stroke [National Institutes of Health Stroke Scale (NIHSS) score, Trial of Org 10172 in Acute Stroke Treatment (TOAST) classification, lesion location], and psychosocial factors comprising family support (family support questionnaire) and social support (social support rating scale) were evaluated with standardized data collection.

Neuroimaging

All patients underwent 3.0T MRI within 72 hours of admission (diffusion-weighted imaging, fluid-attenuated inversion recovery, susceptibility-weighted imaging, magnetic resonance angiography, arterial spin labeling). Infarct volume, white matter hyperintensity burden (Fazekas scale), cortical atrophy (global cortical atrophy scale/medial temporal atrophy scales) and collateral circulation were evaluated independently by two senior neuroradiologists with a focus on prefrontal-striato-thalamic circuit and limbic system involvement.

Serum biomarkers

Blood samples were taken in a fasted state at baseline and week 12, and stored at -80 °C until batch analysis. Measurements: Enzyme-linked immunosorbent assay for NPY, nerve growth factor, substance P, BDNF and vascular endothelial growth; Luminex multiplex assay for IL-1β, IL-6, IL-10 and TNF-α; Electrochemiluminescence detection method for cortisol, adrenocorticotropic hormone (ACTH), neuron-specific enolase (NSE) and S100B; Colorimetric assay for superoxide dismutase activity (SOD) and malondialdehyde concentration (MDA).

Statistical analysis

Statistical analyses were performed using R 4.3.0 software packages. For continuous variables, t-test or Mann-Whitney U test was used as appropriate. The analysis included mixed-effects linear models with time × group interaction terms, repeated measures using an unstructured covariance matrix, complete-case (main) and multiple imputation-based sensitivity analyses (20 imputations) for missing data. Independent predictors (variance inflation factor < 5) were identified through multivariable logistic regression and were included in a bootstrapped, validated nomogram (1000 iterations). Diagnostic performance and clinical utility were evaluated using receiver operating characteristic analysis with DeLong’s method and decision curve analysis. Two-sided P < 0.05 was significant for all tests.

RESULTS

Baseline characteristics

Of 178 patients with a confirmed diagnosis of PSD, 96 were enrolled in the CRG and 82 in the CG. The demographic and clinical characteristics at baseline are shown in Table 1. Baseline characteristics of patients in the two groups were well-balanced (all P > 0.05), including age, sex, body mass index, educational background, previous underlying condition, baseline HAMD-17 score, NIHSS score and MBI. Family involvement scores trended higher (not significant) in the CRG (31.8 ± 7.2 vs 29.5 ± 8.0; P = 0.068), which was adjusted for in sensitivity analyses. The mean time from stroke onset to initiation of rehabilitation did not differ significantly between both the groups (16.8 ± 2.5 days vs 17.2 ± 2.8 days; P = 0.331).

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

Characteristic	CRG (n = 96)	CG (n = 82)	P value
Age (years)	61.5 ± 12.8	63.2 ± 11.5	0.352
Female	52 (54.2)	43 (52.4)	0.809
BMI (kg/m2)	24.8 ± 3.6	25.2 ± 3.4	0.452
Education (years)	10.2 ± 3.8	9.8 ± 4.0	0.490
Married	80 (83.3)	68 (82.9)	0.946
Living alone	14 (14.6)	13 (15.9)	0.814
Employed	38 (39.6)	32 (39.0)	0.939
Hypertension	60 (62.5)	50 (61.0)	0.835
Type 2 diabetes	28 (29.2)	22 (26.8)	0.714
Hyperlipidemia	30 (31.3)	25 (30.5)	0.906
Atrial fibrillation	12 (12.5)	10 (12.2)	0.951
Prior stroke/TIA	18 (18.8)	14 (17.1)	0.775
Tobacco use	28 (29.2)	22 (26.8)	0.714
Baseline HAMD-17	18.2 ± 5.0	18.6 ± 5.3	0.615
Baseline SDS	62.5 ± 8.8	63.2 ± 9.0	0.603
Baseline NIHSS	8.2 ± 3.2	8.6 ± 3.5	0.445
Baseline MBI	45.2 ± 15.5	43.8 ± 14.8	0.534
Baseline MoCA	21.5 ± 4.2	21.2 ± 4.5	0.650
Family involvement (FSQ)	31.8 ± 7.2	29.5 ± 8.0	0.068
Antidepressant use	28 (29.2)1	24 (29.3)2	0.987

¹Comprehensive rehabilitation group: Venlafaxine XR 75-150 mg/day, initiated per protocol for Hamilton Depression Rating Scale-17 ≥ 20 with insufficient response at week 4.

²Control group: Antidepressants (principally sertraline 50-100 mg/day or venlafaxine XR 75-150 mg/day) prescribed at physician discretion. Between-group difference in antidepressant use rate was not statistically significant (P = 0.987).

CRG: Comprehensive rehabilitation group; CG: Control group; BMI: Body mass index; TIA: Transient ischemic attack; HAMD-17: Hamilton Depression Rating Scale-17; SDS: Zung Self-Rating Depression Scale; NIHSS: National Institutes of Health Stroke Scale; MBI: Modified Barthel Index; MoCA: Montreal Cognitive Assessment; FSQ: Family support questionnaire.

Stroke-related clinical features and neuroimaging

Table 2 summarizes stroke-related clinical and neuroimaging parameters at baseline. The groups were closely matched in terms of infarct location, TOAST etiological classification, hemisphere laterality, infarct volume, white matter hyperintensity burden and cerebrovascular collateral status (all P > 0.05), supporting the validity to compare efficacy between treatment groups.

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

Parameter	CRG (n = 96)	CG (n = 82)	P value
Admission NIHSS	8.2 ± 3.2	8.6 ± 3.5	0.445
Frontal lobe involvement	30 (31.3)	24 (29.3)	0.774
Basal ganglia-internal capsule	28 (29.2)	24 (29.3)	0.988
Limbic structures	20 (20.8)	16 (19.5)	0.832
Prefrontal-striato-thalamic circuit	26 (27.1)	22 (26.8)	0.967
Dominant hemisphere	58 (60.4)	50 (61.0)	0.941
Infarct volume (cm3)	10.2 ± 4.5	10.8 ± 4.8	0.395
Multiple infarcts	38 (39.6)	32 (39.0)	0.939
Fazekas total score	2.8 ± 1.2	3.0 ± 1.3	0.292
Periventricular WMH (Fazekas)	1.5 ± 0.6	1.6 ± 0.7	0.328
Deep WMH (Fazekas)	1.3 ± 0.6	1.4 ± 0.6	0.292
GCA frontal score	1.4 ± 0.6	1.5 ± 0.6	0.295
MTA score	1.3 ± 0.6	1.4 ± 0.7	0.348
Microbleed count	1.8 ± 1.4	1.9 ± 1.5	0.678
TOAST classification
Large artery atherosclerosis	32 (33.3)	28 (34.1)	0.895
Cardioembolism	18 (18.8)	14 (17.1)
Small vessel occlusion	34 (35.4)	30 (36.6)
Cryptogenic/other	12 (12.5)	10 (12.2)
ASITN/SIR collateral grade
0-1 (poor)	28 (29.2)	24 (29.3)	0.812
2-3 (moderate)	42 (43.8)	36 (43.9)
4 (good)	26 (27.1)	22 (26.8)

CRG: Comprehensive rehabilitation group; CG: Control group; NIHSS: National Institutes of Health Stroke Scale; WMH: White matter hyperintensity; GCA: Global cortical atrophy scale; MTA: Medial temporal atrophy; TOAST: Trial of Org 10172 in Acute Stroke Treatment; ASITN/SIR: American Society of Interventional and Therapeutic Neuroradiology/Society of Interventional Radiology.

Primary and secondary efficacy outcomes

Table 3 displays primary and secondary efficacy outcomes at 12 weeks. At 6 weeks, HAMD-17 scores were significantly lower in the CRG (10.5 ± 3.8 vs 15.8 ± 4.2; P < 0.001) and remained so at 12 weeks (7.2 ± 3.4 vs 13.8 ± 4.6; P < 0.001). The 12-week remission rate of depression was 67.7% in the CRG compared with 29.3% in the CG (P < 0.001). Mixed-effects model analyses confirmed a significant effect of time × group interaction on HAMD-17 (F = 48.32; P < 0.001), showing that the CRG benefit was progressive across rehabilitation sessions. Significant CRG superiority was also noted for FMA (78.5 ± 14.2 vs 58.3 ± 16.8, P < 0.001), MBI (76.8 ± 13.5 vs 58.6 ± 15.2; P < 0.001), MoCA (24.8 ± 3.5 vs 21.5 ± 4.2; P < 0.001) and SS-QOL (168.5 ± 28.5 vs 142.3 ± 30.2; P < 0.001). All between-group differences exceeded the pre-specified minimal clinically important differences (MCIDs): HAMD-17 ≥ 3 points, FMA ≥ 5.25 (upper limb) or ≥ 6 (lower limb) points, MBI ≥ 1.85 points, and MoCA ≥ 2 points.

Table 3 Primary and secondary efficacy outcomes at 6 and 12 weeks, mean ± SD/n (%).

Outcome measure	CRG (n = 96)	CG (n = 82)	P value	Effect size (Cohen’s d)
Primary outcomes
HAMD-17 at 6 weeks	10.5 ± 3.8	15.8 ± 4.2	< 0.001	1.34
HAMD-17 at 12 weeks	7.2 ± 3.4	13.8 ± 4.6	< 0.001	1.67
ΔHAMD-17 (baseline to 12 weeks)	-11.0 ± 4.2	-4.8 ± 3.5	< 0.001	1.62
Remission rate (HAMD-17 < 7)	65 (67.7)	24 (29.3)	< 0.001
SDS score at 12 weeks	42.5 ± 7.8	55.2 ± 9.5	< 0.001	1.51
Secondary outcomes
FMA at 12 weeks	78.5 ± 14.2	58.3 ± 16.8	< 0.001	1.31
MBI at 12 weeks	76.8 ± 13.5	58.6 ± 15.2	< 0.001	1.27
NIHSS at 12 weeks	2.4 ± 1.8	5.4 ± 2.2	< 0.001	1.50
MoCA at 12 weeks	24.8 ± 3.5	21.5 ± 4.2	< 0.001	0.85
SS-QOL at 12 weeks	168.5 ± 28.5	142.3 ± 30.2	< 0.001	0.89
FSS at 12 weeks	28.5 ± 8.2	38.2 ± 9.5	< 0.001	1.09

CRG: Comprehensive rehabilitation group; CG: Control group; HAMD-17: Hamilton Depression Rating Scale-17; SDS: Zung Self-Rating Depression Scale; FMA: Fugl-Meyer Assessment; MBI: Modified Barthel Index; NIHSS: National Institutes of Health Stroke Scale; MoCA: Montreal Cognitive Assessment; SS-QOL: Stroke-Specific Quality of Life; FSS: Fatigue Severity Scale.

Serum biomarker profiles

Results of serum biomarkers at baseline and 12 weeks are presented in Table 4. There were no baseline differences between groups for any biomarker (all P > 0.05). In CRG, NPY (98.6 ± 18.4 pg/mL vs 76.2 ± 15.8 pg/mL, P < 0.001), BDNF (22.5 ± 5.2 ng/mL vs 15.2 ± 4.8 ng/mL, P < 0.001), and IL-10 (12.8 ± 3.5 pg/mL vs 9.5 ± 3.2 pg/mL; P < 0.001). Concurrently, significant reductions in serum levels were noted for IL-1β (5.8 ± 2.2 pg/mL vs 8.5 ± 2.5 pg/mL; P < 0.001), IL-6 (5.2 ± 1.8 pg/mL vs 8.2 ± 2.4 pg/mL; P < 0.001), TNF-α (11.2 ± 3.0 pg/mL vs 14.8 ± 3.5 pg/mL; P < 0.001), morning cortisol (338.5 ± 82.5 nmol/L vs 378.2 ± 90.5 nmol/L; P < 0.001), NSE (10.2 ± 3.2 μg/L vs 13.8 ± 3.8 μg/L; P < 0.001) and S100B (0.32 ± 0.12 μg/L vs 0.48 ± 0.16 μg/L; P < 0.001). At 12 weeks, SOD activity was significantly higher (108.2 ± 22.8 U/mL vs 88.5 ± 20.2 U/mL; P < 0.001) and MDA lower (5.2 ± 1.8 nmol/mL vs 7.2 ± 2.2 nmol/mL; P < 0.001) in the CRG.

Table 4 Serum biomarker profiles at baseline and 12-week follow-up, mean ± SD/n (%).

Biomarker	CRG baseline	CRG 12 weeks	CG baseline	CG 12 weeks	Inter-group P (12 weeks)
Neuropeptides and neurotrophins
NPY (pg/mL)	76.2 ± 15.8	98.6 ± 18.4a	74.8 ± 16.5	80.5 ± 17.2	< 0.001
BDNF (ng/mL)	13.8 ± 4.6	22.5 ± 5.2a	13.5 ± 4.4	15.2 ± 4.8	< 0.001
NGF (pg/mL)	42.5 ± 10.2	58.8 ± 12.6a	41.8 ± 10.8	46.2 ± 11.5	< 0.001
VEGF (pg/mL)	185.2 ± 42.5	248.6 ± 55.8a	182.8 ± 40.8	198.5 ± 45.2	< 0.001
Inflammatory cytokines
IL-1β (pg/mL)	10.5 ± 2.6	5.8 ± 2.2a	10.8 ± 2.8	8.5 ± 2.5a	< 0.001
IL-6 (pg/mL)	9.8 ± 2.5	5.2 ± 1.8a	10.2 ± 2.8	8.2 ± 2.4a	< 0.001
IL-10 (pg/mL)	8.2 ± 2.8	12.8 ± 3.5a	8.5 ± 3.0	9.5 ± 3.2	< 0.001
TNF-α (pg/mL)	16.8 ± 3.8	11.2 ± 3.0a	17.2 ± 4.0	14.8 ± 3.5a	< 0.001
Neuroendocrine markers
Morning cortisol (nmol/L)	385.2 ± 88.5	338.5 ± 82.5a	382.8 ± 90.2	378.2 ± 90.5	< 0.001
ACTH (pg/mL)	40.2 ± 8.8	33.8 ± 7.5a	41.5 ± 9.2	39.5 ± 8.8	< 0.001
Oxidative stress and neuronal injury
SOD (U/mL)	82.5 ± 18.5	108.2 ± 22.8a	80.8 ± 17.8	88.5 ± 20.2	< 0.001
MDA (nmol/mL)	8.5 ± 2.2	5.2 ± 1.8a	8.8 ± 2.5	7.2 ± 2.2a	< 0.001
NSE (μg/L)	16.8 ± 4.5	10.2 ± 3.2a	17.2 ± 4.8	13.8 ± 3.8a	< 0.001
S100B (μg/L)	0.58 ± 0.18	0.32 ± 0.12a	0.60 ± 0.20	0.48 ± 0.16a	< 0.001

^aP < 0.05 vs respective baseline (paired t-test or Wilcoxon signed-rank).

CRG: Comprehensive rehabilitation group; CG: Control group; NPY: Neuropeptide Y; BDNF: Brain-derived neurotrophic factor; IL: Interleukin; NGF: Nerve growth factor; VEGF: Vascular endothelial growth factor; TNF: Tumor necrosis factor; ACTH: Adrenocorticotropic hormone; SOD: Superoxide dismutase; MDA: Malondialdehyde; NSE: Neuron-specific enolase.

Multivariable logistic regression: Independent predictors of treatment response

All factors with P < 0.10 on univariable analysis were included in multivariable logistic regression. Independent predictors of comprehensive rehabilitation treatment response. The most potent predictor was early initiation of psychological intervention [≤ 2 weeks, odds ratio (OR) = 4.12, 95% confidence interval (CI): 2.28-7.44, P < 0.001], followed by baseline serum NPY ≥ 82.5 pg/mL (OR = 3.38, 95%CI: 1.95-5.86, P < 0.001), and mild-to-moderate depression severity (OR = 2.94, 95%CI: 1.68-5.14, P < 0.001). The remaining predictors were high family involvement (OR = 2.65), intact cognitive status (OR = 2.42), lesion in the non-dominant hemisphere (OR = 2.08) and low baseline IL-1β (OR = 1.86). The Hosmer-Lemeshow test demonstrated adequate fit of the model (χ² = 6.45, P = 0.597) (Table 5).

Table 5 Multivariable logistic regression analysis of independent predictors of treatment response.

Variable	β	SE	Wald χ2	OR (95%CI)	P value
Early psychological intervention initiation (≤ 2 weeks)	1.416	0.302	21.98	4.12 (2.28-7.44)	< 0.001
Baseline serum NPY ≥ 82.5 pg/mL	1.217	0.281	18.75	3.38 (1.95-5.86)	< 0.001
Mild-to-moderate depression (HAMD-17 < 17)	1.079	0.286	14.24	2.94 (1.68-5.14)	< 0.001
High family involvement (FSQ ≥ 28)	0.975	0.264	13.64	2.65 (1.58-4.44)	< 0.001
Intact cognition (MMSE ≥ 24)	0.884	0.261	11.45	2.42 (1.45-4.04)	0.001
Non-dominant hemisphere lesion	0.732	0.273	7.19	2.08 (1.22-3.55)	0.008
Baseline IL-1β < 8.2 pg/mL	0.621	0.269	5.33	1.86 (1.10-3.15)	0.021
Non-significant variables (P ≥ 0.10)
Homocysteine < 15 μmol/L	0.468	0.278	2.83	1.60 (0.93-2.75)	0.093
Baseline NSE < 15 μg/L	0.435	0.265	2.69	1.55 (0.92-2.60)	0.101
Infarct volume < 10 cm3	0.388	0.258	2.26	1.47 (0.89-2.44)	0.133
No diabetes mellitus	0.342	0.260	1.73	1.41 (0.85-2.34)	0.189
Good collateral circulation	0.318	0.272	1.37	1.37 (0.81-2.33)	0.242

Hosmer-Lemeshow goodness-of-fit: χ² = 6.45, P = 0.597. OR: Odds ratio; CI: Confidence interval; NPY: Neuropeptide Y; HAMD-17: Hamilton Depression Rating Scale-17; FSQ: Family support questionnaire; MMSE: Mini-Mental State Examination; IL: Interleukin; NSE: Neuron-specific enolase.

Nomogram model performance and biomarker diagnostic value

The seven-predictor nomogram provided a significantly higher area under the curve (AUC) than any single predictor alone (all DeLong P < 0.01): AUC = 0.87 (95%CI: 0.81-0.92), sensitivity 80.2%, specificity 83.6%, Bootstrap-validated C-index 0.85. The highest single-predictor AUC was observed for NPY (AUC = 0.82, 95%CI: 0.76-0.88), followed by BDNF (AUC = 0.79), IL-1β (AUC = 0.74), IL-6 (AUC = 0.72) and NIHSS score (AUC = 0.68). Calibration curve analysis indicated good agreement between predicted and observed probabilities. The nomogram showed positive net benefit on decision curve analysis across threshold probabilities of 5%-85%. These findings are exploratory and hypothesis-generating; external prospective validation is required before the nomogram or NPY thresholds are applied in clinical practice (Figures 1 and 2).

Figure 1 Receiver operating characteristic curves comparing the nomogram prediction model against individual biomarkers for predicting comprehensive rehabilitation treatment response. The nomogram model integrating seven independent predictors achieved the highest area under the curve (AUC) (0.87), followed by neuropeptide Y alone (AUC = 0.82). All curves were significantly superior to the diagonal reference line (all P < 0.001, DeLong method). AUC: Area under the curve; CI: Confidence interval; NPY: Neuropeptide Y; BDNF: Brain-derived neurotrophic factor; IL: Interleukin.

Figure 2 Forest plot displaying odds ratios and 95% confidence intervals for seven independent predictors of comprehensive rehabilitation treatment response from multivariable logistic regression analysis. All confidence intervals lie entirely to the right of odds ratio = 1.0 (dashed reference line), confirming statistical significance. Colored squares represent point estimates; horizontal lines represent 95% confidence intervals. OR: Odds ratio; CI: Confidence interval; NPY: Neuropeptide Y; HAMD-17: Hamilton Depression Rating Scale-17; FSQ: Family support questionnaire; MMSE: Mini-Mental State Examination; IL: Interleukin.

Depression severity subgroup analysis

Stratified analyses according to depression severity (Table 6) highlighted a marked inverse dose-response relationship between baseline depression severity and remission induced by rehabilitation. In patients with mild depression (HAMD-17: 8-16; n = 62), CRG remission rates achieved 82.4%, significantly higher than in the CG (40.5%, P < 0.001). In the moderate subgroup (HAMD-17: 17-23; n = 72), remission rates were 62.5% vs 21.7% (P < 0.001). In the severe subgroup (HAMD-17 ≥ 24; n = 44), absolute remission rates were lower (CRG 38.5% vs CG 10.5%, P < 0.001) but the absolute risk reduction was still clinically relevant at 28.0 percentage points. Post-treatment NPY elevation declined incrementally with increasing baseline severity (ΔNPY: Mild 26.8 ± 8.5 pg/mL, moderate 21.2 ± 7.8 pg/mL, severe 14.5 ± 6.5 pg/mL; P trend < 0.001), mirroring the severity-dependent gradient in BDNF upregulation and inflammatory suppression seen above.

Table 6 Depression severity subgroup analysis: Rehabilitation outcomes and biomarker trajectories.

Characteristic	Mild (n = 62) HAMD 8-16	Moderate (n = 72) HAMD 17-23	Severe (n = 44) HAMD ≥ 24	P trend
Baseline HAMD-17	11.5 ± 2.2	19.8 ± 1.8	26.5 ± 3.0	< 0.001
Admission NIHSS	5.8 ± 2.0	8.8 ± 2.8	11.5 ± 3.2	< 0.001
Dominant hemisphere, %	50.0	62.5	72.7	0.018
Frontal lobe involvement, %	22.6	33.3	47.7	0.012
CRG remission rate (%)	82.4	62.5	38.5	< 0.001
CG remission rate (%)	40.5	21.7	10.5	< 0.001
12-week MBI	80.5 ± 11.8	72.5 ± 13.0	62.8 ± 14.5	< 0.001
Post-intervention biomarker changes (CRG only)
ΔNPY (pg/mL)	26.8 ± 8.5	21.2 ± 7.8	14.5 ± 6.5	< 0.001
ΔBDNF (ng/mL)	12.5 ± 3.8	8.8 ± 3.2	5.5 ± 2.8	< 0.001
ΔIL-1β (pg/mL)	-5.8 ± 2.0	-4.2 ± 1.8	-2.5 ± 1.5	< 0.001
ΔTNF-α (pg/mL)	-7.2 ± 2.5	-5.5 ± 2.2	-3.2 ± 1.8	< 0.001
Baseline NPY (pg/mL)	82.5 ± 14.5	72.8 ± 15.2	62.5 ± 13.8	< 0.001
Fazekas total score	2.2 ± 1.0	3.0 ± 1.2	4.0 ± 1.3	< 0.001
Family involvement (FSQ)	35.8 ± 6.8	30.5 ± 7.5	26.8 ± 7.0	< 0.001

P trend calculated by Jonckheere-Terpstra test for ordered alternatives. Δ denotes post-intervention minus baseline change in comprehensive rehabilitation group. HAMD: Hamilton Depression Rating Scale; NIHSS: National Institutes of Health Stroke Scale; CRG: Comprehensive rehabilitation group; CG: Control group; MBI: Modified Barthel Index; NPY: Neuropeptide Y; BDNF: Brain-derived neurotrophic factor; IL: Interleukin; TNF: Tumor necrosis factor; FSQ: Family support questionnaire.

DISCUSSION

This retrospective study shows the effectiveness of a comprehensive intervention led by psychological rehabilitation for PSD and establishes a validated nomogram prediction model for individualized decision-making in clinical practice. In addition to contributing a 38.4-percentage-point absolute improvement in remission rate compared with controls and effective motor and cognitive gains, the core findings of neurobiologic remodeling from within the same genes appear to be that NPY was elevated while inflammation was suppressed and that seven independent predictors of treatment response were identified; all these add meaning to precision stroke neurorehabilitation.

The CRG’s 67.7% remission rate dwarfed the 29.3% of standard care (number needed to treat approximately 2.6), performing better than CBT-based interventions[15,16] and outpacing pharmacotherapy alone[26]. The mixed-effects model’s progressive time × group interaction indicates the gradual process of gaining skills associated with CBT and expression of timescale delayed neuroplasticity[27]. While the improvements observed for concurrent FMA and MBI exceed their respective MCID[27], they also emphasize a psychomotor bidirectionality between neuropsychiatric status and motor performance, whereby less depressive withdrawal facilitates movement experience, mediated by changes in cortical representation while further enhance self-efficacy[28].

Early psychological intervention (≤ 2 weeks; OR = 4.12) was the most significant independent predictor of response Indeed, the early post-infarction epoch is marked by maximum synaptic plasticity[26] with transient BDNF-tyrosine receptor kinase B upregulation and long-term potentiation-permissive substrate in peri-infarct cortex[29] a “neuroplastic window” temporally coincident with maximal susceptibility to maladaptive cognitions and HPA hyperactivation. Provision of CBT within this window could harness cortical reorganization potential prior to Hebbian-consolidation of maladaptive representations[27]. These were shown to justify incorporating psychiatric screening and psychological management planning as part of acute stroke admission routines, rather than relegating such consideration to referral upon discharge.

Baseline NPY ≥ 82.5 pg/mL had the strongest single-biomarker predictive value (OR = 3.38, AUC = 0.82). NPY buffers stress reactivity, enhances GABAergic tone and inhibits the HPA axis through Y1/Y2 receptors in the hippocampus, hypothalamus, and amygdala[13,14]. High baseline NPY may thus reflect maintained neurobiological substrate for appropriate responses to the challenges of CBT. This is consistent with in situ synergy for NPY upregulation by CBT, mindfulness, and aerobic exercise through independent pathways[21,30], while post-intervention NPY elevation (mean ± 22.4 pg/mL) in the CRG confirms this dynamic effect. The severity-dependent gradient (mild ± 26.8 pg/mL vs severe ± 14.5 pg/mL) suggests a need for pharmacological NPY augmentation in the setting of severe PSD to extend the therapeutic ceiling of psychological rehabilitation. It should be noted that the NPY cut-point of ≥ 82.5 pg/mL and its AUC of 0.82 were derived from the same cohort used for model derivation; these values are subject to optimism bias and require prospective external validation before they can be considered clinically actionable.

Complementary decrease of IL-1β, IL-6 and TNF-α with an increase in the levels of IL-10 within the CRG indicates a synchronized immune shift from pro-inflammatory to regulatory subtypes, driven by CBT-mediated reductions in sympathoadrenal activation[10] as well as exercise-driven anti-inflammatory myokine release[11] and mindfulness-based cortico-limbic downregulation[31]. Since baseline IL-1β was shown to predict response independently (OR = 1.86), augmenting anti-inflammatory strategies may optimize the neurobiological circumstances for psychological skill acquisition when pathological inflammatory burden exists within patients[10].

Post-intervention reductions in morning cortisol and ACTH levels suggest attenuation of HPA axis reactivity; however, since only a single morning time-point was measured, conclusions about normalization of the full diurnal cortisol rhythm cannot be drawn from the present data and require dedicated longitudinal cortisol sampling in future studies. Prospective studies are warranted to determine whether early cortisol awakening response trajectories (week 4-6) predict eventual CBT response.

Intact cognition (higher the MMSE ≥ 24; OR = 2.42) reflects the CBT’s reliance on working memory, verbal fluency and executive function for skill acquisition and generalization[29]. It also emphasises the immediacy for cognitively adapted CBT (CA-CBT) protocols that use simplified language, visual supports and session completion with relatives as co-therapists for patients with mild cognitive impairment[27], as well as non-verbal therapeutic modalities including music therapy and movement-based mindfulness[28].

Non-dominant hemisphere lesion (OR = 2.08) aligns with hemispheric specialization effect models: Stroke impacting the dominant hemisphere interrupts the linguistic architecture of CBT as well as left prefrontal down-regulation of amygdalar hyperactivation[32]. Multimodal approaches combining verbal CBT with behavioral activation and arts-based methods may provide better results in this subgroup.

Reinforcing the socioecological nature of recovery, high family involvement (OR = 2.65). Family members act as behavioral activation agents, offering prompting and reinforcement between sessions. Family psychoeducation promoting non-critical acceptance and co-participation in goal-setting decrease dyadic distress and increase rehabilitation engagement[19]. The sensitivity analyses adjusting for the family support questionnaire score did not materially affect our primary efficacy conclusions; however, future prospective studies should directly control for family involvement.

The nomogram considering all seven predictors displayed excellent discriminatory performance (AUC = 0.87, Bootstrap C-index = 0.85) and net clinical utility at any of the tested decision thresholds on decision curve analysis vs published stroke rehabilitation prognostic models[33-36]. Its visual structure allows for estimation of probability without computation tools, a practical advantage in resource-limited settings. Such results are hypothesis-generating; the nomogram and relevant cut-points should not be either used to direct clinical decisions or reported in management guidelines until being validated externally within a multicenter cohort of patients or randomized trial. The immediate research priority is a prospective external multicenter validation.

Additional biological advantages increased SOD, decreased MDA, and decreased NSE and S100B indicate reduced oxidative stress and neuronal damage. Although the uptimes observed for SOD and MDA were phenotypically aligned with increased antioxidant defense mechanisms consistent with nuclear factor erythroid 2-related factor 2 (Nrf2)-antioxidant response element patterns in prior studies of exercise or CBT, elucidation of direct mechanistic confirmation via Nrf2 protein expression or transcriptional analysis was not within the framework of this experiment and merits further molecular study.

Limitations include non-randomized retrospective design with potential unmeasured confounders, single-center setting restricting generalizability, the inherent non-blinding of psychotherapy and a 12-week follow-up that was too-short to characterize long-term durability. Peripheral NPY reflect may not correspond with corresponding central dynamics. Ongoing phase II studies will help determine the efficacy of this short treatment interval and inform phase III feasibility. Future directions include multicenter randomized controlled trial (RCT) validation with 12-month follow-up; neuroimaging mechanistic substudies (resting-state functional MRI, diffusion tensor imaging, positron emission tomography neuroinflammation); pharmacological NPY augmentation trials specifically for moderate to severe PSD; consideration of CA-CBT development if efficacious in RCTs; and identification, model testing and functional characterization of polymorphisms within NPY pathway inflammatory pathways through pharmacogenomic investigations.

CONCLUSION

A multi-modal rehabilitation program focused on CBT produced large reductions in depressive symptoms and motor improvement in PSD, with a 38.4-percentage-point increase in remission rate at week 12 relative to standard care. The response of treatment was driven by the time when intervention was made, the neurobiological milieu, cognitive healthiness, family engagement and lesion laterality. Serum NPY became a candidate biomarker of treatment response with an elevation seen post-intervention and in line with restoration of neurobehavioral stress-resilience substrates. An internally validated nomogram may serve as a personalized decision-support tool after multicenter prospective validation. These results facilitate precision stroke neurorehabilitation with actionable biological and psychosocial targets for multimodal rehabilitation plans.