Post-stroke depression update 2025: Mechanisms, prediction, and management

Abstract

Post-stroke depression (PSD) has a stable prevalence of 30%. Beyond stroke severity, lesion location and prior depression, vitamin D deficiency, hyperhomocysteinaemia and insulin resistance are now recognised as independent risk factors. Mechanistically, phenotypic switching of microglia and astrocytes drives neuro-inflammation via NF-κB, JAK-STAT3 and NLRP3 inflammasome signalling. Gut-brain axis studies show reduced Faecalibacterium, increased Enterococcus and decreased short-chain fatty acids. Multi-omics evidence also indicates down-regulation of BDNF/TrkB, glutamate excitotoxicity, hypothalamic-pituitary-adrenal-axis dysregulation and altered miR-146a/34a expression. Predictively, machine-learning models integrating magnetic resonance imaging lesion-network patterns, clinical variables and inflammatory markers achieve area under the curve > 0.90, although external validation is still required. Therapeutically, selective serotonin reuptake inhibitors (sertraline, escitalopram) remain first-line, and evidence for agomelatine, vitamin D and tumor necrosis factor-α inhibitors is increasing. Traditional Chinese medicine (TCM) formulations such as Chaihu Shugan San, Danzhi Xiaoyao San and Shugan Jieyu capsule attenuate neuro-inflammation and restore synaptic plasticity by modulating JAK-STAT3-GSK3β/ PTEN/Akt, Nrf2/HO-1 and NMDAR/BDNF pathways. Additionally, acupuncture, repetitive transcranial magnetic stimulation, exercise and music therapy have proven safe and effective. Looking ahead, PSD research should integrate multi-omics, artificial intelligence and individualised TCM to establish a full-chain paradigm for risk prediction and precision stratification.

Key Words: Post-stroke depression; Risk factors; Neuroinflammation; Gut-brain axis; Precision medicine

Core Tip: Post-stroke depression risk is refined by emerging biomarkers like vitamin D deficiency, hyperhomocysteinemia, and insulin resistance, alongside traditional factors. Multi-omics and artificial intelligence-driven models integrating clinical, neuroimaging, and inflammatory data show high predictive potential for precision stratification. Mechanistically, microglial/astrocytic polarization-driven neuroinflammation and gut-brain axis dysregulation are key therapeutic targets. Beyond first-line selective serotonin reuptake inhibitors, agomelatine, vitamin D supplementation, anti-inflammatory strategies, and evidence-based traditional Chinese medicine formulations that modulate neuro-immune crosstalk offer promising adjunctive treatments. A “predict-mechanism-intervene” paradigm is advocated to translate these advances into personalized management and improve long-term outcomes.

INTRODUCTION

Post-stroke depression (PSD), a prevalent neuropsychiatric complication following cerebrovascular events (including ischemic and hemorrhagic stroke), is characterized by a constellation of symptoms such as persistent low mood, anhedonia, diminished self-worth, appetite disturbance, and sleep irregularities[1]. It is crucial to distinguish PSD as a distinct clinical entity stemming from the complex pathophysiology of stroke, rather than a mere psychological reaction to the event itself[2]. Afflicting approximately one-third of stroke survivors, PSD significantly exacerbates post-stroke cognitive impairment, impedes functional recovery, and substantially diminishes the overall quality of life[3]. Despite being a preventable and treatable condition through early screening, multimodal psychological interventions, and pharmacotherapy [e.g., selective serotonin reuptake inhibitors (SSRI)], the current management of PSD remains suboptimal. Existing therapeutic strategies exhibit limited efficacy in addressing the multifaceted pathophysiology of PSD, which encompasses neuroinflammation, oxidative stress, and impaired neuroplasticity[4]. Consequently, PSD contributes disproportionately to long-term disability, mortality, and healthcare burdens, underscoring an urgent, unmet clinical need.

The past few years have witnessed remarkable advancements in elucidating the risk factors, pathophysiology, and predictive models for PSD. Established risk factors include stroke severity, pre-stroke depression, female sex, advanced age, social isolation, lesions in specific neurocircuits (e.g., frontal lobes, basal ganglia), and comorbidities such as hypertension and diabetes[5]. Concurrently, the underlying mechanisms are increasingly being unraveled, pointing to dysregulation in neurotrophic signaling (e.g., BDNF), neurotransmitter imbalances, and neuroimmune axis dysregulation. Notably, the integration of machine learning and deep learning with multi-omics data (e.g., plasma proteomics) is revolutionizing early diagnosis and prognostic risk assessment, even linking specific mood profiles to aberrant peripheral immune responses[6]. Furthermore, cutting-edge research into modulating neural-immune pathways and the emergence of evidence-based complementary approaches[7], such as acupuncture, is paving the way for novel, precise therapeutic strategies[8].

This review synthesizes the most compelling evidence published between 2022 and 2025. We provide a systematic overview of the PSD landscape, sequentially examining risk factors, pathophysiological mechanisms, predictive models, and therapeutic interventions. More importantly, we propose a forward-looking “predict-mechanism-intervene” translational paradigm, aiming to frame future research and guide the development of personalized medicine for PSD.

PSD RESEARCH LANDSCAPE AND TREND ANALYSIS

The scientific inquiry into PSD has evolved substantially, progressing from initial phenomenological descriptions to a multidisciplinary field at the forefront of neuroscience and psychiatry. To quantitatively delineate this evolution and establish an objective, evidence-based framework for the present review, we conducted a comprehensive bibliometric analysis. This analysis serves not merely as a retrospective but as a strategic roadmap, illuminating the historical trajectory, current hotspots, and future directions of PSD research, thereby providing a robust justification for the narrative structure of this update.

Our analysis was performed using established scientometric tools [(CiteSpace 6.3.R1), VOSviewer (1.6.20), R (version 4.5.1), Origin (2025) and Microsoft Excel]. A systematic retrieval of the Web of Science Core Collection (from January 1, 1984, to September 30, 2025) was conducted using the search query (“post-stroke depression” OR “post stroke depression” OR “PSD”), which yielded 1245 publications. After applying stringent inclusion criteria and excluding records that were not articles or reviews and those not in English, 851 publications remained for in-depth evaluation. From this set, 833 publications were ultimately identified and included for the bibliometric analysis, which encompassed publications, countries, authors, and keywords (Figure 1A). The annual publication output demonstrates a striking, near-exponential growth over the past decade (Figure 1B), a trend that powerfully underscores the escalating global recognition of the clinical and scientific significance of PSD. The collaborative network analysis further reveals the transition of PSD research from isolated efforts to a globally integrated endeavor. China and the United States stand as the most prolific contributors (Figure 1C).

Figure 1 Bibliometric analysis of global research trends in post-stroke depression. A: Flowchart of the study selection process. A systematic search of the Web of Science Core Collection using keywords related to “post-stroke depression” identified 1245 publications. After applying inclusion criteria [e.g., article or review, English language, studies involving human patients, animal models, or cellular models of post-stroke depression (PSD)], 833 publications were included for final bibliometric analysis; B: Annual publication output and citation impact from 1984 to 2024. The bar chart shows the number of publications per year, illustrating the growth of research volume. The line graph represents the average number of citations per year (MeanTcPerYear), indicating the temporal trend of scholarly influence; C: Citation analysis by country. The horizontal bar chart displays the top-contributing countries ranked by their total citation counts, highlighting the most influential countries in the PSD research field. Analyses were performed using CiteSpace (6.3.R1), VOSviewer (1.6.20), R (4.5.1), and Excel (2025).

A dedicated comparative bibliometric analysis between these two nations reveals distinct trajectories and collaboration patterns (Figure 2). Specifically, Chinas publication volume has shown exponential growth since 2010, while the United States has maintained a stable output (Figure 2A). Geographically, research output within China is concentrated in eastern provinces such as Jiangsu and Beijing, with a marked increase in recent years, whereas in the United States, states like California and Massachusetts are established research hubs (Figure 2B and C). At the global level, both countries serve as central, highly interconnected nodes within the international collaboration network (Figure 2D). This expansive cooperation extends to the institutional level, forming a complex web of both domestic and cross-national research partnerships (Figure 2E).

Figure 2 Comparative bibliometric analysis of post-stroke depression research between China (mainland) and the United States (mainland) (1984-2025). A: Annual publication trends (1984-2025). A line graph comparing the yearly number of post-stroke depression (PSD)-related publications from China (mainland, red solid line) and the United States (mainland, blue dashed line). The timeline is divided into four phases (1984-2005, 2006-2012, 2013-2019, 2020-2025) to reflect distinct periods of research policy and investment. China shows exponential growth after 2010, peaking around 2023, whereas United States output remains relatively stable; B: Geographical distribution of publications in China (mainland). A choropleth-bar map illustrating provincial-level research output across four time periods. Each province is marked by a cluster of vertical bars color-coded by period: Red (1984-2005), blue (2006-2012), yellow (2013-2019), and green (2020-2025). Bar height corresponds to publication count. High-output provinces include Jiangsu, Beijing, Shanghai, Guangdong, and Zhejiang, with pronounced growth in the most recent period. Regions such as Tibet and Qinghai show minimal activity; C: Geographical distribution of publications in the United States (mainland). A corresponding choropleth-bar map for United States, using the same period-based color scheme. States with the highest research output include California, Massachusetts, New York, Illinois, and Texas, reflecting established regional research hubs; D: International collaboration network. A node-link diagram generated with VOSviewer, displaying co-authorship links between countries. Node size represents publication volume, and link thickness indicates collaboration strength. China (“peoples r China”) and the United States (“USA”) are the largest and most central nodes, connected to numerous other countries (e.g., Thailand, Belgium), underscoring their pivotal roles in global PSD research networks; E: Institutional collaboration network. A detailed network map highlighting cooperative relationships among major research institutions. Nodes represent institutions (e.g., Harvard University, Stanford University, MIT in the United States; Chinese Academy of Sciences, Nanjing Medical University in China), colored by country/region. Links denote co-authorship, with thicker lines indicating stronger collaboration. The network reveals both domestic clusters and cross-national partnerships, illustrating the complex, globally interconnected nature of PSD research. Analyses were performed using CiteSpace (6.3.R1), VOSviewer (1.6.20), Origin (2025), and Excel (2025).

Most critically, the intellectual evolution of the field is vividly captured through keyword co-occurrence and burst detection analysis. The timeline visualization (Figure 3A) charts a clear conceptual shift: Early research was predominantly anchored in clinical phenomenology and epidemiological risk factors. The subsequent phase witnessed a surge in interest in “randomized controlled trials (RCTs)” and “rehabilitation” strategies. The most recent era, however, is decisively characterized by a pivot towards molecular and systems-level mechanisms. This shift is emphatically highlighted by the keyword burst analysis (Figure 3B), which identifies “neuroinflammation”, “oxidative stress”, and the “gut-brain axis” as the most prominent and rapidly emerging frontiers-a finding corroborated by the keyword heatmap (Figure 3C). This data-driven perspective unequivocally confirms that the field is currently in a phase of deep mechanistic exploration.

Figure 3 Evolution of research fronts and thematic trends in post-stroke depression. A: Keyword co-occurrence network map. Nodes represent high-frequency keywords clustered into distinct research themes (e.g., 0 elderly patient), with node size proportional to frequency and links indicating co-occurrence strength. The multi-color gradient (red, yellow, green, blue) reflects the temporal evolution of keywords from 1989 to 2023, based on a timeline spanning 1995 to 2025, illustrating the shift from clinical phenomenology to mechanistic studies; B: Top 25 keywords with the strongest citation bursts. The bar chart displays burst strength (horizontal axis) and duration (blue-to-red gradient bars) for keywords spanning from 1995 to 2025. Early bursts include “mood disorders” and “recovery”, while recent bursts highlight “randomized controlled trial”, “default mode network”, and “post - stroke depression”, reflecting the evolution of research focus from clinical phenomenology to mechanistic and interventional studies; C: Temporal heatmap of keyword frequency. Rows represent years (1989-2023), columns show keywords, and color intensity (purple to yellow) indicates annual keyword prominence, visualizing the dynamic evolution of research hotspots over time. Analyses performed using CiteSpace (6.3.R1) and R (4.5.1). Clustering and burst detection reveal a paradigm shift toward neuroimmune mechanisms and systems-level interventions in recent decades.

The bibliometric evidence presents an indisputable logical sequence: The field’s foundation in risk identification has naturally progressed to a fervent investigation of underlying pathophysiological mechanisms, with the ultimate goal of informing targeted therapeutic interventions. It is this very trajectory that dictates the narrative flow of this review. We will first dissect the established and emerging risk factors, which form the basis for predictive models. We will then delve into the intricate mechanistic networks, particularly the neuro-immune-inflammatory and gut-brain axes, which currently represent the core of scientific inquiry. Finally, we will evaluate the translation of this knowledge into novel therapeutic strategies, from pharmacotherapy to neuromodulation and integrative medicine. By aligning our structure with the natural evolution of the science itself, this review aims to systematically synthesize the latest evidence and advance the “predict-mechanism-intervene” paradigm, thereby guiding the field toward a future of predictable, subclassifiable, and precisely treatable PSD.

EPIDEMIOLOGY AND NATURAL HISTORY

The global burden of PSD has been steadily rising, a trend exacerbated by the aging global population. PSD represents one of the most frequent neuropsychiatric sequelae of stroke, with a pooled prevalence ranging between 18% and 33% according to meta-analyses[9]. Individual cohort studies report consistent figures, with incidence rates of 29%-31% and 26.51% in specific patient groups[10]. Critically, the risk of developing PSD persists over the long term. The incidence is highest within the first year post-stroke (29%-33%), and the cumulative incidence can escalate to 39%-52% over five years[11]. This underscores PSD not as a transient reaction but as a chronic complication requiring sustained clinical vigilance. The temporal presentation of PSD is heterogeneous. It can emerge at any phase of stroke recovery, with comparable frequencies reported in the acute (< 1 month), subacute (1-6 months), and chronic (> 6 months) phases, each with an incidence of approximately one-third of patients[12]. This suggests that the pathophysiological triggers may evolve over time.

Geographic and socioeconomic disparities significantly influence PSD prevalence. Studies indicate a substantially lower prevalence in developed nations compared to regions like Africa, likely reflecting disparities in healthcare infrastructure, cultural attitudes toward mental health, and socioeconomic determinants[13]. Lesion location is a well-established, though not exclusive, contributor to PSD risk. Evidence consistently implicates disruptions in specific neural circuits. For instance, intraventricular hemorrhage, right-sided lesions, and dysfunction within the cortico-striatal-pallidal-thalamocortical projections are associated with a higher likelihood of PSD[14,15]. Lesions in prefrontal cortex, limbic areas, and basal ganglia are also frequently linked to depressive symptoms, highlighting the importance of network-level dysfunction[16].

The influence of demographic factors like age and sex reveals the complexity of PSD etiology. The relationship between age and PSD risk remains a subject of debate. While some studies suggest younger patients are more vulnerable[17], possibly due to the profound psychosocial impact of a debilitating event early in life, others report a positive correlation with advanced age, potentially reflecting the cumulative effect of age-related biological vulnerabilities and social isolation[18]. Similarly, the role of sex is nuanced. Although some research indicates that female sex is an independent risk factor, potentially linked to greater psychosocial stressors and a higher likelihood of living alone post-stroke[19], other well-designed studies note that female stroke patients are often older and present with more severe depressive symptoms[20]. These apparent contradictions underscore the multifactorial nature of PSD, where biological, psychological, and social determinants intersect.

RISK FACTORS AND PREDICTIVE MODELS

The established understanding of PSD risk is anchored in a set of well-recognized, clinically accessible factors. These traditional determinants include the severity of the initial stroke, the degree of resultant physical disability, a history of pre-stroke depression, and the presence of cognitive impairment[21]. Beyond these clinical metrics, neuroanatomical specificities play a critical role. Lesions affecting structures integral to mood regulation, such as the right amygdala, globus pallidus, and key nodes within the frontal cortical-basal ganglia-thalamic circuits, are consistently associated with a higher incidence of PSD, as evidenced by a single-center lesion-symptom mapping study[22]. Furthermore, the severity of depressive symptoms has been linked to damage in the dorsolateral prefrontal cortex (DLPFC) and inferior frontal gyrus, while symptom progression is associated with lesions in the right insular cortex, putamen, and inferior frontal gyrus[23].

The past decade has witnessed a significant expansion of the PSD risk landscape, driven by the proliferation of large-scale databases and the application of advanced analytical techniques. This has accelerated the identification of promising peripheral biomarkers. Converging evidence from clinical studies underscores the prognostic value of vitamin D deficiency, hyperhomocysteinemia, and insulin resistance (IR). Specifically, lower serum vitamin D levels in acute stroke patients show a significant inverse correlation with inflammatory markers such as interleukin (IL)-6 and high-sensitivity C-reactive protein (hs-CRP), although this finding from a single-center study requires further validation[24], and are independently associated with a higher risk of depression at one month, even after adjusting for seasonal variations, a result also derived from a single-center cohort[25]. Critically, the relationship between vitamin D deficiency and PSD is not direct but is mediated by poor sleep quality, highlighting an interconnected pathway, as suggested by a single-center cohort study awaiting external confirmation[26]. Parallel meta-analyses confirm that patients with PSD have significantly higher baseline levels of homocysteine (Hcy) and hs-CRP compared to non-PSD patients[27,28]. The genetic underpinnings of Hcy metabolism are further elucidated by the finding that the MTHFR rs1801133 AA genotype increases PSD risk, an effect mediated through elevated Hcy levels, as explored in a single-center study aimed at investigating the relationships between Hcy metabolism genes, Hcy levels, and early-onset PSD[29]. Regarding metabolic dysregulation, IR, quantified by HOMA-IR, is reported as an independent predictor of PSD in diabetic stroke patients, conferring a 3.28-fold increased risk at three months [area under the curve (AUC) = 0.760, based on internal validation only][30]. Complementarily, a higher estimated glucose disposal rate, indicative of better insulin sensitivity, is associated with a favorable prognosis and reduced risk of post-stroke adverse outcomes, including depression, as indicated by a prospective UK Biobank study whose findings warrant further testing[31].

While traditional risk factors are easily obtainable, their limitations lie in insufficiently capturing individual heterogeneity. Emerging biomarkers, though susceptible to confounding, significantly enhance predictive sensitivity by reflecting specific biological pathways like inflammation, oxidative stress, and metabolic dysfunction. This recognition has catalyzed a paradigm shift in predictive model construction. The future direction points toward multidimensional models that integrate traditional and emerging factors using machine learning, or the development of subtype-specific models (e.g., for early-onset PSD). This trend is supported by the identification of other novel indices, such as the cardiometabolic index[32], poor sleep (based on a cross-sectional analysis of NHANES data)[33], the oxidative balance score (also derived from a cross-sectional analysis of NHANES data)[34], and the monocyte-to-high-density lipoprotein cholesterol ratio (AUC = 0.660, based on internal validation)[35], all of which contribute to a more nuanced and powerful risk assessment framework. To provide a systematic overview of these emerging biomarkers and their clinical relevance, Table 1[10,30,32-47] summarizes key predictive factors, their pathophysiological mechanisms, and implications for risk stratification in PSD. Crucially, it is important to note that the predictive performances (e.g., AUCs) reported for many of these factors, particularly those from single-center studies or models that have undergone only internal validation, may not generalize to broader, more diverse patient populations, underscoring the pressing need for external validation in future research.

Table 1 Emerging biomarkers and risk factors for prediction and stratification of post-stroke depression.

Category	Risk factor	Explanation	Data source	Reported performance (AUC)	Level of evidence (oxford)1	GRADE quality of evidence2	Ref.
Inflammatory and immunological biomarkers	Higher monocyte-to-HDL cholesterol ratio (MHR)	A pro-inflammatory/pro-atherosclerotic marker reflecting monocyte activation and HDL dysfunction; elevated MHR links systemic inflammation to PSD	Single-center	0.660 (internal validation)	3b	Very low	[35]
	Early Th1-Th2 cytokine imbalance	An early shift in Th1/Th2 ratio (e.g., increased pro-inflammatory Th1 or decreased anti-inflammatory Th2) disrupts neuroinflammation and neurotransmitter regulation-core pathophysiological mechanisms of PSD	Single-center	0.741 (internal validation)	3b	Very low	[36]
	Higher plasma putrescine and spermidine levels	These polyamines modulate oxidative stress and neuroinflammation; elevated levels indicate imbalanced cellular metabolism, contributing to mood dysregulation post-stroke	CATIS trial	Correlation reported (needs external validation)	4	Low	[37]
	Higher serum Dickkopf-1 (Dkk-1) levels	Dkk-1 inhibits the Wnt/β-catenin pathway (critical for neurogenesis/synaptic plasticity); increased Dkk-1 reduces neural repair, raising PSD risk	CATIS trial	Correlation reported (needs external validation)	4	Low	[38]
	Elevated serum growth differentiation factor 15	A stress-responsive cytokine linked to oxidative stress and mitochondrial dysfunction; higher levels predict PSD by exacerbating neuronal damage	CATIS Trial	Correlation reported (needs external validation)	4	Low	[39]
Nutritional and metabolic biomarkers	Homocysteine level	Elevated homocysteine is neurotoxic, damages blood vessels, and disrupts methylation; high levels correlate with PSD via neuronal injury and vascular dysfunction	Single-center	0.881 (internal validation)	3b	Low	[10]
	Insulin resistance	Alters glucose metabolism and promotes systemic inflammation, disrupting brain insulin signaling and contributing to post-stroke mood disorders	Single-center	0.760 (internal validation)	3b	Very low	[30]
	Higher oxidative balance score (OBS)	OBS reflects oxidative-antioxidant balance; higher scores may indicate unresolved oxidative damage to neurons, facilitating PSD	NHANES (cross-sectional)	Association reported (needs external validation)	4	Low	[34]
	Vitamin B intake	Deficiency in B vitamins (B6, B9, B12) impairs neurotransmitter synthesis (serotonin/dopamine) and methylation, increasing PSD susceptibility	NHANES (cross-sectional)	Association reported (needs external validation)	4	Low	[40]
	Lower serum BDNF levels at baseline	BDNF supports neuroplasticity; low baseline levels impair emotional regulation circuits, raising PSD risk	CATIS trial	Correlation reported (needs external validation)	4	Low	[41]
	Helicobacter pylori infection	Chronic infection triggers systemic inflammation (IL-6, TNF-α) and gut-brain axis dysregulation-both implicated in PSD pathogenesis	Single-center	Association reported (needs external validation)	4	Very low	[42]
	MMSE, NIHSS and CSVD burden score	Cerebral small vessel disease (lacunes, white matter disease) causes cumulative brain damage, cognitive decline, and mood dysregulation-associating with PSD	Single-center	0.926 (internal validation)	3b	Low	[43]
	Cardiometabolic index (CMI)	A composite of waist circumference, BMI, blood pressure, and lipids; higher CMI indicates greater cardiometabolic risk, correlating with PSD via inflammation	Single-center	Association reported (needs external validation)	4	Very low	[32]
	Non-HDL-C/HDL-C ratio (NHHR)	Atherogenic lipid profile marker; elevated NHHR associates with vascular damage and systemic inflammation, increasing PSD risk	NHANES (cross-sectional)	Association reported (needs external validation)	4	Low	[44]
	Atherogenic index of plasma (AIP)	Measures plasma atherogenicity [log (TG/HDL-C)]; higher AIP indicates increased cardiovascular risk and links to PSD via shared metabolic/inflammatory pathways	NHANES (cross-sectional)	Association reported (Needs external validation)	4	Low	[45]
	Cardiac history	Pre-stroke cardiac conditions (e.g., MI, heart failure) cause hemodynamic instability, reduced cerebral perfusion, and inflammation-predisposing to PSD	CHARLS Cohort	Association reported (needs external validation)	4	Low	[46]
Sleep and functional impairments	Sleep disorders and short sleep duration	Sleep disruption alters circadian rhythms, reduces serotonin synthesis, and increases amygdala reactivity-strong predictors of PSD	NHANES (cross-sectional)	Association reported (needs external validation)	4	Low	[33]
	Poststroke dysphagia	Difficulty swallowing leads to poor nutrition, dehydration, and social isolation-modifiable factors exacerbating post-stroke mood symptoms	Single-center	Association reported (needs external validation)	4	Low	[47]

¹Level of evidence is based on the oxford centre for evidence-based medicine levels, where 3b indicates systematic review of case-control studies or individual case-control study with non-consecutive patients, and 4 indicates case-series or cross-sectional studies. This provides a quick reference for the robustness of the evidence.

²The GRADE quality of evidence assessment further refines this by evaluating limitations, imprecision, and indirectness. The predominantly 'very low' grading underscores that most current evidence is preliminary and requires confirmation from more robust studies.

Validation status: “Internal validation” indicates model performance was assessed on the same dataset it was trained on. “Needs external validation” indicates that the factor has been identified as a significant predictor, but its predictive power within a formal model has not been tested on an independent cohort, or the study design was cross-sectional/observational without formal predictive modeling. This table highlights a key challenge in the field: While many factors show promising predictive value, the evidence for most is derived from single-center studies or requires further validation in multi-center cohorts, as emphasized by the TRIPOD/PROBAST guidelines. AUC: Area under the curve; CATIS: China Antihypertensive Trial in Acute Ischemic Stroke; CHARLS: China Health and Retirement Longitudinal Study; NHANES: National Health and Nutrition Examination Survey; BMI: Body mass index; PSD: Post-stroke depression; HDL: High-density lipoprotein; IL-6: Interleukin-6; TNF-α: Tumor necrosis factor-α; MMSE: Mini-Mental State Examination; NIHSS: National Institutes of Health Stroke Scale; CSVD: Cerebral small vessel disease; TG: Triglyceride; HDL-C: High-density lipoprotein cholesterol; MI: Myocardial infarction.

PATHOPHYSIOLOGICAL MECHANISMS

The pathophysiology of PSD is increasingly understood as a complex interplay between neuroinflammatory responses, disruptions in neuroplasticity, and large-scale network dysfunction. The core pathways involved in this intricate network are summarized in Figure 4.

Figure 4 Integrated pathophysiological network of post-stroke depression. A: Central ischemic insult. Focal brain ischemia, depicted in the hippocampus and prefrontal cortex, causes neuronal necrosis and serves as the primary trigger, releasing damage signals that initiate downstream pathological cascades; B: Neuroinflammatory Cascade. high-mobility group box 1 emerges as a pivotal hub, released from damaged neurons to activate microglia via RAGE/TLR4, driving NF-κB and NLRP3 inflammasome signaling. This sustains a cycle of M1 microglial activation and A1 astrocytic reactivity, with the latter contributing to excitotoxicity via impaired GLT-1 function; C: Neuroplasticity Impairment. A hallmark of post-stroke depression (PSD) is the downregulation of the BDNF/TrkB/CREB signaling pathway, crucial for neuronal survival and synaptic integrity. This impairment directly links cellular stress to the failure of neural circuit adaptation; D: Gut-brain axis dysregulation. Post-stroke gut dysbiosis-characterized by a loss of beneficial short-chain fatty acid-producing bacteria and an increase in opportunistic pathogens-compromises intestinal barrier function, amplifying systemic inflammation that can exacerbate central neuroinflammation; E: Macroscopic network manifestations. The molecular and cellular perturbations culminate in structural changes (e.g., white matter hyperintensities) and functional default mode network dysfunction, particularly hyperconnectivity in hubs like the posterior cingulate cortex, which underlies cognitive-emotional deficits in PSD.

Neuroinflammation represents a central pillar of PSD pathogenesis, primarily driven by the dynamic and dualistic responses of glial cells[48,49]. Upon activation following stroke, microglia polarize into a pro-inflammatory (M1) or an anti-inflammatory (M2) phenotype[49]. The M1 phenotype exacerbates neuroinflammation by releasing cytotoxic factors, while the M2 phenotype promotes repair[49,50]. This balance is critically influenced by epigenetic regulators; for instance, decreased microglial PCGF1 expression leads to activation of the NF-κB/MAPK pathway, amplifying neuroinflammation and depressive behaviors[51]. Emerging evidence firmly positions high-mobility group box 1 (HMGB1) as a pivotal hub molecule bridging initial ischemic insult to sustained neuroinflammation in PSD. As a prototypical damage-associated molecular pattern, HMGB1 is passively released from necrotic neurons and actively secreted by activated immune and glial cells post-stroke[52], where it initiates and amplifies neuroinflammatory cascades primarily by engaging the RAGE and TLR4 receptors, thereby activating downstream NF-κB signaling and the NLRP3 inflammasome[53,54]. Crucially, recent research reveals a novel mechanistic link to monoaminergic systems; HMGB1 acts as a negative modulator of the serotonin 7 (5-HT7) receptor in microglia, suppressing the neuroprotective cAMP/PKA and Nrf2/xCT/GPX4 pathways, which consequently promotes M2 microglial ferroptosis and exacerbates neuroinflammation, directly contributing to depressive behaviors[55]. Therapeutically, inhibitors targeting the HMGB1 pathway, such as glycyrrhizin[56,57], arctigenin[58], and salvianolic acid A[59], have demonstrated efficacy in ameliorating depressive-like behaviors and neuroinflammation in preclinical models by suppressing the HMGB1/TLR4/NF-κB axis. However, it is critical to note that no selective HMGB1 antagonist has yet advanced to clinical trials specifically for PSD[60], underscoring its promising yet untapped potential as a therapeutic target for interrupting the neuroimmune dialogue in PSD. Furthermore, the microglial NLRP3 inflammasome, upon activation, secretes cytokines via caspase-1 that drive astrolcytes into a neurotoxic A1 state, creating a vicious cycle of inflammation and neuronal dysfunction[61].

Similarly, astrocytes adopt reactive states that exert both detrimental and beneficial effects[49,62]. Neurotoxic A1 astrocytes, induced by microglial signals, contribute to neural damage, while neuroprotective functions include glutamate clearance[62]. Astrocytic dysfunction is a key mechanism in PSD. Impaired glutamate reuptake, due to downregulation of the astrocytic transporter GLT-1, leads to synaptic glutamate accumulation and excitotoxicity, which disrupts synaptic plasticity and is intrinsically linked to depressive pathology[63]. Furthermore, aberrant function of the astrocytic Kir4.1 channel in the lateral habenula drives neuronal hyperexcitability and bursting, directly inducing depression-like behaviors[64]. Antidepressant treatments can target these pathways; for example, vortioxetine, but not escitalopram, modulates astroglial connexin43 expression and L-glutamate release, suggesting a mechanism for its therapeutic effect[65].

Concurrently, impaired neuroplasticity forms another critical pathogenic axis. A hallmark of this is the downregulation of the BDNF/TrkB/CREB signaling pathway, which is crucial for neuronal survival and synaptic plasticity. Electroacupuncture (EA) has been shown to alleviate depressive behaviors by upregulating this pathway, highlighting its central role[66]. Dysregulation of the gut microbiota has emerged as a critical modulator of PSD via the microbiota-gut-brain axis, characterized by reduced microbial diversity, decreased beneficial short-chain fatty acid (SCFA)-producing bacteria (e.g., Bifidobacterium), and increased opportunistic pathogens, which collectively amplify systemic and neuro-inflammation through impaired gut barrier function and elevated pro-inflammatory cytokines (e.g., IL-1β, IL-6)[67-70]. Within this framework, specific pathogens like Helicobacter pylori (H. pylori) exemplify how individual microbes can perturb this homeostasis; although a community-based study suggests that H. pylori infection itself may not be an independent risk factor for depression in the general population, with its effect potentially mediated through dyspepsia symptoms[71], a clinical study specifically in stroke patients identified H. pylori infection as an independent risk factor for PSD, associated with more severe depressive symptoms, significant alterations in gut microbiota composition (e.g., reduced SCFA concentrations), and heightened systemic inflammation[42]. This indicates that in the context of stroke, H. pylori may exacerbate PSD pathogenesis by inducing a deleterious shift in the microbial ecosystem, thereby serving as a specific pathobiont within the broader dysbiosis landscape. Interventions such as fecal microbiota transplantation or prebiotics (e.g., arabinoxylan) that restore overall microbial balance have shown efficacy in alleviating PSD-like behaviors in preclinical models by modulating inflammatory signaling and promoting neuroprotective pathways[67,69,72], underscoring the necessity of targeting the gut microbiome as a whole for future therapeutic strategies.

Finally, these molecular and cellular perturbations manifest at the macroscopic level as structural and network dysfunction. Cerebral small vessel disease, evident as white matter hyperintensities, is an independent predictor of post-stroke cognitive impairment, though its direct link to depression requires further validation[73]. At a functional level, resting-state functional magnetic resonance imaging studies reveal that PSD is associated with altered functional connectivity within the default mode network, characterized by increased centrality and connectivity in hubs like the angular gyrus and posterior cingulate cortex, reflecting a disruption in networks subserving self-referential thought and emotion regulation[74].

SCREENING AND DIAGNOSIS

The accurate identification of PSD is a critical step in initiating timely intervention, yet it remains a clinical challenge due to symptom overlap with other post-stroke conditions. The diagnostic process primarily relies on a combination of standardized assessment tools and careful clinical evaluation to differentiate PSD from its mimics.

The accurate identification of PSD relies on standardized assessment tools, which can be broadly categorized into self-report questionnaires and clinician-administered instruments. The Patient Health Questionnaire-9 (PHQ-9) is a prominent example of the former, while the Hamilton Depression Rating Scale (HDRS) exemplifies the latter[75]. The diagnostic accuracy of these tools has been systematically evaluated. A meta-analysis by Liu et al[75] indicated that the PHQ-9 demonstrates high diagnostic utility for any depression, with a sensitivity of 0.82, specificity of 0.87, and a diagnostic odds ratio of 29. The HDRS shows even higher performance for major depression, with a sensitivity of 0.92, specificity of 0.89, and a diagnostic odds ratio of 94. Both scales maintain high diagnostic accuracy in assessing depressive symptoms during both acute and chronic phases of stroke. It is noteworthy, however, that systematic screening during acute hospitalization identifies few patients with depression. As demonstrated by Shankar et al[76], using the PHQ-2 (the brief version of PHQ-9) during acute stroke admission yielded a positive screening rate of only 4.7%, whereas follow-up assessment at outpatient visits revealed a much higher prevalence of 19.3%. This significant discrepancy highlights the potential for under-detection in the immediate post-stroke period and suggests that the optimal timing for PSD screening may be after the acute hospitalization phase.

Beyond the core symptoms of depression, stroke survivors frequently experience a range of other neuropsychiatric sequelae, such as fatigue, sleep disturbances, irritability, and anxiety[77]. Consequently, a crucial aspect of clinical practice is the differential diagnosis of PSD from other post-stroke complications. Post-stroke fatigue is a distinct and often overlooked condition that can occur during the acute, rehabilitation, and chronic phases[78]. Its hallmark is an early sense of exhaustion upon physical or mental activity that is not effectively relieved by rest, with a reported prevalence ranging widely from 25% to 85%[79]. Post-stroke apathy, characterized by a marked reduction in goal-directed behavior across cognitive, emotional, and social dimensions, affects approximately 30% of stroke patients[80]. Accurately differentiating PSD from post-stroke fatigue, apathy, and other conditions is paramount for ensuring the administration of appropriate and targeted interventions.

THERAPEUTIC AND MANAGEMENT STRATEGIES

The management of PSD has evolved into a multimodal paradigm, encompassing pharmacotherapy, neuromodulation, and integrative medicine approaches, all aimed at addressing its complex pathophysiology.

Pharmacological interventions: Targeting monoamines and beyond

First-line pharmacological treatment for PSD involves SSRIs, which are among the primary therapeutic options. A systematic review of RCTs indicates that escitalopram and sertraline demonstrate comparable efficacy in alleviating anxiety symptoms, improving cognition, and restoring activities of daily living in PSD patients[81,82]. However, escitalopram may offer superior efficacy in relieving core depressive symptoms, with a significant reduction in depression scores observed as early as the first week of treatment. It is critical to note that both agents are associated with a substantial risk of hyponatremia, affecting approximately one-third of patients, though most cases are mild with no significant difference in incidence between the two drugs[83]. SSRI therapy is linked to significant enhancements in post-stroke anxiety, dependency, motor function, and cognition[84]. Notably, treatment with citalopram has been shown to significantly reduce the risk of stroke recurrence and cardiovascular events, underscoring its potential neuroprotective benefits[85].

Agomelatine, a melatonergic agonist, demonstrates antidepressant efficacy comparable to SSRIs such as sertraline and escitalopram in elderly patients with PSD, with a favorable safety and tolerability profile[86]. When combined with EA, agomelatine shows superior improvement in sleep efficiency and cognitive function compared to monotherapy for post-stroke insomnia, suggesting potential synergistic effects[87]. However, as an adjunctive therapy to SSRIs or serotonin-norepinephrine reuptake inhibitors, agomelatine does not exhibit significant augmentation effects on depressive symptoms, despite maintaining a favorable safety profile[88].

Given the established role of neuroinflammation in PSD, interventions targeting associated pathways present novel strategies[89]. For instance, phototherapy significantly reduces pro-inflammatory cytokines, including tumor necrosis factor-α (TNF-α), and alleviates depressive symptoms while improving cognitive function in PSD patients[90]. This suggests that modulation of inflammatory responses may represent a viable treatment approach.

Vitamin D deficiency is highly prevalent in acute stroke patients and is independently associated with the development of PSD[91]. Lower serum vitamin D levels (e.g., ≤ 37.1 nmol/L) significantly increase the risk of PSD at one month, whereas higher levels (≥ 64.1 nmol/L) are protective[91]. Similarly, serum 25-hydroxyvitamin D levels ≤ 11.2 ng/mL at admission are independently associated with depression at six months post-stroke[92]. Preclinical studies suggest that vitamin D3 supplementation mitigates depressive-like behaviors by upregulating hippocampal BDNF signaling[93]. Therefore, vitamin D supplementation represents a crucial preventive and therapeutic measure for PSD.

Traditional Chinese medicine: A systems biology approach

Traditional Chinese medicine (TCM) interprets the core pathology of PSD as obstructed flow of vital energy (Qi) and impaired circulatory function. Stroke-induced stagnation disrupts these pathways, and depression exacerbates the imbalance, creating a vicious cycle that worsens physical and emotional symptoms[94]. Several well-known formulas have shown efficacy. Chaihu Shugan San (CHSGS), Danzhi Xiaoyao San (DZXYS), and Shugan Jieyu Capsule (SGJYC) are prominent examples[95]. CHSGS is composed of seven herbs and is believed in modern pharmacology to promote gastric emptying, regulate gastrointestinal motility, improve depression, and reduce inflammation[96]. Based on network pharmacology analysis, the active components of CHSGS are predicted to exert antidepressant effects by modulating key signaling pathways involved in neurotransmission, including inflammatory mediator regulation of TRP channels, calcium signaling, and cAMP signaling[97]. Furthermore, a systematic review and meta-analysis of animal studies confirms that Chaihu Shugan powder treatment significantly alleviates depressive-like behaviors and is associated with mechanisms involving anti-inflammatory effects and regulation of the hypothalamic-pituitary-adrenal axis[98]. Fan et al[50] demonstrated that CHSGS treatment downregulates STAT3 and PTEN while upregulating GSK3β, indicating modulation of the STAT3-GSK3β/PTEN pathway. DZXYS, a classic formula for mental disorders, can effectively improve depressive symptoms by regulating serotonin, BDNF, cortisol, and IL-6 levels[99]. It also modulates the metabolism of phenylalanine, arachidonic acid, porphyrin, D-arginine, and D-ornithine, and regulates steroid biosynthesis and unsaturated fatty acid biosynthesis, increasing excitability[100]. Wu et al[101] reported that modified DZXYS can reduce Hamilton Depression Scale (HAMD) scores and, experimentally, inhibit microglial M1 polarization and alleviate neuroinflammation by enhancing autophagy via the PI3K/Akt/mTOR pathway. SGJYC primarily alleviates low mood, mental anxiety, somatic anxiety, and autonomic dysfunction, and is most commonly used in combination with paroxetine, sertraline, or fluoxetine[102]. Wang et al[103] found it relieves liver-qi depression by regulating the ERK-CREB-BDNF pathway. A systematic review of 15 RCTs (1240 patients) by Feng et al[104] indicated the SGJYC group had higher total response rates and HAMD scores than the control group (e.g., paroxetine, citalopram) with a better adverse event profile. To provide a systematic overview of the active ingredients and molecular mechanisms underlying the efficacy of these TCM formulations, Table 2[50,103,105-111] summarizes key compounds, their targets, and the signaling pathways involved in alleviating PSD, offering a deeper insight into the pharmacologica basis of traditional remedies.

Table 2 Pharmacological mechanisms of selected traditional Chinese medicine formulations in post-stroke depression.

Prescription/preparation	Active ingredients	Mechanism of action	Level of evidence (oxford)1	GRADE quality of evidence2	Ref.
Shuyu capsules relieve liver	Adhyperforin, etc.	Reversing the disruptions of the p-ERK, p-CREB and BDNF	5	Not applicable (pre-clinical)	[103]
Chaihu Shugan powder	Seven Chinese herbs including bupleurum	Chaihu-Shugan-San inhibits neuroinflammation in the treatment of post-stroke depression through the JAK/STAT3-GSK3β/PTEN/Akt pathway	5	Not Applicable (pre-clinical)	[50,105]
Danzhi Xiaoyao San	Eight Chinese herbs including Paeoniae Radix Alba	Reducing neuroinflammation through PKCγ/p38/NF-κB signaling pathway	5	Not applicable (pre-clinical)	[106,107]
Jiao-tai-wan (JTW)	Jatrorrhizine	JTW could exert antidepressant effects by modulating neuroinflammation via inhibition of the STING pathway	5	Not applicable (pre-clinical)	[108]
Flavones	Apigenin, etc.	Flavones exert protective effects against depression in mice, primarily by stimulating neurotrophic factors and modulating inflammatory pathways	5	Not applicable (pre-clinical)	[109]
Dendrobium officinale	Dendrobine (DEN) and erianin (ERI)	DEN and ERI alleviated LPS-induced microglial activation and neuroinflammation by binding to PDE4B and preventing TLR4/NF-κB signaling pathway	5	Not applicable (pre-clinical)	[110]
Corydalisyanhusuopolysaccharides (CYP)	CYP	Corydalis yanhusuo polysaccharides regulates HPA-axis mediated microglia activation and inhibits astrocyte A1 transformation to improve depression-like behavior	5	Not applicable (pre-clinical)	[111]

¹Oxford Centre for Evidence-Based Medicine 2011 levels of evidence: Level 5 refers to mechanism-based reasoning or expert opinion, which encompasses findings from experimental/mechanistic studies that form the basis for future clinical research.

²GRADE assessment: The evidence summarized in this table is derived from preclinical studies elucidating potential mechanisms of action. As such, it is classified as mechanistic exploration and a formal GRADE rating (high, moderate, low, very low) is not applicable (N/A). This evidence is crucial for understanding biological plausibility but requires validation in human clinical trials before any conclusions regarding clinical efficacy can be drawn.

HPA: Hypothalamic-pituitary-adrenal; LPS: Lipopolysaccharide.

Neuromodulation and non-pharmacological therapies

Repetitive transcranial magnetic stimulation (rTMS) is an effective neuromodulation technique. Its modalities include high-frequency rTMS (HF-rTMS, > 1 Hz), low-frequency (≤ 1 Hz), theta-burst stimulation, and bilateral rTMS. By delivering magnetic pulses to specific cortical areas, particularly the DLPFC, rTMS modulates emotion-regulation networks, improves neural circuit function, and promotes neuroplasticity. HF-rTMS applied to the left DLPFC can improve depressive symptoms in the subacute phase of subcortical ischemic stroke. rTMS demonstrates significant therapeutic potential for PSD. Specifically, HF-rTMS targeting the left DLPFC significantly improves depressive symptoms in patients with subcortical ischemic stroke during the subacute phase[112]. The treatment effect may be predicted by specific depressive symptom dimensions at admission. Furthermore, rTMS treatment can promote the synthesis and release of monoamine neurotransmitters (DA, NE, 5-HT), regulate inhibitory/excitatory amino acid neurotransmitters (Gly, Glu), reduce inflammatory responses (IL-1β, IL-6, TNF-α), and consequently improve clinical outcomes and enhance immune function in PSD patients[113].

Acupuncture demonstrates potential therapeutic effects for PSD, though the evidence quality requires cautious interpretation. An overview of systematic reviews indicated that acupuncture may improve depressive symptoms, stroke-related symptoms, and activities of daily living in PSD patients, but the majority of systematic reviews were of very low quality due to methodological limitations[114]. Specifically for EA, a meta-analysis of RCTs demonstrated its association with reduced scores on the HAMD, Self-Rating Depression Scale, and TCM Depression Scale, along with a lower incidence of adverse events compared to control groups[115]. Furthermore, a RCT by Xie et al[116] investigating mind-regulating acupuncture based on the “microbiota-gut-brain axis” theory reported increases in serum 5-HT and BDNF levels and modulation of gut microbiota composition. While these findings are promising, further high-quality research is necessary to conclusively establish the efficacy and safety of acupuncture for PSD.

Non-traditional interventions like music and exercise therapy are gaining prominence. Music therapy facilitates emotional release and improves mental health by directly stimulating the hypothalamus and limbic system[117]. Exercise promotes the release of neurotransmitters like dopamine. Mao et al[118] pointed out that multimodal exercise (MME) combines different types of exercise involving multiple senses. Supportive music and imagery helps patients create mental images while listening to music, presented through writing or painting. Combining MME with supportive music and imagery can effectively divert patient attention, promote the secretion of enzymes, stimulate the vagus nerve, alleviate anxiety and depression, and reduce post-stroke psychological distress. To consolidate the evidence from recent RCTs on various PSD treatments, including neuromodulation and non-pharmacological interventions, Table 3[8,81,83,86,119-134] provides a comprehensive summary of key findings from the past five years, highlighting efficacy, safety, and functional outcomes.

Table 3 Evidence from recent randomized controlled trials for post-stroke depression interventions.

Intervention approach	Ref.	Study design	Year	Sample size	Evidence summary and critical appraisal	Key limitations/uncertainties	Evidence maturity	Level of evidence (oxford)1	GRADE quality of evidence2
Acupuncture	Kalaoğlu et al[8]	RCT	2024	54 patients	The study was negative on its primary outcomes. While secondary outcomes showed signal of potential benefit, the results are inconclusive	Very small sample size; high risk of performance bias; primary outcome was not met	Pilot	1	Very low
SSRIs (escitalopram vs sertraline)	Yan and Hu[81]	RCT	2024	60 patients	Suggests comparable efficacy between escitalopram and sertraline for anxiety and functional outcomes in PSD. However, the difference in HAMD scores was not consistent across timepoints	Small sample size; short follow-up period; some outcome measures showed no significant difference	Early phase	1	Low
SSRIs (escitalopram and sertraline)	Naseralallah et al[83]	RCT	2024	401 patients	Focuses on safety, finding both drugs are associated with a comparable and increased risk of mild hyponatremia	Study was not primarily designed to assess efficacy for depression; highlights an important adverse effect	Early phase (for safety profile)	1	Moderate (for safety outcome)
Electroacupuncture vs escitalopram	Ma et al[119]	RCT	2024	150 participants	Suggests electroacupuncture may be non-inferior or superior to escitalopram at 10 weeks for mild-to-moderate PSD, with modulatory effects on inflammation	Unblinded design (high risk of performance bias); specific mechanisms remain exploratory; requires larger-scale replication	Early phase	1	Low
Pharmacotherapy (Edaravone Dexborneol)	Xu et al[120]	RCT	2024	93 patients	Suggests potential for preventing early PSD and modulating inflammatory cytokines	Single, relatively small RCT; mechanism is exploratory; requires confirmation in larger trials focused on depression treatment	Pilot	1	Low
SSRI (fluoxetine)	Tay et al[121]	RCT	2023	1500 participants	A large RCT found fluoxetine reduced depressive scores but increased apathy scores, demonstrating a differential effect on depressive vs apathetic symptoms	Complex effects (beneficial for depression but potentially detrimental for apathy); risk-benefit profile needs careful consideration	Established but with trade-offs	1	Moderate
SSRI (sertraline)	Stuckart et al[122]	RCT	2021	114 patients	Suggests a potential benefit for preventing incident depression and functional recovery, but the groups were not balanced at baseline	Non-randomized comparison (major limitation); baseline severity differed between groups, confounding results	Pilot	1	Very low
SSRIs vs nootropics	Arcadi et al[123]	RCT	2021	44 patients	SSRIs showed a large effect size for depression/anxiety compared to nootropics in a very small sample	Very small sample size; preliminary nature of findings	Pilot	1	Very low
Agomelatine vs SSRIs	Yao et al[86]	RCT	2021	165 patients	Agomelatine, sertraline, and escitalopram were all more effective than control, with no significant differences between them	Lacks a placebo control; unable to establish absolute efficacy vs no treatment	Early phase	1	Moderate
Bright light therapy + escitalopram	Xiao et al[124]	RCT	2020	106 patients	Combination therapy improved sleep and depression scores more than escitalopram monotherapy	Single study; unblinded design for light therapy; focuses on PSD with insomnia subtype	Pilot	1	Low
Vitamin D (risk factor)	Tan et al[125]	Meta-analysis	2025	3537 patients (7 studies)	Observational data links low vitamin D levels in acute stroke phase to higher risk of developing PSD	Evidence is associative, not interventional; does not prove causation or treatment efficacy	Early phase (for association)	1	Low (for association)
Agomelatine vs SSRIs/SNRIs	Chen et al[126]	Meta-analysis	2024	857 patients (9 studies)	Agomelatine was comparable to SSRIs/SNRIs for depression reduction but showed better functional improvement (Barthel Index) and safety profile	Limited to short-term (6-12 weeks) studies; majority of included RCTs may be of moderate quality	Early phase	1	Moderate
SSRI (Fluoxetine)	Wu et al[127]	Meta-analysis	2023	6584 patients (14 RCTs)	Meta-analysis indicates fluoxetine reduces depression/anxiety risk but did not improve functional outcomes (e.g., mRS, BI) compared to placebo	Lack of benefit on key functional scales; mixed evidence profile	Mixed quality	1	Moderate
SSRIs (Prevention)	Zhou et al[128]	Meta-analysis	2021	5370 patients (10 RCTs)	Early SSRI use reduces the risk of PSD occurrence, but did not improve functional independence	No functional benefit despite preventing depression; significant heterogeneity in some analyses	Mixed quality	1	Moderate
SSRIs (prevention)	Richter et al[129]	Meta-analysis	2021	6560 patients (6 RCTs)	Confirms early SSRIs reduce PSD incidence, but at the cost of increased risk of bone fractures and nausea	Clear trade-off between benefit and harms (fractures, nausea)	Established but with trade-offs	1	High
SSRI (paroxetine)	Li et al[130]	Meta-analysis	2020	212 patients (4 studies)	Finds no significant advantage of paroxetine over control treatments, indicating limited evidence for its use	Very limited number of small studies; fails to establish efficacy	Mixed quality (leaning towards ineffective)	1	Low
Animal studies	Dong et al[131]	Animal experiment	2025	Rats model	Suggests a potential mechanism for vortioxetine in improving post-stroke recovery in rats	Pre-clinical evidence; direct applicability to human PSD patients is unknown	Pilot (pre-clinical)	5	N/A (preclinical)
Animal studies	Zhang et al[132]	Animal experiment	2025	Rats model	Proposes a novel mechanism for cilostazol in preventing PSD in mouse models	Pre-clinical evidence; requires validation in human trials	Pilot (pre-clinical)	5	N/A (preclinical)
Animal studies	Wei et al[133]	Animal experiment	2022	Rats model	Explores the mechanism of fluoxetine on neuronal differentiation in rat models	Pre-clinical evidence; mechanistic study	Pilot (pre-clinical)	5	N/A (preclinical)
Animal studies	Shyu et al[134]	Animal experiment	2021	Rats model	Investigates antidepressants for post-stroke pain and comorbid depression in rats	Pre-clinical evidence; focuses on a specific pain comorbidity	Pilot (pre-clinical)	5	N/A (preclinical)

¹Evidence maturity: Labels such as ‘pilot’ or 'early phase' describe the developmental stage of the evidence, indicating the need for further validation or larger studies. Level of evidence (Oxford Centre for Evidence-Based Medicine): This hierarchy is based primarily on study design, where level 1 represents the highest quality (e.g., randomized controlled trials, meta-analyses).

²GRADE quality of evidence: This dynamic assessment starts with the study design and then adjusts the quality rating (high, moderate, low, very low) based on limitations, imprecision, inconsistency, and other factors, providing a more nuanced estimate of confidence in the evidence.

PSD: Post-stroke depression; SSRI: Selective serotonin reuptake inhibitor; N/A: Not applicable; RCT: Randomized controlled trial; HAMD: Hamilton Depression Rating Scale.

Given the complexity of PSD pathophysiology and the array of available interventions, translating this evidence into clear clinical practice requires a structured and evidence-graded framework. To this end, we propose an integrated clinical management algorithm that synthesizes the core principles of prediction, mechanism-based stratification, and personalized intervention. This algorithm, detailed in Figure 5, provides a step-by-step visual guide for clinicians. To ensure transparency and facilitate deeper scrutiny of the evidence underpinning each recommended intervention within the algorithm, a comprehensive summary of the supporting data, including detailed study designs, sample sizes, and evidence grades assessed using the GRADE framework, is provided in the Supplementary Table 1. Together, these resources offer a practical toolkit, guiding the clinician from initial patient assessment in the acute phase through risk stratification to the selection of evidence-based therapies tailored to individual patient profiles.

Figure 5 A “predict-risk stratify-intervene” clinical management algorithm for post-stroke depression. This integrated decision-support tool translates post-stroke depression pathophysiology into a structured, time-sensitive clinical pathway. It begins with multimodal risk assessment, proceeds to mechanism-based subtyping, and culminates in subtype-specific interventions annotated with GRADE-derived evidence grades. The algorithm emphasizes the integration of traditional clinical markers, emerging biomarkers, and advanced analytics for precision management, with evidence ratings directly informing therapeutic confidence.

FUTURE DIRECTIONS AND A FULL-CHAIN TRANSLATIONAL PARADIGM

Substantial advances in neuropsychiatry and psychology have deepened our understanding of PSD, and existing therapeutic modalities demonstrate encouraging efficacy. However, a more profound dissection of its pathophysiology is imperative to address unresolved challenges. Critically, PSD arises from the complex interplay of biological, psychological, and social factors, often amplified by comorbidity and disability. Therefore, future research and clinical practice must consistently report functioning and quality-of-life outcomes alongside symptom assessment, adopting a brief “multidimensional profile” (encompassing biological, psychological, social, and functional domains) to guide risk stratification and personalized care.

To this end, future investigations should operationalize a “full-chain” translational paradigm into measurable deliverables. First, the priority is to develop and validate a multicenter, externally validated prediction model that integrates multi-omics data and artificial intelligence (AI). Second, this should be followed by conducting subtype-stratified RCTs based on inflammatory, metabolic, and gut-brain axis signatures to test targeted interventions. Third, there is a need to establish standardized neuromodulation protocols, such as rTMS with active sham comparators, to ensure efficacy and reproducibility. Fourth, future trials must predefine comprehensive endpoint bundles, including objective biomarker endpoints (e.g., BDNF, inflammatory panels) alongside patient-reported outcomes, to fully capture intervention effects.

The application of advanced experimental systems, such as brain organoids and microfluidic platforms, will enable more accurate modeling of PSD pathogenesis, effectively bridging translational gaps between animal studies and human clinical data. This refined mechanistic knowledge will facilitate precision medicine approaches, including patient stratification based on inflammatory-metabolic-gut signatures and the personalization of TCM through alignment of herbal formulations with individual metabolic markers and genetic profiles. Concurrently, the growing emphasis on neuroimmune mechanisms and the gut-brain axis promises to reveal novel therapeutic opportunities. Probiotic supplementation and targeted dietary interventions represent promising non-pharmaceutical strategies for modulating these pathways.

Several other critical research priorities warrant continued emphasis. The search for genetic determinants of PSD susceptibility remains crucial. The current reliance on subjective clinical scales further underscores the need to validate objective biomarkers in accessible biofluids. Furthermore, non-pharmacological interventions require rigorous evaluation to optimize efficacy. Key clinical questions, such as the independent contribution of PSD to mortality and its interactions with other post-stroke sequelae (e.g., cognitive impairment, anxiety), demand clarification to develop integrated treatment protocols. The establishment of comprehensive long-term management strategies is critical for improving functional recovery and quality of life. Particular focus should be directed toward vulnerable populations. Ultimately, the validation of these next-generation strategies will depend on conducting large-scale, multi-center, international RCTs.

LIMITATIONS

Notwithstanding the progress summarized herein, the current body of research on PSD is subject to several important limitations. A primary constraint is the scarcity of predictive models that have undergone rigorous external validation. Furthermore, the field lacks large-scale, globally representative cohort studies, which are essential for generating findings with broad generalizability.

Research on TCM formulations faces specific challenges. The considerable heterogeneity and highly individualized nature of TCM interventions complicate the standardization of clinical trials and the interpretation of results. Wider global promotion of TCM research, coupled with studies incorporating larger, more diverse subject populations, is necessary to refine optimal herbal prescriptions and more fully evaluate their therapeutic potential. While certain risk factors such as stroke severity and cognitive impairment are well-established, the associations of other factors, including age and sex, with PSD remain controversial and require further clarification. The evidence base for non-pharmacological interventions, including music therapy, is still nascent, constrained by a paucity of large-scale, multi-center, international RCTs. The translational validity of animal models for PSD is another concern, as the equivalence between depression-like behaviors in rodents and the complex human condition of depression is uncertain, potentially limiting the clinical relevance of preclinical findings.

Finally, while key pathophysiological mechanisms are emerging, a more profound understanding of the intricate neurobiological cascades underlying PSD is imperative. Future research would benefit significantly from enhanced integration across neurological, psychiatric, psychological, and biological disciplines to address these multifaceted limitations comprehensively.

CONCLUSION

In summary, the field of PSD research is undergoing a pivotal transformation, moving beyond phenomenological description towards a mechanistic and data-driven discipline. The integration of multi-omics with AI is refining our ability to predict risk, while the elucidation of neuro-immune-inflammatory and gut-brain axes provides a more holistic pathophysiological framework. These advances collectively validate the “predict-mechanism-intervene” paradigm, which emphasizes a continuous translational cycle from risk stratification to mechanism-based therapeutic targeting. The future management of PSD therefore lies in leveraging these interconnected strategies to achieve personalized, timely, and effective interventions, ultimately improving long-term outcomes for stroke survivors.