1) Study Population

Diagnostic Criteria: Stroke diagnosis must comply with the 2024 AHA/ASA Guidelines for the Primary Prevention of Stroke (38) or the Chinese Guidelines for Cerebrovascular Disease Prevention and Treatment (2024) (39).

Emotional Disorder Diagnosis: Depression: Meets DSM-5/ICD-10 criteria for depressive episode, with SDS score ≥ 53 or HAMD-17 score ≥ 8; Anxiety: GAD-7 score ≥5 or HAMA score ≥ 7; Mixed Type: Concurrent depressive and anxiety symptoms with HADS mean score ≥ 8.

Disease Course: Emotional disorders emerging between 2 weeks and 1 year post-stroke.

Age: 18–75 years, no gender restrictions.

2) Interventions

TCM Intervention Types: Chinese Herbal Medicine: Compound formulas; single herbs; proprietary Chinese medicines;

Acupuncture: Body acupuncture; scalp acupuncture; electroacupuncture;

Other Therapies: Qigong; Tai Chi; Baduanjin; Five-Element Music therapy;

Combination Therapies: TCM integrated with Western medicine; multimodal TCM approaches.

Intervention Details: Treatment frequency and duration must be reported; herbal formulas require specification of decoction methods and dosages.

3) Comparator Groups

Control Types:

Active Control: Conventional Western therapy (e.g., pharmacotherapy, cognitive behavioral therapy).

Negative Control: Placebo (e.g., sham acupuncture, herbal placebo).

No-Intervention Control: Basic stroke care only (e.g., aspirin, rehabilitation).

Permitted Co-Interventions: Control groups may receive foundational stroke treatments (e.g., antiplatelet agents).

4) Outcome Measures

Primary Outcomes:

Clinical Response Rate: Total effective rate; (To control for heterogeneity, only studies that explicitly define the total effective rate will be included. The definition must be based on standardized scales recognized in this study and must clearly report the thresholds for “marked improvement” and “improvement.” Studies using self-defined criteria, vague definitions, or failing to report the specific calculation method will be excluded).

Standardized Scale Scores: Pre-post treatment differences in HAMD-17, HAMA, SDS, GAD-7, or HADS scores.

Secondary Outcomes:

Quality of Life: Changes in scores assessed by standardized QoL instruments, including but not limited to the 36-Item Short Form Health Survey (SF-36), the Stroke-Specific Quality of Life Scale (SS-QOL), or the WHO Quality of Life-BREF (WHOQOL-BREF).

5) Study Design

Design: Parallel-group randomized controlled trials (RCTs) with intervention duration ≥4 weeks.

Sample Size: ≥20 participants per group to minimize bias.

Publication Period: Studies published between January 2000 and May 2025 (excludes methodologically immature early studies and obsolete therapies).

Language: Chinese or English publications.

1) Study Population

Severe cognitive impairment or aphasia precluding assessment cooperation;History of psychiatric disorders (e.g., schizophrenia, bipolar affective disorder);Substance abuse (e.g., alcohol dependence);Pre-existing unresolved emotional disorders prior to stroke onset (to avoid confounding efficacy evaluation).

2) Interventions

Non-traditional TCM methods (e.g., infrared “acupuncture,” unvalidated device-based therapies);Interventions lacking TCM theoretical basis (e.g., proprietary Chinese medicines administered without syndrome differentiation).

3) Outcome Measures (Outcomes)

Sole use of non-validated or self-designed assessment tools; Failure to distinguish stroke-related emotional disorders from other complications (e.g., fatigue, pain) in outcome reporting.

4) Study Design

Non-controlled studies; Duplicate publications or inaccessible full texts.

This study will implement a standardized literature management workflow: Retrieved citations will undergo automated and manual deduplication in EndNote 2025, with the deduplicated library subsequently uploaded to the Covidence platform (40). Within this system, two independent reviewers will conduct preliminary title/abstract screening for all records; a randomly selected 10% subset will undergo dual independent screening to quantify inter-rater agreement via Kappa statistics (target κ≥0.8), with discrepancies resolved through consensus discussion, while the remaining records will be screened independently. Studies passing initial screening will proceed to full-text assessment, wherein corresponding authors of publications with incomplete data will be contacted for supplementation; all exclusion decisions will be categorically documented in Covidence per PRISMA-P standards, automatically generating auditable exclusion records. The platform’s built-in functionality will ultimately export a PRISMA 2020-compliant flow diagram to dynamically visualize the screening process (41) (provisional diagram: Figure 1). For data extraction, customized Covidence electronic forms will facilitate dual independent data entry, automatically flagging discordant entries for arbitration until consensus is achieved, ensuring dataset accuracy and methodological transparency.

Dual Independent Extraction: Two reviewers will independently extract data using a structured extraction template encompassing:

Study Characteristics: Authors, publication year, journal, country, funding sources.

Design Information: Intervention types, comparator groups, total study duration, diagnostic criteria.

Sample Characteristics: Sample size per group, age/gender/racial composition.

Outcome Metrics: Baseline and post-intervention scale scores (HAMD-17, HAMA, SDS, GAD-7, HADS, SF-36, SS-QOL, WHOQOL-BREF).

TCM Syndrome Patterns: With reference to the TCM syndrome classification in the Guidelines for the Integrated Clinical Diagnosis and Treatment of Mental Disorders - Post-Stroke Mental Disorders (42), a standardized syndrome reference table will be developed to serve as the basis for data extraction and subsequent subgroup analysis, as shown in Table 3.

Follow-up Data: Scale scores categorized by endpoint: short-term (≤90 days), mid-term (91–364 days), and long-term (>364 days).

Consistency Validation: Inter-rater agreement will be quantified using the Kappa statistic (κ≥0.8 required). If the naming of syndrome patterns in the included studies varies, two TCM clinical experts will perform a consensus classification based on the above diagnostic criteria, preferentially assigning them to the syndrome type with the highest match to core symptoms. Studies that cannot be clearly classified will be included in an “other syndrome patterns” subgroup or excluded. In case of any other disagreements, a third reviewer will arbitrate to reach a consensus.

All extracted data will undergo cross-verification and organization within the Covidence literature management platform, with final exportation in structured.csv format to support subsequent statistical analysis.

Intervention group classification will be completed during the data extraction phase. The intervention groups in each trial will be categorized according to the following classes: the TCM interventions will be classified as follows: Chinese herbal compound formulas with clearly defined compositions and syndrome differentiation basis; acupuncture therapies conforming to the STRICTA reporting guidelines (43); integrated TCM therapies; and other TCM therapies. Control condition types are classified as: conventional Western medicine treatment, where the categories of anti-affective disorder drugs included in this study are based on evidence-based therapeutic agents recommended by the American Psychiatric Association medication guidelines (44), including selective serotonin reuptake inhibitors (SSRIs, e.g., sertraline, fluoxetine), serotonin-norepinephrine reuptake inhibitors (SNRIs, e.g., venlafaxine), and tricyclic antidepressants (TCAs, e.g., amitriptyline), etc.; placebo control; and no-treatment control. If classification discrepancies cannot be resolved by a third-party reviewer, consultation will be sought with the project advisory group, comprising TCM clinical experts and psychiatric specialists, to determine the final classification through consensus.

This study will generate a network plot to visualize the available evidence structure. This plot will depict all possible direct comparisons between interventions, with the thickness of the connecting lines between nodes, or numerical annotations, reflecting the number of included studies or sample size for each comparison pair. The reference node will be set as the active control condition, and different intervention combinations will be presented as distinct nodes (45). This figure will intuitively display the connectivity of the evidence network and the comparison pathways, providing a topological basis for subsequent statistical analysis. To address the potential issue of sparse network connectivity resulting from overly granular classification, we will adhere to the principle of “clinical similarity first.” Based on clinical criteria of “similarity in therapeutic principle” and “similarity in herbal composition or acupoint selection,” two reviewers will independently assess and subsequently submit cases for arbitration by an expert advisory panel to merge similar interventions into a single node.

Studies passing full-text screening will undergo risk of bias assessment. This assessment will be independently performed by two reviewers using the Cochrane Risk of Bias tool on the Covidence platform (46). The tool assesses key domains including the method of random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessors, completeness of outcome data, and risk of selective outcome reporting.

The summary measures for this study will report scores for anxiety, depression, comorbid conditions and quality of life. For studies employing continuous data, the effect size between treatment groups will be presented as the mean difference (MD). The mean difference will be calculated as the follow-up score minus the baseline score (with some studies directly reporting the difference). Any study unable to provide any two of the following—baseline data, change values, or follow-up data—will be excluded. Furthermore, studies failing to report the standard deviation (SD) of the score change or the follow-up SD (or where this cannot be calculated from confidence intervals or standard errors) will also be excluded. These criteria aim to reduce the risk of bias while maintaining the informational integrity of the evidence network. Data will also be summarized via treatment rankings and the surface under the cumulative ranking curve (SUCRA) to aid in interpreting the relative effectiveness of all intervention types within the network.

To validate the effectiveness of indirect comparisons within the network meta-analysis (NMA) and ensure the scientific validity and clinical plausibility of comparative results across different interventions, a table of trial characteristics that may act as effect modifiers will be compiled from the collected data. This table will include: ① Stroke-related factors (severity, type); ② PSED-related factors (time of onset, subtype); ③ Patient-related factors (age, sex, cognitive function); and ④ Co-interventions. By analyzing whether the distribution of potential effect modifiers is balanced across studies, this approach will assess whether these factors introduce heterogeneity that interferes with treatment effects, thereby supporting the evaluation of the transitivity assumption (47, 48). If clinically significant imbalances in key factors are identified, this will be explicitly noted in the interpretation of indirect comparisons and will serve as the basis for conducting pre-planned meta-regression or subgroup analyses.

The primary objective of data synthesis in this study is to compare the efficacy differences among different Traditional Chinese Medicine (TCM) therapies. As network meta-analysis (NMA) can compensate for the lack of direct evidence through indirect estimations, it is suitable for achieving this objective. Currently, no head-to-head clinical trials have been conducted between distinct TCM therapies; therefore, studies will be pooled for analysis according to the intervention group classifications defined above. If quantitative synthesis of study results is infeasible, findings will be presented through a descriptive systematic review. All quantitative analyses will be performed using Stata 18 and RStudio software.

If included studies provide direct comparison data between different Traditional Chinese Medicine (TCM) interventions, an exploratory pairwise meta-analysis will be performed. A random-effects model will be employed for analysis, with forest plots used to display individual study results and pooled effect sizes; separate independent analyses will be conducted for each primary outcome measure. Publication bias and small-study effects will be assessed using funnel plots and Egger’s regression test. If significant bias is detected (Egger’s test P < 0.10), the trim-and-fill method will be applied to adjust effect size estimates (49).

Given that this study examines the effectiveness of diverse Traditional Chinese Medicine (TCM) interventions for post-stroke emotional disorders (PSED), subgroup analyses will be implemented as follows: First, studies will be stratified into subgroups according to distinct PSED types. Subsequently, meta-regression models will be applied to perform further subgroup analyses incorporating variations in 1) forms of TCM interventions, 2) TCM syndrome patterns, 3) treatment durations, or 4) outcome data types, among other factors. These analyses will quantify inter-subgroup differences and test their statistical significance.

Given that the transitivity assumption holds, we will conduct subgroup-specific random-effects network meta-analyses for PSED subtypes and TCM syndrome patterns (where data volume permits) using the “gemtc” package (version 1.0.2) in R 4.3.1 software under a Bayesian framework with non-informative priors. The NMA model will account for treatment effect correlations in multi-arm trials, comparing acupuncture, moxibustion, tuina, Chinese herbal medicine, traditional mind-body exercises (e.g., Tai Chi, Qigong), and other TCM therapies. Results will be presented in tables showing pairwise effect sizes between all interventions within each subgroup alongside probability rankings for optimal interventions (including mean ranks, 95% credible intervals, and surface under the cumulative ranking curve values). Model convergence will be evaluated using the Gelman-Rubin statistic, with values <1.1 indicating adequate convergence (50, 51).

The robustness of findings will be verified by testing different priors and comparing fixed-effect versus random-effects models to exclude influences from key assumptions or data selections. Given the heterogeneity arising from inconsistent definitions of the total clinical effective rate, we will construct a primary analysis model and a sensitivity analysis model and compare the consistency of effect estimates and intervention rankings between the two models to test the robustness of the results concerning this outcome measure. If substantial differences are observed, they will be highlighted in the results and discussion sections.

According to Cochrane guidelines, I² and τ² statistics will be used to assess heterogeneity. If significant statistical heterogeneity or clinical imbalance in effect modifiers is identified, meta-regression analysis will be conducted to explore potential sources. The pre-specified effect modifiers will serve as candidate covariates. Random-effects meta-regression models will be used to examine the associations between these continuous or categorical covariates and the treatment effect size. This analysis will first be conducted within direct comparisons that have a sufficient number of studies. If the network structure and data permit, we may also employ network meta-regression within a Bayesian framework to evaluate the influence of covariates on the relative effects across the entire network.Publication bias, small-study effects, and related uncertainties will be detected and quantified through funnel plots and Egger’s regression. Since included studies may contain both two-arm and multi-arm trials, design inconsistency and loop inconsistency will be evaluated using the design-by-treatment interaction model; if inconsistency is detected within the network, closed loops will undergo further examination (52). The certainty of evidence for treatment effect estimates derived from the network meta-analysis will be assessed using a four-step approach specifically designed for NMA (53, 54). First, all direct comparisons within the network will be rated using the five standard GRADE domains. Second, for indirect comparisons, the certainty will be based on the assessment of the two contributing direct comparisons, with a focus on evaluating transitivity by comparing the distribution of potential effect modifiers across the trials forming the indirect comparison. If clinically important imbalances that threaten transitivity are identified, the evidence will be downgraded for indirectness. Third, for the mixed evidence derived from the NMA model, the assessment will start from the higher certainty rating between the available direct and indirect evidence. Network inconsistency will then be evaluated globally using the design-by-treatment interaction model. If significant and unexplained inconsistency is detected, downgrading will be applied. Finally, the certainty of evidence for each comparison will be categorized as high, moderate, low, or very low. Pre-specified operational downgrading criteria are as follows: For imprecision, downgrading will be considered if the 95% credible interval crosses a pre-defined minimal important difference or spans both the null value and a clinically important effect. For inconsistency, downgrading will be considered if evidence from direct comparisons shows a heterogeneity indicator I² > 50% with non-overlapping confidence intervals across studies, or if the P-value from network inconsistency tests is < 0.05. For indirectness, a clinically relevant imbalance in the distribution of key effect modifiers among the trials underlying an indirect comparison will serve as a basis for downgrading. Given that traditional rank probabilities based on any non-zero effect difference may be fragile and could overstate clinical superiority (54), a sensitivity analysis will be conducted to test the robustness of treatment rankings. This involves setting a series of thresholds representing minimal clinically important differences (e.g., for continuous outcomes: SMD = 0.2, 0.5, 0.8; for dichotomous outcomes: RR = 0.9, 0.8, 0.7) and recalculating the probability of each intervention being optimal. This analysis will evaluate the robustness of ranking results against clinical significance criteria.