Our search strategy encompassed several major electronic databases. These included international databases such as PubMed, EMBASE, Web of Science, and the Cochrane Library, along with widely used Chinese databases like China National Knowledge Infrastructure (CNKI), Wanfang, VIP, and China Biomedical Database (CBM). Search period extended from their initial entries through August 14, 2025. Keywords employed in the search included: post-stroke thalamic pain, thalamic pain, central post-stroke pain, herbal medicine, traditional Chinese medicines, Chinese herbal medicine, Chinese and Western medicine, Chinese patent medicine, and randomized controlled trial. For each database, the search strategy was refined to match its specific features, with an example provided for PubMed in Table 1.

Study Design: Only randomized controlled trials (RCTs), conducted either in China or abroad and reported in Chinese or English, were considered.

Total effective rate was extracted as reported in the original studies. It was defined based on the criteria used in each individual trial, typically categorizing patients as cured, markedly improved, improved, or non-responders. The total effective rate was analyzed as the proportion of patients who achieved a clinical response.

Pain intensity was assessed using the Visual Analog Scale (VAS), Pain Rating Index (PRI), or Present Pain Intensity (PPI) scale. Psychological distress was evaluated using the Self-Rating Anxiety Scale (SAS), Hamilton Anxiety Scale (HAMA), Self-Rating Depression Scale (SDS), or Hamilton Depression Scale (HAMD). Adverse events were also recorded.

Preclinical Evidence: Publications limited to animal experiments or laboratory-based studies were excluded.

Early-stage screening focused on titles and abstracts was performed by two investigators (Xuran Zhang and Ling Cheng) in an independent manner. Records that clearly did not meet the inclusion criteria were excluded. Full texts of potentially eligible studies were further assessed to determine final inclusion. The findings of the included research were cross-checked by the two investigators. Any disagreements or concerns about their inclusion were resolved through discussion with a third investigator. For the included literature, the data extraction process captured key study characteristics, including authorship, publication date, sample size, demographic characteristics (e.g., age), interventions, trial duration, and any reported adverse effects.

Both evaluators independently screened and appraised the methodological rigor and the extent of bias across the included trials. Disagreements underwent initial discussion between evaluators to reach consensus on inclusion. Any persistent disagreements were arbitrated by a senior researcher to arrive at a final conclusion. The Revman (5.4) criteria were utilized to evaluate the listed studies’ methodological quality in terms of possible bias. We appraised the risk of bias by examining several critical aspects: firstly, the methods for generating the random allocation sequence and ensuring its concealment until assignment were scrutinized for potential selection bias. Secondly, we considered whether blinding was applied to participants, treatment providers, and outcome assessors. Further appraisals focused on the completeness of the outcome data, whether all pre-specified outcomes were reported, and the likelihood of any other bias influencing the results.

The statistical computations were conducted using RevMan (5.4) along with Stata version 16. For dichotomous variables, relative risk (RR) was selected as the effect measure. Mean difference (MD) was utilized for continuous outcomes measured on the same scale. A 95% confidence interval (CI) accompanied each effect size. The Cochrane Q test and I² statistics were used to assess heterogeneity across studies. A common-effect model was used when p > 0.1 and I² < 50%, whereas a random-effects model was adopted otherwise. When necessary, subgroup analyses were performed to identify possible causes of heterogeneity. Differences reaching statistical significance were noted if the p value, obtained from the Z-test, was below 0.05. Publication bias was assessed using funnel plots and Egger’s test.

Utilizing Grading of Recommendations Assessment, Development and Evaluation (GRADE) profiler, strength of the evidence of the included trials was evaluated. An independent assessment of the evidence certainty was conducted by two reviewers (Junying Zhai and Tianwei Zhang) in accordance with the GRADE criteria. A consensus process was employed to resolve any discrepancies between their evaluations. Primary criteria included publication bias, inconsistency, indirectness, imprecision, and study limitations. Following GRADE framework, the evidence was stratified into one of four certainty levels: high, moderate, low, or very low.

To reduce the risk of random errors, we performed trial sequential analysis to assess the reliability of the findings with TSA software (version 0.9.5.10). Boundaries were observed using trial sequential monitoring and required information size (RIS) calculations. These aimed to achieve a 25% relative risk reduction, a two-sided alpha level of 0.05 and a beta level of 0.10. Low-bias studies were included to assess control group incidence, mean difference, and variance.

Based on our search approach, 1,269 articles were initially found overall. These records were managed using EndNote X9.2, resulting in 1,027 unique articles after duplicates were removed. A total of 998 studies were removed during the first-round screening, and the remaining 29 were retrieved for full-text review. One record could not be obtained in full text, leaving 28 articles for in-depth review. Full-text screening excluded one non-randomized study, seven trials involving acupuncture, and four studies in which anticonvulsants were used only in the control group. Finally, 16 eligible articles were incorporated in the analysis (16–31). Figure 1 provides an overview of the literature screening process and corresponding outcomes.

The analysis comprised 16 RCTs with 1,122 patients confirmed to have PSTP. Of these, 564 patients were in the experimental group receiving combined herbal medicine and WM, while 558 patients were in the control group receiving only WM. Intervention duration across all studies ranged from 14 to 56 days. Random sampling methods were explicitly stated in 13 studies. Individual RCTs varied in sample size, with the largest study including 120 cases and the smallest including 38 instances. Table 2 provides a comprehensive summary of study characteristics evaluated within this meta-analysis.

WM treatments administered in the control group consisted of carbamazepine, flupentixol melitracen tablets, gabapentin, neurotropin, fluoxetine, pregabalin, amitriptyline, and flupirtine maleate. Herbal medicine interventions included Shaoyao Gancao Decoction, modified Shentong Zhuyu Decoction, Naodesheng Pill, Zhongfeng Huichun Tablet, Jiawei Shentong Zhuyu Decoction, Xingnao Zhitong Decoction, Yiqi Quzhuo Decoction, Huoxue Zhuyu Decoction, Buyang Huanwu Decoction, Xiao Huoluo Dan, Daqinjiao Decoction, MuDan Granule, and modified Huangqi Guizhi Wuwu Decoction. Table 3 provides the detailed composition of the herbal medicine formulas used in the included studies.

All 16 randomized controlled trials (RCTs) reported baseline comparability between the experimental and control groups. Although randomization was mentioned in all studies, several (19, 22, 27, 29) did not specify the exact methods used. Allocation concealment procedures were not addressed in any of the reviewed studies, indicating a potential gap in methodological rigor. Regarding blinding, only one study (21) employed a double-blind methodology, while the remaining studies did not implement blinding. Loss to follow-up and withdrawals were mentioned in 4 studies (17, 21, 24, 30), whereas in 12 studies, no cases of loss to follow-up were documented. Figure 2 displays the evaluation of the extent of bias across selected research.

Funnel plots were generated for VAS scores and total effective rate, and Egger’s test was performed to further evaluate funnel plot asymmetry. Mild asymmetry was observed in the funnel plot for VAS scores, but it did not reach statistical significance according to Egger’s test (p = 0.14), implying that publication bias was minimal (Figure 12). In contrast, significant asymmetry was identified in funnel plot of the total effective rate. With a significant Egger’s test (p < 0.001), the presence of publication bias was acknowledged (Figure 12), which may result from the underreporting of studies with negative or nonsignificant results. To account for this bias, the trim-and-fill approach was utilized, estimating the existence of approximately seven missing studies (Figure 13). After imputation of these theoretical studies, the adjusted pooled risk ratio retained its statistical significance (RR = 1.242, 95% CI: 1.168–1.320).

We next evaluated the overall quality of evidence with the GRADE profiler, followed by a trial sequential analysis to reinforce our conclusions. The certainty of evidence regarding the overall efficacy rate, VAS score, PPI, and PRI score as outcome markers was assessed using the GRADE profiler. The certainty of evidence was deemed to be very low for PRI, low for VAS score and total effective rate, and moderate for PPI (Table 4). These ratings were primarily due to several factors that downgraded the quality of evidence across studies, including inadequate methodology, small sample sizes, significant heterogeneity among studies, and potential publication bias. Despite these limitations, the findings from the meta-analysis provide preliminary evidence regarding the potential benefits of combining herbal medicine with WM for PSTP.

TSA was conducted on the VAS, PPI, PRI, HAMA, SAS, HAMD, SDS, total effective rate, and rate of adverse reactions to minimize the potential for spurious positive results caused by random error. Figure 14 illustrates that the cumulative Z-curve (green solid line) consistently exceeded both the TSA boundaries (red curve) and the conventional significance thresholds (green dashed line), indicating a robust outcome. Specifically, the cumulative Z-curve for VAS, PRI, HAMA, SAS, HAMD, SDS, total effective rate and rate of adverse reactions, surpassed the required information size (RIS) threshold, suggesting sufficient data to draw reliable conclusions. Although the cumulative patient count for PPI was below the RIS estimate, the TSA boundary was crossed, suggesting that the current evidence may be sufficient for this outcome. These TSA results provide additional statistical support for the potential benefit of combining herbal medicine with WM in treating PSTP.

Based on 16 RCTs including 1,122 participants, the meta-analysis suggested that adjunctive herbal medicine may have potential benefits in the treatment of PSTP. The adjunctive therapy exhibited substantial efficacy in reducing pain without significantly increasing adverse events. By using the TSA, we were able to show that the main outcome indicators (the total effective rate, VAS, and PRI score) obtained the RIS, and the cumulative Z-curve consistently reached beyond the trial sequential monitoring boundary (TSMB). For the PPI outcome, the cumulative Z-curve demonstrated a notable early breach of the TSMB prior to reaching the RIS. These findings provide further support for the potential benefit of combining herbal medicine with WM in managing PSTP. GRADE assessment classified the evidence as low for total effective rate and VAS, very low for PRI, and moderate for PPI scores. Owing to methodological limitations in the included studies, these findings should be interpreted with caution.

Pharmacological management of PSTP remains limited. Tricyclic antidepressants (e.g., amitriptyline) are commonly prescribed but can cause side effects, including hypotension and arrhythmias (32). Serotonin-norepinephrine reuptake inhibitors, including duloxetine and venlafaxine (33), along with gabapentinoids provide alternative options but often require combination therapy due to modest efficacy. Investigations have also focused on non-pharmacological interventions, including rTMS, invasive motor cortex stimulation (34–37), and acupuncture (38–40), underscoring the need for multimodal treatment strategies.

Our meta-analysis indicates that combining herbal medicine with WM may provide additional benefits in improving clinical response (total effective rate), reducing pain intensity (VAS, PPI, and PRI) and alleviating anxiety and depression in patients with PSTP, with a safety profile comparable to WM therapy alone. The total effective rate was frequently reported across included studies as a composite measure of clinical response. However, it is not a standardized outcome in international pain research. Therefore, these findings should be interpreted alongside validated pain-related scales, including VAS, PRI, and PPI, which provide more robust and comparable measures of pain intensity. From a clinical perspective, these findings may be relevant, as mood disorders are common comorbidities in PSTP and may further contribute to disease burden (41). For clinicians, an integrative approach could be considered as a potential adjunct to standard pharmacological treatments (e.g., gabapentinoids or antidepressants), although it’s routine clinical application requires further confirmation.

The minimal clinically important difference (MCID) is a widely used indicator for interpreting the clinical relevance of treatment effects. Although MCID values for chronic pain have been reported in previous studies, no established MCID specific to post-stroke thalamic pain currently exists (42). Therefore, evidence from related chronic pain conditions suggests that the MCID for VAS is approximately 1.5–2.0 points (43). In the present study, baseline VAS scores ranged from 3.25 to 6.78, indicating moderate pain severity, which may influence the interpretation of MCID thresholds. The pooled mean difference of −1.50 suggests a clinically meaningful trend of pain reduction, although the exact clinical significance in PSTP remains uncertain. Future studies should establish condition-specific and baseline-adjusted MCID thresholds for PSTP to better interpret treatment effects.

Among the herbal interventions examined, Shentong Zhuyu Decoction was the most frequently investigated (n = 4), followed by Xingnao Zhitong Decoction (n = 2). These prescriptions largely adhere to the therapeutic approach of “activating blood circulation and resolving stasis,” which is consistent with the traditional Chinese medicine theory that blood stasis plays an important role in the pathogenesis of pain symptoms (44). The potential benefits of herbal medicine may be explained by its multi-component and multi-target pharmacological actions. Blood-activating herbs such as Honghua (Carthami Flos) and Danshen (Salvia miltiorrhiza) improve microcirculation and inhibit platelet aggregation (45). Taking safflower as an example, its key bioactive component hydroxy safflor yellow A demonstrates potential in alleviating thalamic pain by modulating the MAPK/p38/iNOS pathway (46) to reduce nitric oxide release, while also suppressing HPA axis hyperactivity, oxidative stress, and neuroinflammation (47, 48). Evidence from related central post-stroke pain research suggests that the PIK3CG/NLRP3 signaling axis may be involved in the analgesic effects of Danshenol B (49). Meanwhile, anti-inflammatory and neuroprotective herbs such as Chuanxiong (Ligusticum chuanxiong) and Danggui (Angelica sinensis) modulate the NF-κB/p38 signaling pathways, thereby reducing neuroinflammation and central sensitization (50). In addition, flavonoids and other bioactive compounds from herbal medicine have demonstrated antidepressant and anxiolytic effects by regulating the hypothalamic–pituitary–adrenal axis and neurotransmitter balance (51–53). These mechanisms provide a plausible explanation for the dual improvement in pain and mood symptoms observed in our analysis.

To our knowledge, this is the first systematic review and meta-analysis to evaluate herbal medicine as an adjunct to WM for the management of PSTP, providing an updated evidence-based synthesis of this therapeutic approach. This study has several strengths. We employed TSA to evaluate the main outcome measures. The results demonstrated that the cumulative Z-curves for the total effective rate, VAS, and PRI scores all crossed the trial sequential monitoring boundaries, thereby providing further statistical support for the robustness of the meta-analysis findings and reducing the risk of type I errors. In addition, beyond assessing analgesic efficacy, this study comprehensively evaluated the intervention’s effects on comorbid anxiety and depression. This multidimensional assessment aligns with the common clinical presentation of PSTP, which frequently involves affective disturbances, offering a more holistic view of patient outcomes.

The interpretation of our findings is limited by several factors. First, the methodological quality of the included RCTs was generally low. None of the included studies reported allocation concealment, and only one trial implemented a double-blind design. Most studies were at unclear risk of bias in several domains, including selection, performance, and detection bias. These limitations reduce the overall confidence in the pooled effect estimates and are reflected in the overall low-to-moderate certainty of evidence as assessed using the GRADE framework. In addition, a small proportion of included studies were unpublished master’s theses rather than peer-reviewed journal articles. Although all studies were assessed using standardized risk of bias tools and incorporated into the GRADE framework, the inclusion of gray literature may raise concerns regarding reporting quality. Second, substantial heterogeneity was observed across VAS, PRI, and PPI, likely due to differences in herbal formulations, types of Western medicine, treatment duration, and baseline characteristics. The lack of standardization of herbal prescriptions may also have contributed. Subgroup analyses and meta-regression were not feasible due to limited study numbers. Third, the evaluation of therapeutic efficacy relied largely on subjective scales (VAS, PRI, and PPI). These instruments are susceptible to patients’ psychological and emotional states and lack specificity for neuropathic pain assessment, potentially compromising the accuracy of outcome measurements (54). Future investigations should incorporate more objective measures, for instance, the Leeds Assessment of Neuropathic Symptoms and Signs (LANSS) (55, 56) or electrophysiological measures like quantitative EEG (57). Fourth, the included RCTs had relatively short follow-up durations, which may be insufficient to assess delayed adverse events and long-term safety outcomes. Finally, all included RCTs were conducted in China, which may limit the external validity and generalizability of the findings to other healthcare systems and populations.