The search scope includes seven databases: PubMed, EMBASE, Cochrane Library, Web of Science (WOS), Chinese National Knowledge Infrastructure (CNKI), Wanfang Data Information Site, and VIP Information Database. The time range is from the establishment of the databases to November 21, 2024. The search strategy includes both subject headings and free terms, approximately as follows: (Stroke OR Brain Infarction OR Cerebral Infarction OR Cerebral Hemorrhage) AND (Medicine, Chinese Traditional) AND (Nursing). For the detailed search strategy, please refer to Supplementary material 2.

Two researchers (YC and HL) independently screened eligible studies based on the inclusion and exclusion criteria. In case of disagreement, the two researchers resolved the issue through discussion.

Inclusion criteria:

Published, peer-reviewed journal articles with a RCT design, targeting stroke patients, written in Chinese or English.

Exclusion criteria:

Full text not available.

Two researchers (YC and HL) independently extracted the required information from the included studies and stored it in an Excel sheet. Disagreements were resolved through discussion, and missing information was obtained by contacting the authors. If there was no response, this was noted in the discussion section. The time point for outcome data was selected as the endpoint of the full treatment course. If there were multiple experimental groups, the group with the best therapeutic effect was chosen for the experimental group data.

Extract the following information:

First author and publication year;

Two researchers, YC and ZW, independently assessed the risk of bias of the included studies using the Cochrane Risk of Bias Tool 2 (RoB-2) (11). In cases of disagreement, a third reviewer (MJ) was consulted to reach a final decision.

Grades of Recommendation, Assessment, Development, and Evaluation (GRADE) is an internationally recognized system for assessing the quality of evidence and the strength of recommendations. Developed by the GRADE Working Group, it is widely used in systematic reviews, clinical practice guidelines, and health technology assessments (12). The GRADEpro GDT tool was used to evaluate the quality of evidence for three outcome indicators. Since all included studies were RCTs, the initial level of evidence was rated as High. The quality was then adjusted based on the following five downgrading factors: risk of bias, inconsistency, imprecision, indirectness, and publication bias. The final evidence levels were categorized as High, Moderate, Low, or Very Low.

The meta-analysis will be conducted using the meta package in R 4.4.2. At the end of the study, FMA and MAS will be used to measure limb motor function scores, and BI will be used to measure ADL scores. Standardized mean differences (SMD) will be used for summarizing the data. I² will be used to quantify statistical heterogeneity, and when combining effect sizes, if the heterogeneity is less than 50%, a fixed-effects model will be used; if greater than 50%, a random-effects model will be applied. Funnel plots and Egger’s test will be used to detect publication bias, and if missing studies are identified, the trim-fill method will be applied to handle missing data. Sensitivity analysis will be performed to assess the robustness of the combined results. Meta-regression and subgroup analysis will be performed based on the following factors: hospital care vs. home care, hemorrhagic stroke vs. ischemic stroke, and high/medium/low risk of bias. A p-value of < 0.05 will be considered statistically significant.

Following the PRISMA guidelines and the predefined search strategy, a total of 428 studies were initially identified. After removing duplicates, 302 studies remained. Based on title and abstract screening, 239 studies were excluded. Among the remaining 63 full-text articles, 40 were excluded due to the following reasons: non-RCT (n = 13), inconsistent interventions (n = 3), outcome measures not aligned with inclusion criteria (n = 20), and full-text not accessible (n = 9). Ultimately, 18 studies were included in the meta-analysis (Figure 1).

Among the 18 included studies (5–7, 13–27), various TCM nursing techniques were employed. Specifically, 13 studies applied acupoint massage, 4 used suspended moxibustion, 1 used herbal soaking, 4 used auricular point pressing, 4 applied acupoint application, 2 used herbal fumigation, 3 employed herbal hot compresses, 4 used cupping therapy, 1 applied herbal iontophoresis, 1 used wax therapy, and 1 employed herbal warm-moist compress. 16 studies adopted hospital-based nursing models, while 2 studies used home-based care models. Regarding stroke type, 6 studies focused on ischemic stroke, 1 on hemorrhagic stroke, and 11 on all types of stroke. In terms of outcomes, 8 studies reported the FMA, 5 reported the MAS, and 12 reported the BI. Detailed information is presented in Table 1.

Among the 18 included studies, 12 (66.67%) were rated as having a low risk of bias, 1 (5.56%) as having some concerns, and 5 (27.78%) as having a high risk of bias (Table 1). Specifically, 11 studies (61.11%) employed an appropriate method for random sequence generation, while only 2 studies (11.11%) clearly reported allocation concealment. None of the studies reported double-blinding of participants and personnel, and 5 studies (27.78%) mentioned blinding of outcome assessors. All studies showed no deviations from the intended interventions, no missing outcome data, and no selective reporting of results (Figure 2).

In terms of motor function, the pooled results from 8 studies demonstrated that TCM nursing interventions significantly improved FMA scores in stroke patients (n = 614, SMD = 1.92, 95% CI [0.54~3.30], p = 0.0066; heterogeneity: χ² = 3.89, df = 7, p < 0.0001, I² = 96.4%; Figure 3A). Similarly, 5 studies showed that TCM nursing interventions significantly improved MAS scores (n = 421, SMD = −0.82, 95% CI [−1.60~−0.03], p = 0.0416; heterogeneity: χ² = 0.75, df = 4, p < 0.0001, I² = 90.3%; Figure 3B). Regarding ADL, 12 studies revealed a significant improvement in BI scores following TCM nursing interventions (n = 878, SMD = 1.34, 95% CI [0.89~1.79], p < 0.0001; heterogeneity: χ² = 0.56, df = 11, p < 0.0001, I² = 86.4%; Figure 3C). For all three outcomes, I² values exceeded 50%, indicating substantial heterogeneity. To explore the sources of heterogeneity, sensitivity analyses, meta-regression, and subgroup analyses were subsequently conducted.

Funnel plots and Egger’s tests were employed to assess potential publication bias. The funnel plot for FMA showed asymmetry (Figure 4A), and the Egger’s test indicated significant publication bias (p = 0.0006). After applying the trim-fill method, one study was imputed, resulting in a corrected effect size that differed from the original (SMD = 1.26, 95% CI [−0.53~3.06], p = 0.1682; Figures 4B,C). For MAS, the funnel plot also exhibited asymmetry (Figure 4D), and the Egger’s test yielded a p-value of 0.0085. However, the trim-fill analysis indicated no missing studies. Regarding BI, the funnel plot appeared asymmetric (Figure 4E), with an Egger’s test p-value of 0.0013. One study was imputed using the trim-fill method, and the adjusted effect size remained consistent with the original result (SMD = 1.19, 95% CI [0.67~1.71], p < 0.0001; Figures 4F,G). The above results suggest that the meta-analysis findings for FMA, MAS, and BI may be subject to publication bias. After controlling for publication bias using the trim-fill method, MAS and BI still showed significant differences, indicating that their results are relatively robust. However, for FMA, the results were no longer significant after adjusting for publication bias, which suggests that the initially observed significant difference may have been due to bias.

Sensitivity analysis showed that the combined effect sizes for FMA and BI remained statistically significant after excluding any single study (Figures 5A,C), indicating good robustness of the meta-analysis results and suggesting that the observed heterogeneity was not driven by any individual study. For MAS, heterogeneity dropped to I² = 0% after excluding the study by Wang (20), suggesting that this study was the primary source of heterogeneity. Furthermore, upon excluding this study, the effect remained statistically significant (p = 0.0001), whereas exclusion of any other single study resulted in non-significant p-values (all p > 0.05; Figure 5B). These findings indicate that the MAS-related results are not stable and highlight the need for additional large-scale studies to provide more reliable evidence for clinical practice.

Meta-regression and subgroup analyses were conducted to explore potential sources of heterogeneity based on care setting (hospital-based vs. home-based), stroke type (ischemic vs. hemorrhagic), and risk of bias (high, unclear, or low). Meta-regression results indicated that none of these factors significantly contributed to the heterogeneity in any of the outcome measures (all p > 0.05). Subgroup analyses further confirmed these findings, as within-group heterogeneity remained high (I² > 50%) across all subgroups, suggesting that these factors were not the primary sources of heterogeneity. However, some notable differences in effect sizes were observed between subgroups:

1 Fugl-Meyer Assessment (FMA):

A significant difference was found between care settings (χ² = 5.76, df = 1, p = 0.0164), with hospital-based care (SMD = 2.45, 95% CI [0.78~4.11]) showing greater improvements than home-based care (SMD = 0.37, 95% CI [0.06~0.68]; Figure 6A). Risk of bias also influenced the effect size significantly (χ² = 13.46, df = 2, p = 0.0012), with studies at high risk (SMD = 2.83, 95% CI [0.29~5.36]) and unclear risk (SMD = 2.13, 95% CI [1.59 ~ 2.68]) reporting larger effects than those at low risk (SMD = 0.66, 95% CI [0.05~1.27]; Figure 6B). No significant difference was observed across stroke types (χ² = 1.48, df = 2, p = 0.4774; Figure 6C).

2 Modified Ashworth Scale (MAS):

No significant difference in effect size was found among different risk of bias levels (χ² = 1.03, df = 1, p = 0.3111; Figure 6D).

3 Barthel Index (BI):

A significant difference was observed between care settings (χ² = 10.80, df = 1, p = 0.0010), with hospital-based care (SMD = 1.43, 95% CI [0.98~1.88]) being more effective than home-based care (SMD = 0.33, 95% CI [−0.14~0.81]; Figure 6E). No significant differences were found across different levels of risk of bias (χ² = 0.60, df = 2, p = 0.7391; Figure 6F) or stroke types (χ² = 1.15, df = 2, p = 0.5621; Figure 6G).

The evaluation results show that the Egger’s test for all three outcome indicators exhibited significant differences, leading to the assessment of “publication bias strongly suspected.” For FMA, the RoB-2 assessment revealed that half of the studies had a risk of bias, with the heterogeneity test showing I² > 50% and no identifiable source of heterogeneity. As a result, both “Risk of bias” and “Inconsistency” were rated as “serious,” leading to an overall evidence quality of Very Low. For MAS, although the heterogeneity test (I² > 50%) indicated high heterogeneity, sensitivity analysis identified the source of heterogeneity (20), so “Inconsistency” was rated as “not serious,” resulting in an overall evidence quality of Moderate. For BI, due to the heterogeneity test (I² > 50%) showing high heterogeneity and no identified sources of heterogeneity, “Inconsistency” was rated as “serious,” resulting in an overall evidence quality of Low (Figure 7).

Hu et al. (28) conducted an umbrella review suggesting that TCM nursing may be effective for stroke management; however, the credibility of the findings was limited. Notably, their study was based on the Technical Operation Regulations of Traditional Chinese Nursing Practices issued by the Chinese Association of Traditional Chinese Medicine in 2006—a guideline that is now outdated and lacks contemporary clinical relevance. In contrast, the National Administration of Traditional Chinese Medicine introduced an updated set of 18 standardized TCM nursing techniques in 2015. These techniques span a broad spectrum, from external herbal applications to modern acupoint-based interventions, and are more representative of current TCM nursing practices. To build upon and update existing evidence, the present study systematically synthesized all available literature to date on TCM nursing interventions for stroke. However, due to the limited number of studies using the same nursing technique, which did not support a meta-analysis, we combined different TCM nursing techniques for a unified analysis. The meta-analysis revealed that TCM nursing significantly improved motor function and ADL in stroke patients, as measured by FMA (SMD = 1.92, p = 0.0066), MAS (SMD = −0.82, p = 0.0416), and BI (SMD = 1.34, p < 0.0001). However, these aggregated effect sizes were accompanied by substantial heterogeneity and potential publication bias. To address these concerns, we conducted sensitivity analyses, meta-regression, and subgroup analyses to explore the sources of heterogeneity, and used Egger’s test along with the trim-fill method to assess and adjust for possible publication bias.

The Cochrane risk of bias assessment showed that, regarding randomization, more than half of the studies reported random allocation methods, while only a small number mentioned allocation concealment. Nonetheless, scientifically appropriate randomization methods—such as computer-generated sequences or random number tables—can prevent investigators or participants from knowing the group assignments in advance. Therefore, studies that reported proper randomization methods were considered to have a low risk of bias in the randomization process.

In terms of blinding, none of the studies reported blinding of participants or personnel. In drug trials, placebo use is the preferred approach to achieve double-blinding; however, for interventional therapies, it is often difficult to blind the operator. Certain traditional Chinese nursing techniques, such as acupoint application or external herbal treatments, may allow double-blinding through the use of identical-appearing and similarly scented preparations. In contrast, techniques like massage, moxibustion, and auricular point pressing are challenging to blind from the practitioner. Although both the operators and participants might have been aware of the treatment, given that all studies showed no deviation from the intended interventions, this domain was judged as low risk. Additionally, only a few studies reported blinding of outcome assessors. However, considering that the outcome measures were not based on subjective perception (e.g., pain scores), but rather were relatively objective, the risk of bias in outcome measurement was considered low. Finally, none of the studies had missing outcome data or selectively reported results. In summary, the overall risk of bias in the 18 included studies was low, with the primary source of high risk stemming from inadequate randomization methods.

In terms of reporting bias, Egger’s test suggested potential publication bias across all three outcome indicators. After applying the trim-fill method, the summary effect sizes for MAS and BI remained robust, while the result for FMA was no longer significant after controlling for publication bias. This suggests that the initially observed significant difference for FMA may have been due to bias. These results indicate that TCM nursing techniques are more effective in improving spasticity and ADL in stroke patients, whereas their effect on upper and lower limb motor function recovery remains debatable.

All three pooled outcomes exhibited I² values greater than 50%, indicating substantial heterogeneity. To investigate its sources, we conducted sensitivity analysis, meta-regression, and subgroup analyses. Sensitivity analysis showed that for MAS, heterogeneity disappeared (I² = 0%) after excluding the study by Wang (20), suggesting that this study was the primary source of heterogeneity. Upon re-examination, we found that the control group in this study received conventional care combined with herbal hot compress, whereas other studies’ control groups received conventional care alone, which may partly explain the observed heterogeneity. For FMA and BI, excluding any single study did not significantly alter heterogeneity, indicating that heterogeneity was not driven by an individual study. Meta-regression analyses showed that care setting, stroke type, and risk of bias were not significant sources of heterogeneity for any of the outcomes (all p > 0.05). Subgroup analyses further demonstrated that within-group heterogeneity remained high (I² > 50%), suggesting that these factors could not explain the heterogeneity either.

We speculate that part of the heterogeneity may stem from differences in the specific nursing techniques applied in the intervention groups. Across the included studies, various combinations of 18 TCM nursing techniques were used, which could contribute to inconsistency. Using identical nursing techniques might reduce between-study heterogeneity; however, currently, there are insufficient RCTs focusing on the same technique to allow for separate meta-analysis. In addition, heterogeneity may also stem from inconsistencies in outcome assessment timing, with time spans ranging from 2 weeks to 3 months post-treatment. Since 4 studies did not provide specific time points for outcome assessments, we did not conduct a subgroup analysis (14–16, 26). Considering the above, we employed a random-effects model to obtain relatively reliable results despite the presence of heterogeneity.

Subgroup analyses based on study characteristics revealed additional findings beyond heterogeneity exploration. Both hospital-based and home-based care models significantly improved limb motor function and ADL in stroke patients, with hospital-based care showing greater improvements. Home-based care is an important component of nursing practice. A bibliometric analysis based on the WOS indicated that the number of publications on home-based care has been steadily increasing, with “quality of life” identified as a major research focus (29). However, home-based care is still at an early stage of development and remains relatively underutilized in clinical practice; for example, among the 18 included studies, only two adopted a home-based care model (5, 13). Moreover, current research in this area remains relatively narrow in scope, lacking interdisciplinary integration, which warrants further exploration and refinement in future practice. In addition, subgroup analyses revealed no significant differences in outcome measures across different types of stroke, suggesting that Chinese medicine nursing techniques are similarly effective for both hemorrhagic and ischemic stroke patients. Finally, studies with a higher risk of bias reported larger effect sizes, highlighting the need for cautious interpretation of results and emphasizing the importance of more rigorous study designs in future research.

The GRADE results indicate that FMA was rated as Very Low, meaning there is almost no confidence in the effect estimate: the true value is likely to differ significantly from the estimated value. MAS was rated as Moderate, suggesting moderate confidence in the effect estimate: the true value may be close to the estimate, but there is still a possibility that they could differ substantially. BI was rated as Low, indicating limited confidence in the effect estimate: the true value could differ significantly from the estimated value. The main downgrading factors were publication bias, risk of bias, and heterogeneity. Future studies should aim to improve in terms of multicenter design, large sample sizes, and rigorous trial methodology.

Several animal studies have begun to elucidate the biological mechanisms underlying the effects of commonly used TCM nursing techniques for stroke treatment. Research has shown that moxibustion can reduce lipid peroxidation levels after cerebral ischemia–reperfusion, decrease reactive oxygen species accumulation, inhibit ferroptosis of neuronal cells, and alleviate neurological impairment (30). Other studies have demonstrated that acupoint application can lower the expression of inflammatory cytokines following ischemia–reperfusion injury, exert anti-apoptotic effects, significantly reduce infarct volume, and improve neurological deficits (31). Additionally, auricular acupressure has been found to ameliorate cerebral ischemic injury, potentially by regulating immune-inflammatory responses (32). However, animal studies on other TCM nursing techniques remain limited, possibly due to the technical challenges of replicating interventions such as acupoint massage, cupping, and scraping therapy in animal models. This area warrants further exploration.

Based on the available literature, inflammatory response appears to be a key mechanism through which TCM nursing techniques exert therapeutic effects in stroke. Current studies mainly focus on changes in inflammatory cytokines; future research could delve deeper into upstream mechanisms, such as microglial phenotype modulation and pyroptosis, to better understand the underlying processes.

Based on the latest guidelines issued by the National Administration of Traditional Chinese Medicine, this study systematically retrieved RCTs from seven major databases to evaluate the effects of TCM nursing techniques on motor function and ADL in stroke patients—an effort that represents the first of its kind both domestically and internationally. Furthermore, given the considerable heterogeneity among the included studies, we employed three approaches—sensitivity analysis, meta-regression, and subgroup analysis—to explore potential sources of heterogeneity. Nevertheless, there is still high heterogeneity within the subgroups, which may be related to differences in the TCM nursing techniques and outcome assessment time points across studies. Additionally, the lack of allocation concealment and blinding reporting in most studies indirectly led to the poor GRADE assessments for FMA and BI. Therefore, the above results should be interpreted with caution.

During the course of this study, we identified several issues in the current TCM nursing technique-related RCTs, such as inconsistent interventions and the lack of blinding. We recommend that future RCTs adopt more uniform intervention protocols to reduce heterogeneity, which would also be beneficial for the development of clinical guidelines or industry standards. Additionally, some TCM nursing techniques, such as acupoint application and external use of Chinese medicine, are suitable for blinding with placebos, which can be achieved using visually and olfactorily similar substances.