All RCTs about SXNI for AIS were eligible for inclusion. The language of publication was not limited. We excluded case-control studies, retrospective studies, case reports, animal experiments, cell experiments, reviews, meta-analyses, clinical experiences, commentaries, and conference abstracts.

Patients who were older than 18 years and diagnosed with AIS according to the recognized guidelines or criteria and confirmed by computed tomography (CT) or magnetic resonance imaging (MRI) were included without limitations on sex, culture, nationality, or race. The duration of the stroke should have been less than 14 days. We excluded patients with hypoxic-ischemic encephalopathy and postpartum apoplexy.

The SXNI group received SXNI combined with conventional treatments (CTs), and the non-SXNI group received CTs alone or CTs combined with other injections. According to the Chinese guidelines for the diagnosis and treatment of AIS 2018; Chinese Society of Neurology Chinese Stroke Society, 2018), CTs include IVT, EVT, antiplatelet drugs, neuroprotective agents, and symptomatic supportive treatment, such as the management of blood pressure and blood lipid and sugar levels. Other injections were defined as any monodrug of non-Ginkgo biloba extract, such as compound Danshen injection, Shuxuetong injection, and so on. The duration of treatment had to be between 14 and 30 days. There were no restrictions on daily dose or frequency.

The primary outcome was clinical efficacy assessed by the CER. According to the scoring criteria for clinical neurological deficit (CND) in stroke patients (Fourth Session of Chinese National Cerebrovascular Conference, 1996), improvements in CND were classified into five categories: a) recovery: 90%–100% reduction in CND with a disability level of 0; b) significant improvement: 46%–89% reduction in CND with a disability level of 1–3; c) progress: 18%–45% reduction in CND; d) inefficacy: less than 17% reduction in CND; e) deterioration: greater increase in CND. CER = (a + b + c) cases/total cases × 100% (National Administration of Traditional Chinese Medicine, 1994; Zheng, 2002).

The secondary outcomes included a) neurological function assessed by the Chinese Stroke Scale (CSS) and the NIHSS; b) ADL assessed by the Barthel Index (BI); c) AEs. The CSS consists of eight dimensions with a total score of 45. A higher CSS score indicates that a patient’s stroke is more severe (Fourth Session of Chinese National Cerebrovascular Conference, 1996). The NHISS consists of 15 items with a total score of 42 and is a validated tool for assessing the severity of acute stroke. The greater the NIHSS score is, the more severe the neurological deficit (Brott et al., 1989; Kwah and Diong, 2014). The highest BI score is 100, and the higher the BI score is, the better the individual’s ability to perform ADLs (Mahoney and Barthel, 1965). The common AEs related to SXNI are headache, flushing, and so on.

Seven databases were searched from their inception to January 2023, including PubMed, the Cochrane Library, Embase, the China National Knowledge Infrastructure (CNKI), the WanFang Database, the Chinese Scientific Journal Database (VIP), and the Chinese Biological Medicine Database (CBM). We also searched the Chinese Clinical Trial Registry (http://www.chictr.org.cn/) and the U.S. Clinical Trial Registry (https://clinicaltrials.gov/). We screened the reference lists of all included studies to identify more eligible studies. The search terms included (“stroke” OR “palsy” OR “apoplexy” OR “apoplexia” OR “cerebrovascular disorders” OR “cerebral infarction” OR “infarct of brain” OR “cerebral hemorrhage” OR “intracerebral hemorrhage” OR “brain hemorrhage”) AND (“shuxuening” OR “shuxuening injection” OR “Ginkgo Leaf” OR “Folium Ginkgo” OR “Chinese patent medicine”). The detailed search strategy is described in Supplementary Appendix 1.

We not only removed reduplicated articles using NoteExpress 3.6 but also performed manual de-duplication. Then, two reviewers independently screened the literature based on predefined eligibility criteria. We first excluded irrelevant literature by reading the titles and abstracts and then reading the full texts to select eligible studies. When two or more studies reported the same RCT, we selected the study with the largest sample size, appropriate outcomes, and the earliest publication. We consulted a third reviewer to resolve any disagreements.

We developed standardized data extraction forms that included study information (e.g., author, publication year, city, and country), characteristics of participants (e.g., age, sex, and course of stroke), study design (e.g., sample size, interventions, administration methods, duration of intervention, and SXNI manufacturing company), and outcomes (e.g., CER, NIHSS, CSS, BI, and AEs). Two reviewers extracted the data independently and cross-checked the extracted data. We contacted the authors via email when the information in the article was unclear or insufficient. If there were disagreements on data extraction, a third reviewer was consulted.

The quality of the included studies was evaluated by two reviewers using the Cochrane risk of bias (ROB) assessment tool (Higgins and Green, 2011). The ROB tool includes seven items: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other biases. Each item can be judged as “low risk”, “high risk” or “unclear risk”. If a study had a “low risk” of more than four domains, its overall quality was judged as high. Any disagreements between the two reviewers were resolved by a third reviewer.

We used R software (version 4.1.2, https://cran.r-project.org/) and Review Manager software (version 5.4, Cochrane Collaboration, Copenhagen, Denmark) to analyze the data. Dichotomous data are expressed as risk ratios (RRs) and 95% confidence intervals (CIs). Continuous data assessed using the same scale are expressed as weighted mean differences (WMDs) and 95% CIs, whereas standardized mean differences (SMDs) and 95% CIs were used for the different scales. Statistical heterogeneity was assessed by I ² and p values from the Cochrane Q test. I ² < 50% and p ≥ 0.05 indicated no heterogeneity among studies. A random effects model was used if heterogeneity existed; otherwise, the fixed effects model was used. p < 0.05 was considered to indicate a statistically significant difference.

We tested the stability of the meta-analysis results by alternating the random and fixed effects models. Moreover, we excluded each study in turn to test the robustness of the results.

We performed a subgroup analysis based on the duration of intervention, daily dose, and different injections used in the control group. We also conducted mixed-effects meta-regression analysis using sample size (n ≤ 107 vs. n > 107), quality of studies (high vs. low), publication year, mean age (≤60 years old vs. > 60 years old vs. mixed), and total dose of SXNI (≤280 mL vs. > 280 mL) as covariates to search for sources of heterogeneity.

Publication bias was assessed with “funnel” and Egger’s tests when the number of included studies was more than 10.

We evaluated the quality of evidence for each outcome using the Grading of Recommendation, Assessment, Development, and Evaluation (GRADE) tool (https://www.gradepro.org/). The GRADE included five domains: a) risk of bias, b) inconsistency, c) indirectness, d) imprecision, and e) other considerations, such as publication bias and number of included studies. The overall level of evidence for every outcome was judged as high, moderate, low, or very low. Any disagreement regarding the level of evidence was resolved through discussion and consultation.

A total of 4,061 records were retrieved, 2075 of which were duplicate publications. After removing 1,775 records that did not meet the criteria by screening the titles and abstracts, we read the full texts of 211 articles. Finally, 116 eligible studies were included in the meta-analysis. The details of the study selection process are displayed in Figure 1. The full citation details of the included and excluded studies are described in Supplementary Appendix 2.

All included studies were conducted in China and published in Chinese, with publication years distributed between 2002 and 2023. The most frequent provinces of author affiliation were Henan, Shandong, Guangdong, and Hebei. The sample sizes were between 40 and 438, and the duration of AIS ranged from less than 12 h to 14 days. The intervention duration ranged from 14 to 30 days. The daily dose of SXNI varied from 2 to 30 mL. The total dose of SXNI was between 28 and 900 mL. Thirty-six studies with 3,568 participants compared the efficacy of SXNI plus CTs vs. CTs alone, and eight studies with 8,833 participants compared the efficacy of SXNI plus CTs vs. CTs plus other injections. Of the 116 studies, 102 reported the CER, 14 reported the BI, 44 reported the CSS, and 16 reported the NIHSS. The characteristics of the included studies are described in Table 1.

Fifteen (12.9%) studies were judged to be of moderate to high quality (Table 2; Figure 2). Of 116 studies, 110 (94.8%) mentioned the word “randomization”, 15 (12.9%) used random number table methods, and the remaining studies used incorrect or undetailed randomization methods. Five (4.3%) studies mentioned the word “blinding” without a detailed description of the specific methods. None of the studies reported allocation concealment. Three (2.6%) studies were judged as high risk because they did not conduct intention-to-treat analysis and did not specify the reason for case dropout. Nine (7.8%) studies were assessed as high risk for selective reporting based on the description of the methods in the article. All included studies were free of other sources of ROB.

Seventy-one studies reported AEs, and 49 reported no AEs. Twenty-two studies reported AEs related to the SXNI group or non-SXNI group, five of which did not report the number of AEs in the SXNI group, and two did not report the number of AEs in the non-SXNI group. SXNI plus CTs (1.53%, 61/3,994) was similarly safe to CTs alone or CTs plus other injections (1.32%, 50/3,797). The most common AEs related to SXNI were cardiovascular system events, and the five most common symptoms were dizziness, flushing, palpitations, nausea, and headache. The most common AEs related to the non-SXNI group were digestive system events, and the five most common symptoms were palpitations, nausea, dizziness, flushing, and vomiting. Moreover, all these symptoms in both the SXNI group and the non-SXNI group were mild and disappeared after discontinuation of the drug and symptomatic treatment. The symptoms of AEs are detailed in Supplementary Table S3.

We used the GRADE tool to systematically evaluate the quality of each outcome. The levels of evidence for CER and BI scores were low in SXNI plus CTs compared with CTs alone for AIS, while the levels of evidence for the NIHSS score and CSS score were very low (Table 3). Additionally, the level of evidence for the CER, NIHSS, CSS, and BI scores were all very low for SXNI plus CTs compared with CTs plus other injections for AIS (Table 3). The grade was decreased mainly because of the low methodological quality of the included studies, high heterogeneity among the included studies, and potential publication bias.

A total of 116 studies involving 12,401 participants were included in this review to assess the efficacy and safety of SXNI as an add-on therapy for patients with AIS. The meta-analysis results showed that SXNI plus CTs was superior to CTs alone or CTs plus other injections in improving patients’ CER, NIHSS, CSS, and BI scores, suggesting that SXNI combined with CTs can significantly improve clinical efficacy, reduce neurological deficits, and promote the recovery of ADL in patients with AIS. Subgroup analysis and mixed-effect meta-regression analysis indicated that duration of intervention, daily dose of SXNI, total dose of SXNI, sample size, mean age, different injections used in the control group, quality of studies, and publication year were not sources of heterogeneity. Except for the BI score between SXNI plus CTs and CTs plus other injections, the sensitivity analysis of the remaining indicators demonstrated that the meta-analysis results were robust. Sensitivity analysis revealed that the difference in BI scores between SXNI plus CTs and CTs plus other injections was not significant when five studies were removed one at a time (Chen, 2021; Li, 2014; Ma, 2016; Zhang et al., 2009; Zhao, 2015). In terms of safety, based on the results of this study, we cautiously believe that SXNI is relatively safe and reliable. Publication bias was found in the CER and NIHSS scores. Only 15 (12.9%) studies were judged to be of moderate to high quality. The levels of evidence for the CER, NIHSS, CSS, and BI scores were very low to low quality. Therefore, the results of this study should be considered with caution.

According to traditional Chinese medicine theory, the main pathological changes in AIS are qi deficiency and blood stasis. Qi is regarded as the main material basement that constitutes the human body and maintains life activities and an energy which manifests simultaneously on the physical and emotional-mental-spiritual level, and it has functions of promoting, warming, defense, transformative action and containment) (Jiang et al., 2023; Tian et al., 2024). Therefore, clinical treatment of this disease is mainly carried out by supplementing qi, promoting blood circulation, and dredging meridians and collaterals. SXNI is an extract of G. biloba leaves that promotes blood circulation, eliminates blood stasis, unblocks meridians, tonifies qi, and strengthens the brain (Ding, 2018; Zhang, 2019). From the perspective of Western medicine, SXNI can improve blood rheology, inhibit platelet aggregation, and reduce blood viscosity (Zhang et al., 2005). Ginkgolides can inhibit glutamate receptor-gated calcium channels, significantly reduce Ca²⁺ content in brain cells, prevent intracellular Ca²⁺ overload, and reduce cascade reaction-induced brain cell necrosis (Chandrasekaran et al., 2001). SXNI can improve the energy metabolism and nutrition of hypoxic brain cells via its antioxidant properties (Smith et al., 2002; Nash and Shah, 2015) and reduce the apoptosis of neuronal cells in brain tissue through the inhibition of multiple inflammatory responses and the modulation of oxidative stress levels (Dong et al., 2023). Moreover, SXNI can increase superoxide dismutase activity against oxidative damage caused by ischemia to protect muscle cells (Gao et al., 2007; Koyama et al., 2013; Powers and Lennon, 1999), and Ginkgolide B also can promote muscle regeneration by reviving osteocalcin–GPRC6A signalling (Wang et al., 2023). This may be the potential mechanism by which SXNI can reduce neurological deficits and improve limb function in AIS patients.

A centralized hospital monitoring study of the safety of SXNI injection in 9,735 patients showed that SXNI had a low incidence of AEs and good safety (Qu et al., 2023), which is consistent with our results. However, with the expanding application scope and increasing use frequency of SXNI in clinical practice, it is often abused. Reports of AEs related to SXNI were not uncommon. More than 80% of patients experienced AEs within 2 hours after administration (Chen, 2018; Gong and Shang, 2023), which prompted us to focus on monitoring the patient’s condition during this time period. The occurrence of AEs to SXNI may be related to many factors, such as patient age, type of solvent, dosage, over-the-counter use, and coadministration (Wu, 2019). Therefore, we should pay attention to selecting the appropriate drug solvent, strictly follow the instructions for use, and try to avoid using other injections in combination.

In this review, 12.9% of the included studies were defined as moderate to high quality. Correct randomization and allocation concealment are important conditions for ensuring the minimization of selection bias; however, only 15 (12.9%) studies reported correct randomization methods, and none of the studies reported detailed allocation concealment, blinding, or implementation. Of all studies, 2.6% did not conduct intention-to-treat analysis and did not specify the reason for case dropout, 7.8% may not have completely reported outcomes, and none was registered in advance. Although all outcomes except CER showed significant heterogeneity, our subgroup analysis and mixed-effect regression analysis did not reveal the source of heterogeneity. Sensitivity analysis revealed differences in BI scores between SXNI plus CTs and CTs plus other injections when five studies (Chen, 2021; Li, 2014; Ma, 2016; Zhang et al., 2009; Zhao, 2015) were excluded one by one. The different manufacturers of SXNI, large sample size spans, different age-inclusion criteria, and populations in the different regions of these five studies may be potential sources of heterogeneity. The overall level of evidence for each outcome ranged from very low to low quality. The reasons for this were mainly the low methodological quality of the included studies, high heterogeneity among the included studies, and potential publication bias. Consequently, future RCTs with large sample sizes from multiple centers should be rigorously conducted in accordance with the consolidated standards of reporting trials (CONSORT) and should be registered in advance on the registration platform (Schulz et al., 2010).

Our results suggested that SXNI in the treatment of AIS could not only increase CER and improve neurological function and ADL but also had good safety, which was consistent with the results of five systematic reviews previously published in 2011, 2012, and 2016 (Yan, et al., 2011; Xi et al., 2012; Zheng et al., 2012; Cai and Jiang, 2012; Tan, et al., 2016). We included patients who had AIS within 14 days; however, Xi et al. and Zheng et al. included patients whose stroke occurred within 72 h and 7 days, respectively. According to the use requirements of SXNI, we included only adult stroke patients, but previous studies did not limit the age of AIS patients, which may have affected the scope of its application. Compared with previous meta-analyses (Yan, et al., 2011; Xi et al., 2012; Zheng et al., 2012; Cai and Jiang, 2012; Tan, et al., 2016), our study used more comprehensive outcomes (including CER, neurological function, ADL, and AEs) to evaluate the efficacy and safety of SXNI as an add-on therapy for AIS. However, Yan et al. included only 14 RCTs and did not report neurological deficit scores for outcome measures. Cai et al. only observed the CER of SXNI in AIS patients based on 22 RCTs. Tan et al. did not evaluate the effect of SXNI on the ability of patients with AIS to perform ADL. Moreover, the studies included in the previous meta-analyses (Yan, et al., 2011; Xi et al., 2012; Zheng et al., 2012; Cai and Jiang, 2012; Tan, et al., 2016) were all published before 2016, and many RCTs focusing on SXNI for AIS have been reported since 2016; therefore, our literature search deadline was 2023 to update the evidence.

Our findings supported the clinical use of SXNI as an add-on treatment for AIS patients with a low level of evidence. Moreover, because no long-term follow-up studies have been performed to assess the effect of SXNI on the long-term prognosis of AIS patients, our study can only confirm that an intervention duration of 14–30 days with SXNI can effectively improve the neurological function and ADL of AIS patients. The usual daily dose of SXNI was 20 mL, and the most frequent treatment duration was 14 days. The five most common symptoms related to AEs were dizziness, flushing, palpitations, nausea, and headache, which is consistent with previous studies (Qu et al., 2023; Gong and Shang, 2023). Cardiovascular system events were the most common AEs of SXNI for AIS patients. Therefore, in clinical practice, SXNI should be used in strict accordance with the instructions, and patients should be monitored within 2 h after SXNI administration.

In this review, most (87.1%) included studies were judged as low quality. The lack of a detailed description of allocation concealment and blinding methods is a common problem in most studies. In all, 7.8% of the studies may have had a selective reporting bias. The included studies were all single-center RCTs conducted in China, which may mask the efficacy of SXNI in stroke patients of different ethnicities. Most of the studies were conducted in cities, and there is a lack of studies on SXNI for AIS in rural or primary medical institutions, while AIS patients in rural and primary medical institutions are in urgent need of effective and convenient drugs or technologies. SXNI has the advantages of being easy to use and safe. If SXNI is indeed effective in improving the prognosis of AIS patients, it will compensate for the deficiency caused by limited access to IVT and/or EVT. Large-sample, multicenter, placebo-controlled RCTs, especially in rural or primary medical institutions, should be conducted in the future. Moreover, all RCTs should be registered in an internationally recognized clinical trial registration platform and reported in accordance with the CONSORT guidelines.

Some limitations of this study should be noted. First, all included trials were conducted in China and published in Chinese, which may have led to selection bias and publication bias and reduced the generalizability of the results to the applicable population. Second, most of the included studies were of low quality and had significant heterogeneity, which led to a low level of evidence.

There are also some advantages in our study. First, we carried out an extensive literature search, including seven databases and two clinical trial registration platforms, to obtain as many relevant studies as possible. Second, two reviewers independently performed literature screening, data extraction, and quality evaluation and cross-checked to ensure the accuracy of the data. Third, we conducted a sensitivity analysis, subgroup analysis, and mixed-effects meta-regression analysis to test the robustness of the meta-analysis results. We also assessed the level of evidence for each outcome based on the GRADE tool.