Two researchers (SML and STY) independently searched all relevant studies up to June 2023 from four English databases (PubMed, Embase, CENTRAL, and Web of Science) and four Chinese databases (SinoMed, VIP information database, China National Knowledge Infrastructure, and Wanfang Data Information Site) without no limitation on language or areas (The strategy was described in Supplementary Table S3). Furthermore, we searched ClinicalTrials.gov and the Chinese Clinical Trial Registry, and tracked citations on relevant systematic reviews.

We excluded the following studies: 1) Case reports, animal studies, case reports, and meta-analysis, etc.,; 2) studies treated with other traditional chinese medicine such as Chinese botanical drug, massage, scraping, cupping; 3) studies with incomplete or inadequate data; 4) studies involving patients with other serious diseases such as severe liver, brain, and kidney diseases, etc.

Two researchers (SML and STY) independently conducted the study screening and data extraction according to the inclusion and exclusion criteria. EndNote X9 was used to manage the records and removed duplicates. The extracted data included 1) sample characteristics: authors, publication year, sample size, mean age of participants, types of CHD, and the injection method of SXNI; 2) study design: randomization, allocation concealment, and blinding; 3) information of outcomes: clinical efficacy rate, angina symptoms, and other valuable outcomes. A third researcher (JZC) reviewed and confirmed the final data extraction sheets. For missing data, efforts were made to contact the authors of the respective articles. If unsuccessful, the study had to be excluded from the analysis.

Two researchers (SML and STY) assessed the quality of included studies using the Cochrane risk of bias tool 2.0 (RoB2.0), in which six domains were evaluated: 1) randomization process, 2) deviations from the intended interventions, 3) missing outcome data, 4) measurement of the outcome, 5) selection of the reported outcome, and 6) overall bias. Any discrepancies that arose during the assessment were resolved through discussion with a third researcher (JZC). Each entry was either rated as low, high, or unclear.

Statistical analyses were performed using Review Manager version 5.4 (Cochrane Collaboration, Copenhagen, Denmark) and STATA 15.1 (StataCorp, College Station, TX, United States). For dichotomous data, the relative risk (RR) was calculated, while for continuous variables, the mean difference (MD) was calculated. Confidence intervals (CIs) were set at 95%, and a p < 0.05 indicated statistical significance for the overall effect. Heterogeneity among studies was quantified using the inconsistency index (I2). If I2 < 50% and p > 0.05, the fixed-effects model was used to pool the data; otherwise, the random-effects model was used. The mean and standard deviation for variable changes before and after the treatment are calculated with the methods provided in Chapter 16.1.3.2 of Cochrane Handbook Version 5.0.2 (Corr = 0.40).

Preplanned subgroup analyses were conducted to investigate the potential influence of specific characteristics (duration of treatment, dose of SXNI, types of CHD, and the mode of administration) on the efficacy of SXNI. Furthermore, univariate meta-regression analyses using sample size, mean age, and year of publication as co-variables were performed to identify the confounding factors that may result in heterogeneity. Sensitivity analysis was performed by excluding studies with a high degree of overall bias or by excluding single study to assess the stability of the results and query the sources of heterogeneity.

When the number of studies exceeded 10, we assessed the possibility of publication bias using funnel plots and Egger’s regression test. In addition, we used the trim-and-fill analysis to estimate the number of missing studies through iterative methods, and assess the effect of publication bias on the interpretation of the results.

We used Grading of Recommendation, Assessment, Development, and Evaluation (GRADE) to assess the quality of evidence for each outcome, in which five domains were evaluated: 1) study limitations were assessed according to RoB2.0; 2) consistency was evaluated using I2 values and the agreement of 95% confidence; 3) indirectness; 4) precision was examined by the optimal information sample size; and 5) publication bias and the number of included studies (Gonzalez-Padilla and Dahm, 2021). Similarly, the quality of evidence by GRADE was also decided with consensus.

After the completion of this study, two researchers (FCN and ZCY), who had no conflict of interest with the study, assessed the methodological quality and risk of bias using the Modified Quality Assessment Scale for Systematic Reviews (AMSTAR-2) (Shea et al., 2017) and Risk of Bias in Systematic Reviews (ROBIS) tool (Whiting et al., 2016). In case of any disagreements, they resolved them through discussion. If necessary, the opinion of a third expert (HHY) was sought to provide further insights. Furthermore, we will refine this meta-analysis according to the results of reviewing until all entries were satisfied.

A total of 1,735 studies were initially searched. Among them, 1,027 were duplicates, and 657 were excluded after reading the title or abstract. Finally, 51 trials were reviewed at the full-text level for further evaluation, of which 12 were excluded (Reasons of excluding studies were provided in Supplementary Table S5), and the remaining 39 studies (Cui et al., 2008; Gao et al., 2009; Wang and Zhang, 2009; Chai, 2010; SUn, 2010; Zhong, 2010; Zhou, 2010; Zhuang and Wu, 2010; Gao and Feng, 2011; Sui, 2011; Yang, 2011; Zhang and Yu, 2011; Zhao and Liu, 2011; Ma and Xi, 2012; Meng et al., 2012; Xu, 2012; Zhang, 2012; Liu and Zhang, 2013; Yuan, 2013; Dai, 2014; He et al., 2014; Ren and Wang, 2014; Zhu, 2014; Li, 2015; Liu and Cheng, 2015; Liu et al., 2015; Wang, 2015; Chen et al., 2016; Wu, 2016; Zhang, 2016; Zhou, 2017; Liu, 2018; Zhang et al., 2018; Sun et al., 2019; Cai, 2021; Wang, 2021; Wang, 2021; He et al., 2022; Zhang, 2022) were included in the final review (Figure 1).

A total of 39 RCTs involving 3,779 patients with CHD were included. All trials were conducted in China from 2008 to 2022. There was a total of 1,920 participants in the experimental groups and 1,859 in the control groups. The average age of participants ranged from 48 to 72 years old. Regarding disease subtypes, one trial only recruited patients with stable angina pectoris (SAP) (Zhang, 2012), and nineteen studies recruited patients with unstable angina pectoris (UAP) (Chai, 2010; Zhou, 2010; Gao and Feng, 2011; Sui, 2011; Yang, 2011; Ma and Xi, 2012; Liu and Zhang, 2013; Yuan, 2013; He et al., 2014; Zhu, 2014; Li, 2015; Liu and Cheng, 2015; Liu et al., 2015; Wang, 2015; Wu, 2016; Liu, 2018; Sun et al., 2019; Wang, 2021; Zhang, 2022). The other studies did not specify the type of CHD. All the studies employed a two-arm design, with SXNI combination therapy as the experimental group and conventional therapy as the control group. In experimental groups, the majority of studies used the intravenous route as the mode of administration for SXNI, and only two studies (Sun et al., 2019; Zhang, 2022) used intramuscular injection; the daily dosage of SXNI ranged from 3 mL to 30 mL, and the duration of treatment spanned from 1week to 1 month. In the control groups, conventional therapy, including anticoagulants, antiplatelets, and receptor blockers, was used. The characteristics of the included studies were indicated in Table 1.

Twelve studies (Zhuang and Wu, 2010; Liu and Zhang, 2013; Li, 2015; Liu et al., 2015; Wang, 2015; Wu, 2016; Zhang, 2016; Zhou, 2017; Zhang et al., 2018; Sun et al., 2019; Wang, 2021; Zhang, 2022) provided a sufficient randomization process to generate random sequences with a low risk of bias; while one RCT (He et al., 2022) was assessed as high risk due to the possibility of patients being grouped based on physician preferences. The remaining studies did not provide specific details of randomization, and therefore we evaluated the risk of bias as unclear. Additionally, none of the studies provided information on allocation concealment, blinding of participants, and outcome assessment, so we evaluated the risk of bias as unclear. All studies included in the analysis published complete data regarding the outcomes, leading us to rate the risk of bias as low. We were concerned about the selection of the reported results from two studies (Sui, 2011; Wu, 2016), and as a result, we evaluated the risk of bias as unclear. Overall, one study had a domain rated as having an extreme risk of bias, all studies had at least one domain rated as having an uncertain risk of bias due to a lack of information (Figure 2).

The GRADE approach was used to systematically assess the quality of 17 outcome indicators (Table 2). The results revealed moderate-quality evidence for clinical efficacy, angina duration, CRP, EGG changes, TC, TG, LDL, PV, HS, and HCT. In addition, low-quality evidence was indicated for the angina attack frequency, BNP, cardiac index, HDL, and FIB. Furthermore, very low-quality evidence was observed for LVEF and LVEDD. The main reasons for downgrading the quality of evidence were the poor methodological quality of the included RCTs, high heterogeneity among the studies, small sample sizes, potential publication bias, and the number of including RCTs.

Two evaluators (FCN and ZCY) who had no conflict of interest with this study used AMSTAR-2 and ROBIS to evaluate the methodological quality and risk of bias in this study. The results showed that no significant methodological errors were found in this systematic review, and the risk of bias was low (Supplementary Table S13, S14).

This study evaluated the efficacy and safety of SXNI combination therapy in treating CHD. A total of 39 trials involving 3,779 patients with CHD were included. The meta-analysis results indicated that SXNI combination therapy had a significant impact on improving clinical efficacy and decreasing the frequency and duration of angina, suggesting that SXNI significantly improve the clinical symptoms and the quality of life of patients with CHD. Furthermore, the results of the cardiac function indexes proved that SXNI combination therapy significantly decreased LVEDD and increased CI, CO, and LVEF, suggesting that SXNI could effectively improve heart function and avoid the development of chronic heart failure in patients with CHD. Given that blood lipids indexes and thrombosis factors are strongly associated with the severity of CHD and all-cause mortality, the meta analysis results indicated SXNI combination therapy had a significant impact on improving blood lipids indexes (e.g., increased HDL, decreased TC, TGD and LDL) and thrombosis factors (e.g., decreased FIB, PV, HCT, and HS), which indicates that SXNI could reduce the severity of disease and improve the quality of life of patients. At the same time, the therapy also resulted in decreased CRP levels and increased ECG efficacy rate and BNP levels. In terms of safety, based on available evidence, we can only conservatively assume that SXNI does not increase the occurrence of adverse reactions. In summary, SXNI could improve the symptoms, signs and quality of life of patients, reduce the severity of CHD, and improve clinical efficacy. However, the level of evidence was very low to moderate due to the potential publication bias and heterogeneity.

Similar to the results of this study, an animal experiment found that SXNI inhibits oxidative stress and endoplasmic reticulum stress (ERS). This, in turn, regulates the TLR4/NF-κB pathway, leading to a reduction in inflammatory responses and a reduced risk of thrombosis (Wang et al., 2019). A cell experiment found that SXNI can reduce the generation of superoxide and inhibit apoptosis, resulting in a reduction in infarct size and an improvement in cardiac function (Li et al., 2019). In addition, SXNI exhibited significant improvements in ST-segment changes in the ECG of rats with myocardial ischemia and reduced the incidence of arrhythmia (Guo et al., 2012). Another animal experiment discovered that ginkgo flavonol glycosides (GFGs), the active ingredient of SXNI, mitigated heart injury in mice by downregulating the tumor necrosis factor (TNF)-like weak inducer of apoptosis/fibroblast growth factor-inducible 14 (TWEAK/Fn14) axis and reducing the generation of oxygen-free radicals (Xiao et al., 2019). Furthermore, A study based on RNA-seq and network pharmacological analysis showed that SXNI effectively protects heart from injuries via a common Tnfrsf12a-mediated pathway involving atherosclerosis signaling and inflammatory response (Lyu et al., 2018). Overall, SXNI can protect cardiomyocytes and reduce the size of the myocardial infarct by regulating blood lipids, reducing oxidative stress, reducing inflammatory factors, and regulating platelet aggregation, thus improving cardiac function and symptoms of CHD.

We assessed the quality of evidence for outcome indicators to guide clinical practice. First, the majority of studies had an acceptable risk of bias, except for one study which was deemed to have an extreme risk of bias. We downgraded the risk level by 1 grade if more than 25% of participants in a comparison were from studies with a high risk of overall bias. Second, heterogeneity was used to evaluate the inconsistency, and significant heterogeneity was found in 12 outcome indicators. Despite the high degree of heterogeneity among the studies, we chose to reduce the level by 1 grade given that the sensitivity analysis confirmed the robustness of the conclusions. Third, no major differences were found in the characteristics (e.g., age, sex, diagnosis, and treatment parameters) of the different studies, and no serious issues of indirectness were found. Fourth, in term of imprecision, we defined the optimal information sample size of the evidence as 400, and we would chose to reduce the level by 1 grade if the total sample size was less than 400 participants. Fifth, the asymmetry of the funnel plot and the results of the Egger’s test indicated potential publication bias in four outcome indicators. Therefore, we reduced the level by 1 grade. Furthermore, if the number of RCTs for an outcome indicator was less than 6, we also reduced the level by 1 grade.

Overall, all outcome indicators had at least one downgrade factor. The quality of evidence for 10 outcome indicators were rated as “moderate”, 5 outcome indicators were rated as “low”, and 2 outcome indicators were rated as “very low”. In addition, two outcome indicators (CO and adverse reactions) were not assessed for the quality of their evidence as only descriptive analyses were performed.

We assessed the impact of treatment duration, dose of SXNI, mode of administration, and subtypes of disease on the clinical efficacy. Notably, the results from subgroup analyses were not fully consistent. Both SAP and UAP are the most representative disease subtypes of CHD. However, we found SXNI combination therapy showed a positive effect in the treatment of UAP but not on SAP. We surmised that the differences in efficacy across disease subtypes could be attributed to the number of included studies (only 1 study recruited patients with SAP, but 14 studies recruited patients with UAP). Therefore, it is crucial to conduct rigorously designed and higher-quality RCTs to confirm the effectiveness of SXNI in the treatment of SAP. Furthermore, our findings indicated that SXNI could enhance the clinical efficacy regardless of the duration of treatment (less than 2 weeks, 2 weeks, or over 2 weeks), SXNI dose (less than 20 mL/day, 20 mL/day, or over 20 mL/day), and mode of administration (intravenous injection or intramuscular injection).

In view of the high heterogeneity of outcome indicators, sensitivity analysis, subgroup analysis, and meta-regression tests were used to eliminate the heterogeneity. We found that the source of clinical heterogeneity in HDL was due to the inconsistency of treatment duration, and the higher risk of overall bias in the included studies was the source of methodological heterogeneity in LVEDD and LVEF. The meta-regression analysis showed that mean age may be the source of heterogeneity in PV, but the age-based subgroup analysis did not show significantly decreased levels of heterogeneity among the studies. Despite our efforts to eliminate heterogeneity, some outcome indicators still exhibited high levels of heterogeneity, prompting us to interpret the results with cautions for clinical practice when involving in the effect of SXNI for CHD by decreasing the frequency and duration of angina, decreasing CRP, improving blood lipids (decrease TC, TGD and LDL) and thrombosis factors (e.g., decrease FIB, PV, HCT, and HS).

Additionally, we identified publication bias in the FIB index. Notably, while the sensitivity analysis confirmed the robustness of the existing conclusions, the trim-and-fill analysis found that some RCTs showing negative results remained unpublished, and the existing conclusions may be overturned when the negative trials are published. Therefore, this conclusion needs to be treated with caution.

Previous SRs also showed that SXNI is effective and safe in treating CHD. However, this study has the following advantages over previous studies. First, our study comprehensively evaluated the efficacy of SXNI in CHD by examining 19 outcome indicators. Second, this study was conducted strictly in accordance with PRISMA. Furthermore, ROBIS and AMSTAR-2 were used to evaluate the risk of bias and methodological quality of this SR, which enhanced the credibility of the findings. Third, subgroup analyses were conducted to explore the influence of certain characteristics on the efficacy of treatment, which could have clinically significant. Fourth, we used trim-and-fill analysis to assess the impact of publication bias on the interpretation of the results. Finally, this study avoided the flaws of previous studies and updated the evidence.

However, this study unavoidably has the following limitations. First, the majority of the included studies had a risk of bias which reduced the credibility of the evidence. Second, considering that traditional Chinese medicine is also popular in other countries (Wang et al., 2023), while the studies included in this SR were all conducted in China which may lead to bias. Third, although sensitivity analyses confirmed the robustness of these conclusions, existing conclusions need to be treated with caution due to heterogeneity and publication bias. These issues were not fully addressed and could affect the reliability of the results. Fourth, although long-term efficacy is an important part of clinical evaluation, our study did not evaluate the long-term efficacy of SXNI. Notably, previous studies have shown that administration of Shuxuening in combination with potassium aspartate and magnesium, atorvastatin calcium, Shengmai injection, pantoprazole sodium, or high-dose medication was a risk factor for suspected allergic reactions. Furthermore, the incidence of allergic reaction was also influenced by the vehicle, duration of treatment, single dose, and indicated vs. off-label use (Wang et al., 2018). However, due to the different criteria used to evaluate adverse events in each study, reliable safety conclusions could not be reached.

Based on the conclusions and limitations of this SR, several beneficial and feasible suggestions can be proposed for future research.

Given the inconsistent findings for different subtypes of CHD, future studies should pay more attention to evaluating the efficacy of SXNI, specifically in subtypes such as SAP. Considering the high mortality rate of CHD, future trials should take long-term efficacy as a part of the outcome indicators. In addition, since this study did not draw an accurate safety conclusion, future studies should standardize the monitoring and recording of adverse reactions. Notably, the trim-and-fill analysis found that several RCTs showing negative findings remained unpublished, which could influence the existing findings. Therefore, the bias of selective reporting should be avoided in the future research. Finally, the effects of treatment duration, mode of administration, and dose of SXNI need to be further investigated.

In terms of clinical study design, more large-sample, multi-center, long-period RCTs should be conducted, and strictly follow the Consolidated Standards of Reporting Trials (CONSORT) guidelines. In addition, the sample size should be reasonably estimated, random allocation and allocation concealment methods should be selected, and blind method should be implemented as far as possible. At the same time, the selection of outcome indicators should be able to fully evaluate the efficacy and safety of SXNI, and the measurement of outcome indicators should use a unified standard.