A comprehensive and systematic literature search was conducted across eight electronic databases: PubMed, Embase, Web of Science, the Cochrane Library, China National Knowledge Infrastructure (CNKI), Wanfang Database, VIP Database, and the China Biomedical Literature Database (CBM). The search covered all articles published from the inception of each database up to 28 February 2025.

The search strategy was developed to be highly sensitive, combining controlled vocabulary terms (e.g., MeSH and Emtree) with free-text keywords related to two core concepts: “salvianolic acid A” and “ischemic stroke.” The search strategy for PubMed is provided as an example: (((“Salvianolic Acid A” [MeSH Terms] OR “SalA” [All Fields])) AND ((“Ischemic Stroke” [MeSH Terms] OR “Brain Ischemia” [MeSH Terms] OR “MCAO” [All Fields]))). This strategy was adapted for the syntax of each of the other databases. For Chinese databases, the search strategy utilized the Chinese translations of the keywords. The complete, detailed, and reproducible search strategies for all eight databases are provided in Supplementary Material l.

No language restrictions were applied during the initial search phase, but only articles published in English or Chinese were included for full-text screening. Two authors independently performed the literature search, and the results were cross-checked to ensure completeness.

The inclusion criteria are as follows: (1) studies involving rats or mice as subjects, with the establishment of MCAO or I/R models; (2) the experimental group receives SalA intervention post-surgery (selecting the group with the best therapeutic effect if different concentrations are used); (3) the model group receives a placebo or no treatment; (4) no restrictions on animal species, gender, age, weight, or sample size; (5) the primary outcome measures include neurological function scores, infarct area or proportion, and all biomarkers indicating IS.

The exclusion criteria for the literature are strictly defined as follows: (1) animal models of non-cerebral ischemia or global cerebral ischemia; (2) reviews, in vitro studies, and trials; (3) studies in which SalA was used to intervene in other diseases or was not used at all; (4) studies with insufficient data, including unpublished data; and (5) duplicated studies (Figure 2).

The data extraction table is completed by two independent individuals to ensure accuracy. Additionally, the reference list includes the following information: (1) first author’s name and year of publication; (2) detailed information about the animal model, such as species, gender, and weight; (3) induction method of MCAO and I/R animal models; (4) intervention group: drug used, dosage, and administration method; (5) control group: type, dosage, time, and duration of the nonfunctional substance given; (6) main conclusions of the study; and (7) outcome measures. When animals in the experimental group receive different doses, only the highest dose is recorded. If data are presented in a graphical form, they are measured using digital ruler software. When data are incomplete, we attempt to contact the authors to obtain further information.

Two research workers independently assessed the quality of the literature using the SYRCLE animal experiment bias risk assessment tool within Review Manager 5.4 software, evaluating 10 criteria: selection bias (sequence generation, baseline characteristics, and allocation concealment), performance bias (random housing and blinding), detection bias (random outcome assessment and blinding), attrition bias (incomplete outcome data), reporting bias (selective outcome reporting), and other potential sources of bias. Each criterion was rated as follows: low risk (method correctly applied), unclear risk (method application unclear), or high risk (method incorrectly applied or not applied). The assessments were carried out independently by both research workers and cross-checked. In case of disagreements, the research workers discussed and reached a consensus; if no consensus was achieved, a third research worker was consulted to make the final decision.

Each item was scored one point if it was met, for a maximum quality score of 10. Discrepancies in scoring were resolved through discussion. We generally consider a score above 6 to be the average score. Importantly, no studies were excluded based on their quality score. The risk-of-bias assessment was instead used to evaluate the overall quality of the evidence base and to inform subsequent sensitivity analyses to investigate potential sources of heterogeneity.

Given that the outcome measurements in this study are continuous variables, the standardized mean difference (SMD) with a 95% confidence interval (95% CI) was used to combine the effect size. When p < 0.05, it indicates a significant difference between the intervention and control groups. The heterogeneity evaluation metric used is I-squared (I ² ). When I ² < 50%, it indicates that the studies retrieved are homogeneous, and a fixed-effect model is used to combine the effect size. When I ² > 50%, it indicates that the quality of the studies retrieved is heterogeneous, and a random-effects model is applied to combine the effect size. To identify potential causes of heterogeneity, subgroup analyses were performed based on animal species, methods, and stages of administration. To assess potential publication bias, the Egger test was used. Additionally, Origin 2021 was used for dose–response interval analysis.

A total of 15 animal experiments from 2016 to 2025 were retrieved, all published in English journals, with a total of 564 animals included: 282 in the intervention group and 282 in the control group. The animal species included SD rats (Feng et al., 2017; Huang and Wang, 2023; Li J. et al., 2025; Ling et al., 2021; Liu et al., 2020; Song et al., 2019; Yang M. Y. et al., 2024; Yang R. et al., 2024; Zhang S. et al., 2022; Zhang W. et al., 2016; Zhang W. et al., 2018; Zhao et al., 2020), Wistar rats (Yang et al., 2022), C57BL/6 mice (Mahmood et al., 2017), and ICR mice (Chien et al., 2016; Yang M. Y. et al., 2024). Anesthesia was induced using pentobarbital sodium (Huang and Wang, 2023; Li J. et al., 2025; Yang et al., 2022), chloral hydrate (Feng et al., 2017; Ling et al., 2021; Liu et al., 2020; Mahmood et al., 2017; Zhang W. et al., 2016; Zhang W. et al., 2018), isoflurane (Feng et al., 2017; Ling et al., 2021; Liu et al., 2020; Mahmood et al., 2017; Zhang W. et al., 2016; W. Zhang et al., 2018), and an unspecified anesthetic (Zhao et al., 2020; Table 1).

All animal models used in this study were described according to the inclusion and exclusion criteria from the literature, reflecting the pathological process of IS. The intervention group was treated with SalA monotherapy. The control group received an equal volume of physiological saline, drug carrier treatment, or no treatment. In evaluating the effect of SalA on IS, 12 studies reported cerebral infarction volume, nine studies reported NODs, seven studies reported brain edema volume, five studies reported tumor necrosis factor-a (TNF-α), five studies reported interleukin 6 (IL-6), five studies reported IL-1β, three studies reported Caspase-3, two studies reported Bcl-2-associated X protein (Bax)/β-actin, two studies reported B-cell lymphoma-2 (Bcl-2)/β-actin, two studies reported Bcl-2/Bax, two studies reported the expression of Caspase-3 protein, two studies reported ZO-1/β-actin, two studies reported occludin/β-actin, two studies reported p-AKT/AKT, and two studies reported the positive expression rates of neuronal nuclei antigen (NeuN) and terminal urine nucleotide extraction ligation (TUNEL) in cells (Table 1).

The methodological quality of the 15 included studies was evaluated using the SYRCLE animal experiment bias risk assessment tool to identify both strengths and areas for improvement in the current body of preclinical literature. The results, summarized in Figure 3, indicate a consistent adherence to certain reporting standards, with total quality scores for the studies ranging from 5 to 7 out of a possible 10.

Encouragingly, all included studies were published in peer-reviewed journals and reported compliance with animal welfare regulations, establishing a foundational level of scientific rigor. Moreover, a majority of experiments demonstrated good laboratory practice by reporting temperature control (80%) and using anesthetics without significant intrinsic neuroprotective activity (87%). These aspects enhance the reliability of the reported findings. However, the assessment also highlighted several opportunities for methodological enhancement in future preclinical research. Key areas such as sample size calculation, allocation concealment, and blinded assessment of outcomes were not consistently reported across the studies (0%, 0%, and 27%, respectively).

Although these methodological gaps are common in exploratory preclinical research and represent a known challenge in the field, they also underscore the importance of a systematic review. By synthesizing data from these studies, our meta-analysis provides a comprehensive overview of the current evidence landscape. The identified risks of bias, particularly in selection and detection, are considered significant potential sources of the statistical heterogeneity observed. Acknowledging these limitations allows for a more nuanced interpretation of the pooled results and provides clear, actionable recommendations for improving the design and reporting of future animal studies in this area. Therefore, the current body of evidence was deemed suitable for a meta-analysis aimed at summarizing existing findings and guiding future, more robust research.

The literature search initially identified 898 records. After removing duplicates and screening titles and abstracts, 281 full-text articles were assessed for eligibility. This process resulted in the final inclusion of 15 preclinical studies that met all criteria (Figure 2). The key characteristics of these 15 studies, which involved a total of 564 animals, are summarized in Table 1. The methodological quality assessment, as detailed in Figure 3, revealed that although all studies adhered to certain reporting standards, significant risks of bias were prevalent, particularly in the domains of sample size calculation, allocation concealment, and blinded outcome assessment, which are critical for the interpretation of the following results.

The meta-analysis demonstrated that SalA treatment led to significant improvements in the three primary measures of ischemic brain injury. The main findings of the meta-analyses for all outcomes are summarized in Table 2.

The analysis of 12 studies confirmed that SalA positively reduced the infarct area in IS compared to the control group [SMD = −4.67, 95% CI = (−5.98, −3.36), p < 0.00, n = 223, and n (treatment/model) = 112/111]. Due to significant heterogeneity between studies (I ² = 87.8 and p < 0.001), the heterogeneity for this analysis was I ² = 85.5%, and a subgroup analysis based on the administration method was performed (Figure 6A).

Subgroup analysis based on the administration method showed a positive reduction in heterogeneity. In the intravenous injection subgroup, heterogeneity remained high (I ² = 76.8% and p < 0.001), but SalA still showed positive effects in the intravenous injection group [SMD = −3.91, 95% CI = (−5.17, −2.65), p < 0.001, and n = 130]. In the gastric gavage subgroup, heterogeneity decreased (I ² = 74.3% and p < 0.05), and SalA showed positive effects with this administration method [SMD = −3.82, 95% CI = (−6.87, −0.77), p < 0.001, and n = 24]. In the tail vein injection subgroup, n = 20/22 (treatment/model), heterogeneity was high (I ² = 92.6% and p < 0.001), but SalA still showed positive effects with this administration method [SMD = −7.39, 95% CI = (−18.60, 3.82), p < 0.001, and n = 42]. As only one study was included, heterogeneity was not tested in the intraperitoneal injection group [SMD = −18.35, 95% CI = (−23.83, −12.87), p < 0.001, and n = 24]. Despite high heterogeneity in the overall analysis, the effects of SalA were particularly positive in the intravenous and tail vein injection subgroups, suggesting that different administration methods may have an impact on the therapeutic effects (Figure 6B).

According to the meta-analysis of nine studies, SalA showed positive effects in alleviating the neurological deficit scores in IS. Compared to the control group, SalA reduced the neurological deficit scores [SMD = −6.39, 95% CI = (−9.091, −3.688), p < 0.001, and n = 164]. However, due to significant heterogeneity between studies (I ² = 66.8% and p < 0.001), we performed a subgroup analysis based on different drug concentrations (20 mg/kg, 10 mg/kg, and 5 mg/kg) (Figure 7A).

Subgroup analysis based on drug concentrations: 20 mg/kg subgroup: this group had high heterogeneity (I ² = 71.5% and p = 0.007), showing considerable variability between studies. SalA positively reduced the neurological deficit scores at the concentration of 20 mg/kg [SMD = −3.130, 95% CI = (−4.307, −1.954), p < 0.001, and n = 82]. This subgroup accounted for 54.88% of the total analysis; 10 mg/kg subgroup: this group had lower heterogeneity (I2 = 52.5% and p = 0.122). SalA also positively reduced the neurological deficit scores at the 10 mg/kg concentration [SMD = −1.893, 95% CI = (−2.952, −0.835), p < 0.001, and n = 50]. This subgroup accounted for 36.82% of the total analysis; 5 mg/kg subgroup: only one study was included in this group, with zero heterogeneity (I ² = 0%), indicating almost no variability. However, SalA still significantly reduced the neurological deficit scores [SMD = −3.823, 95% CI = (−5.844, −1.801), p < 0.001, and n = 32] (Figure 7B). The results of this meta-analysis show that SalA significantly improved the neurological deficit scores in IS models across different concentration groups, and the overall effect was statistically positive. Heterogeneity varied at different concentrations, with the 20 mg/kg group showing considerable variation between studies. Future research may need to further explore the relationship between drug concentration and therapeutic effects, in particular, the effects and heterogeneity at higher concentrations.

The analysis of seven studies showed that SalA had a positive effect in reducing the brain edema area after IS injury compared to the control group [SMD = −5.291, 95% CI (−7.607, −2.975), p < 0.001, and n = 88]. Due to significant heterogeneity between groups (I ² = 83.5% and p < 0.001) (Figure 8A), we performed a subgroup analysis based on animal models for the seven studies. The results showed that SalA positively reduced the brain edema area in all model groups [MACO group: SMD = −6.82, 95% CI = (−13.55, −0.10), p < 0.001, and n = 46 (weight = 39.48%); tMACO group: SMD = −5.1, 95% CI = (−9.56, −0.64), p < 0.001, and n = 18 (weight = 28.49%)]. The I/R group and MACO/R group each included only one study, with zero heterogeneity, but still showed positive therapeutic effects [MACO/R group: SMD = −5.54, 95% CI = (−8.23, −2.87), p < 0.001, and n = 12 (weight = 15.09%); I/R group: SMD = −3.26, 95% CI = (−5.08, −1.44), p < 0.001, and n = 12 (weight = 16.94%)] (Figure 8B).

SalA demonstrated a strong anti-inflammatory effect by significantly reducing the levels of key pro-inflammatory cytokines. The pooled analysis of five studies showed a marked reduction in TNF-α (n = 70; SMD = −4.14 and 95% CI = [−6.52, −1.77]; Supplementary Figure S1) and IL-6 (n = 56; SMD = −5.72 and 95% CI = [−8.71, −1.38]; Supplementary Figure S2). Similarly, an analysis of another five studies confirmed a significant decrease in IL-1β levels (n = 70; SMD = −4.77 and 95% CI [−6.75, −2.79]; Supplementary Figure S3). All these analyses exhibited very high heterogeneity (I ² > 70%), likely stemming from variations in the timing of sample collection and the specific assay methods used across studies.

The meta-analysis of apoptosis-related proteins indicated a consistent and significant antiapoptotic effect of SalA. The pooled data from two studies (n = 18) showed that SalA treatment significantly upregulated the Bcl-2/Bax ratio (SMD = 6.532 and 95% CI = [3.93, 9.13]; Supplementary Figure S6). Akt, as a key component of the antiapoptotic pathway, is influenced by the therapeutic effects of SalA, which affect the ratio of p-Akt/Akt. In two studies, the following data were found: n = 20; SMD = 3.447, 95% CI = [1.96, 4.93], and p < 0.001 (Supplementary Figure S7). Complementing this finding, data from three studies (n = 30) revealed a significant reduction in Caspase-3 levels (SMD = −3.495 and 95% CI = [−4.60, −2.39]; Supplementary Figure S8). Notably, both of these analyses exhibited low-to-no heterogeneity (I ² = 0%), suggesting a more robust and consistent effect of SalA on these specific molecular targets.

The analysis of tight junction proteins, key indicators of BBB integrity, showed that SalA had a positive and significant effect on ZO-1 expression based on two studies (n = 24; SMD = 4.97 and 95% CI = [3.20, 6.74]; I ² = 36.5%; Supplementary Figure S9). In contrast, for occludin, the pooled effect from two studies was not statistically significant (SMD = 1.45 and 95% CI = [−0.39, 3.34]; I ² = 72.6%; Supplementary Figure S10).

Publication bias for the three primary outcomes was visually inspected using funnel plots (Figures 9A–C) and formally tested with Egger’s regression test. No significant publication bias was detected for cerebral infarct volume (t = −8.88 and p > 0.05), NDS (t = −4.3 and p > 0.05), or brain edema (t = −3.42 and p > 0.05); however, the power of these tests is limited by the small number of studies in each analysis.

To evaluate the stability of the primary outcome measures (including cerebral infarction area, NDS, and cerebral edema volume), a sensitivity analysis was conducted. The analysis showed no significant differences in the results, indicating the robustness of the study findings (Supplementary Figure S11).

To our knowledge, this is the first systematic review and meta-analysis to quantitatively synthesize the preclinical evidence for SalA in treating ischemic stroke. Our pooled results demonstrate that SalA administration in animal models leads to a statistically significant reduction in cerebral infarct volume, neurological deficits, and brain edema. Furthermore, our analyses provide quantitative support for the multi-target nature of SalA, showing consistent modulatory effects on key pathways of inflammation, apoptosis, and blood–brain barrier integrity. However, these promising findings must be interpreted with considerable caution. The certainty of this evidence is significantly limited by the high risk of bias identified across the included studies—particularly, the universal lack of sample size calculation and blinding—and the substantial statistical heterogeneity observed in most of our primary analyses.

Our meta-analysis confirmed that SalA’s neuroprotective effects are associated with potent anti-inflammatory and antiapoptotic activities. The significant reduction in pro-inflammatory cytokines such as TNF-α, IL-6, and IL-1β (Supplementary Figures S1-S3) provides robust evidence for its anti-inflammatory role. Several included studies attributed this effect to the inhibition of pathways, including NF-κB (Zhang et al., 2018). Similarly, the strong and consistent effect on increasing the Bcl-2/Bax ratio and decreasing Caspase-3 levels (Supplementary Figures S6-S8), which notably showed low heterogeneity, underscores a reliable antiapoptotic mechanism. These findings suggest that SalA may act by rebalancing the cellular response to ischemic stress, shifting it away from cell death and toward survival. The positive effect on the tight junction protein ZO-1 (Supplementary Figure S9) further suggests that SalA contributes to the restoration of blood–brain barrier integrity, a critical step in mitigating secondary injury.

Taken together, our analysis of molecular markers suggests that SalA exerts its neuroprotective effects through a multi-pronged approach, targeting key pathological cascades in ischemic stroke. We have integrated these findings into a schematic diagram that illustrates the potential interconnected mechanisms of action, highlighting SalA’s role in modulating inflammation, apoptosis, and blood–brain barrier stability (Figure 10).

A critical finding of this review is the overall moderate-to-low methodological quality of the current preclinical evidence base, as detailed in our risk-of-bias assessment (Figure 3). The prevalent lack of randomization, allocation concealment, and blinding introduces a high risk of selection and performance bias, which likely contributes to the overestimation of treatment effects. This low methodological quality is also a primary driver of the high statistical heterogeneity (I ² > 70%) observed in our main results. Although our exploratory analyses (Figure 4) suggested that factors such as animal model and timing of administration may contribute to this variability, the limited number of studies precluded a formal meta-regression to quantify these sources. This underscores an urgent need for the field to adopt more rigorous experimental designs and adhere to reporting guidelines, such as ARRIVE, to enhance the reliability and reproducibility of preclinical stroke research.

This study is the first to integrate animal experimental evidence of SalA through a systematic review and meta-analysis, clarifying its multi-target effects. Although previous studies have explored the role of Salvia miltiorrhiza acid A in ischemic stroke, there is a lack of dedicated systematic reviews or meta-analyses. Therefore, our manuscript fills this gap by systematically summarizing the research progress on Salvia miltiorrhiza acid A in this field. Through meta-analysis, we evaluate the effects of different doses and timing of administration on its therapeutic efficacy, providing guidance for clinical application. Additionally, we identify potential biases and shortcomings in the existing studies through quality assessment, offering directions for future research improvements.

This meta-analysis has several limitations. First, as discussed, the conclusions are based on primary studies with significant methodological flaws. Second, the lack of systematic safety reporting in the included studies prevented any quantitative assessment of SalA’s adverse effects. Third, although, in our study, Egger’s tests did not detect significant publication bias, this may be due to the limited number of studies, and the potential for bias cannot be entirely ruled out. Consequently, the immediate clinical translation of SalA based on this evidence would be premature. Instead, these findings should serve as a strong impetus for conducting future, higher-quality preclinical studies. Future research should prioritize the following: (1) the use of standardized, blinded, and randomized study designs; (2) perform a priori sample size calculations; (3) systematically report on safety and adverse events; and (4) directly compare different dosages and administration windows to define an optimal therapeutic protocol.