We conducted a comprehensive search of four electronic databases, including PubMed, Cochrane Library, Web of Science, and Scopus, spanning from the inception of these databases until December 30, 2024. The search imposed no restrictions regarding the year, gender, or geographical location of studies; however, it was confined to articles published in English. The keywords utilized in the search encompassed “edaravone-dexborneol” and “acute ischemic stroke (AIS),” employing the following search strategy: (“Edaravone Dexborneol”[Title/Abstract] OR “Edaravone and Dexborneol”[Title/Abstract] OR “dual neuroprotection”[Title/Abstract]) AND (“acute ischemic stroke”[MeSH] OR “ischemic stroke”[Title/Abstract]) AND (“efficacy”[Title/Abstract] OR “safety”[Title/Abstract] OR “outcome”[Title/Abstract]).

We included the studies according to the following inclusion and exclusion criteria:

Inclusion criteria:

Exclusion criteria: Reviews, conference abstracts, and non-English language studies were excluded. There were no other restrictions on the study design or population.

The eligibility screening process was conducted in two phases: first, the titles and abstracts were screened, followed by the retrieval of full-text articles for those abstracts deemed eligible to undergo further assessment against specified eligibility criteria for the meta-analysis. Any disagreements that arose were addressed through discussion. For data extraction, an online data extraction template was developed based on a pre-specified uniform data extraction sheet, encompassing the following domains: (1) characteristics of the included studies, (2) characteristics of the study population, (3) risk of bias domains, and (4) outcome measures.

The revised Cochrane risk-of-bias tool (RoB 2) for RCTs was employed to assess the risk of bias in the included clinical trials (18). This thorough evaluation addressed critical domains, including the randomization process, sequence allocation concealment, deviations from intended interventions, and the implementation of suitable statistical methods to evaluate intervention effects. Furthermore, it examined outcome measurement, reporting biases, and the overall risk of bias. Studies were categorized according to their methodological quality as exhibiting low risk, some concerns, or high risk of bias. Any discrepancies that arose were resolved by a final determination made by the senior author. For observational studies, the Newcastle-Ottawa Scale (NOS) was utilized (19). This tool allows researchers to adopt a point-based system to classify studies as ‘good,’ ‘fair,’ or ‘poor.’ Disagreements were addressed through dialogue, and a third author was consulted when necessary to ensure consensus.

Whenever data were presented as medians and interquartile ranges (IQR), they were converted to means (M) and standard deviations (SD) utilizing the equations established by Wan et al. (20). In instances where the SD was not provided, it was derived from the standard error using the formula for a single sample: SD = SE * √n (21), where n represents the sample size. In cases where the mean change (MC) between baseline and endpoint was unavailable, it was computed from the pre-treatment and post-treatment means using the equation MC = Mpost-treatment - Mpre-treatment. Furthermore, if the standard deviation of the mean change was not specified, it was calculated from the pre-treatment and post-treatment standard deviations employing the formula SD_change = √(SD_pre² + SD_post²–2*r*SD_pre*SD_post) (22). Missing effect sizes (d) and effect size correlations (r) were determined through the following formulas: d = (Mtreatment - Mcontrol)/SDpooled, where SDpooled = √[(SD²treatment + SD²control)/2], (23) and r = d/√(d² + 4) (24).

A thorough visual inspection of the final forest plots was conducted, complemented by an assessment utilizing I-square and Chi-Square tests, specifically Cochran’s Q test, to evaluate the extent of heterogeneity. In instances of significant heterogeneity, defined as Chi-Square p < 0.1, a sensitivity analysis was executed to address this heterogeneity. This analysis was performed using R programming version 4.4.1 for Windows (25).

A variety of statistical methods were utilized to evaluate publication bias. The Fail-Safe N test, employing the Rosenthal methodology, was conducted to ascertain the number of additional null studies necessary to render the overall meta-analysis results non-significant. Begg and Mazumdar’s rank correlation test was implemented to assess the relationship between effect sizes and their standard errors, thereby identifying potential asymmetry within the funnel plot. Additionally, Egger’s regression test was performed to evaluate the linear relationship between effect size and precision, offering further insight into potential funnel plot asymmetry (Table 1 and Figure 1). Collectively, these methods provide a thorough assessment of potential publication bias in the analysis (26–28).

The analysis was conducted utilizing the standardized mean difference as the primary outcome measure. A random effects model was employed to fit the data. The degree of heterogeneity, represented as tau², was estimated using the restricted maximum-likelihood estimator (Viechtbauer, 2005). In addition to the estimation of tau², the Q-test for heterogeneity (Cochran, 1954) and the I² statistic were also reported. Should any degree of heterogeneity be detected (i.e., tau² > 0, independent of the Q-test results), a prediction interval for the true outcomes is provided.

To assess potential outliers and influential studies within the model context, studentized residuals and Cook’s distances were analyzed. Studies exhibiting a studentized residual exceeding the 100 × (1–0.05/(2 × k))th percentile of a standard normal distribution were classified as potential outliers, employing a Bonferroni correction with a two-sided alpha level of 0.05 for k studies included in the meta-analysis. Furthermore, studies with a Cook’s distance exceeding the median plus six times the interquartile range of Cook’s distances were deemed influential.

Rank correlation and regression tests were utilized to evaluate funnel plot asymmetry, employing the standard error of the observed outcomes as predictors. All analyses were executed using R Statistical Software (version 4.1.2; R Core Team, 2021), while the risk of bias assessment was conducted using Jamovi version 2.6 for Windows.

Since this study utilized publicly available data and did not involve direct interaction with human or animal subjects, ethical approval was not necessary. Furthermore, this study was registered with PROSPERO, assigned the identification number CRD42024626320.

Our search initially identified 344 potentially relevant studies. After removing 154 duplicates, 190 records were retained for title and abstract screening. Following this screening, 31 studies were deemed eligible for a full-text review. Of these, six randomized controlled trials (RCTs) and one observational (6, 7, 14, 29–32) encompassing a total of 2,942 patients met the inclusion criteria and were included in the analysis. Twenty publications were excluded during full-text screening: three RCTs did not meet the outcome inclusion criteria, and one publication was only available as an abstract. Further details are provided in Figure 2.

Of the seven included studies, six were parallel RCTs and one cohort study conducted in China that investigated the effects of Edaravone and Dexborneol in ischemic stroke patients (6, 7, 14, 29–32). Six RCTs and one cohort with 2,942 patients were included, of whom 1931 (65.64%) were males. The treatment regimens involved intravenous and sublingual administration of Edaravone and Dexborneol, either alone or in combination with other standard ischemic stroke treatments, for durations ranging from 10 to 14 days. The participant numbers varied from 85 to 1,200, with a mean age of 60–68. These studies consistently reported significant improvements in functional outcomes, including enhanced modified Rankin Scale (mRS) scores and reduced National Institutes of Health Stroke Scale (NIHSS) scores, indicating better stroke recovery. Xu et al. (29) and Li et al. (5, 30) also observed improved activities of daily living (ADL). In contrast, several studies, such as Hu et al. (31) and Xu et al. (29), noted reductions in post-stroke depression (PSD), inflammation, and hemorrhagic transformation. Improvements in cognitive function, including better MoCA and Stroke Impact Scale scores, have also been highlighted in several trials. Additionally, the studies demonstrated a favorable safety profile, with fewer adverse events compared to standard treatments, as noted by Xu et al. (6).

Notably, the included studies varied in dosing regimens, administration routes, and concomitant therapies. Most trials administered edaravone dexborneol intravenously at 37.5 mg twice daily for 10–14 days, although some explored higher (62.5 mg) or lower (12.5 mg) IV doses (14), and one large RCT (7) evaluated a 36 mg sublingual formulation. Comparator groups also differed, ranging from placebo or standardized ischemic stroke care to active comparators such as edaravone alone (6, 14) and alteplase therapy (5, 30). These variations in regimen and background therapy may partially account for the heterogeneity observed in NIHSS outcomes (Table 2).

After the quality evaluation of the seven included studies, there were two low-risk studies (6, 7), two with some concerns (14, 29), and two high-risk studies (30, 31). The quality evaluation form according to the RoB2 scale (the risk of bias was classified into three levels: “low risk of bias, “some concerns,” and “high risk of bias”), as shown in Figure 3. The observational study by Chen et al. (2024) (32) was of good quality, up to the NOS score (9/9).

Investigations into the use of edaravone dexborneol for acute ischemic stroke (AIS) consistently revealed a favorable safety profile, characterized by mild adverse effects and a low incidence of severe complications (Table 3). Fu (2024) (7) reported similar rates of adverse events between the sublingual edaravone dexborneol group (89.8%) and the placebo group (90.1%), indicating a comparable safety margin. Li (2024) (30) noted that common side effects, including nausea, vomiting, dizziness, chest tightness, fever, and headache, occurred at comparable rates in both the edaravone dexborneol injection and control groups, further underscoring its tolerability. Xu (2019) (14) identified 29 serious adverse events, such as pruritus, skin rash, acute liver injury, and kidney damage, with two severe cases associated with high-dose edaravone dexborneol; however, no significant differences in serious adverse events were observed among the treatment groups. Hu (2023) (31) demonstrated a safety advantage of edaravone dexborneol by significantly reducing the incidence of hemorrhagic transformation (20.29% vs. 39.73% in the control group), a serious complication of AIS. Similarly, Xu (2024) (29) reported only mild adverse effects, such as dizziness and nausea, with no serious reactions noted, whereas Xu et al. (2021) (6) found similar rates of adverse events between the edaravone dexborneol and edaravone groups (54 vs. 47 patients with serious adverse effects, respectively). Collectively, these findings reinforce the consistent safety outcomes associated with edaravone dexborneol, with adverse effects typically being mild, dose-dependent, and on par with those of standard treatments, highlighting its acceptability in the management of AIS.

The quality of evidence for the outcomes assessed in this review was evaluated using the GRADE of Recommendations Assessment, Development, and Evaluation framework. Among the included studies, the certainty of the evidence ranged from high to very low. The primary outcome, the National Institutes of Health Stroke Scale (NIHSS), was rated as moderate, and the modified Rankin Scale (mRS) was rated as moderate due to the combination of RCTs and observational studies and the moderate quality of overall bias. Overall, the strength of evidence was downgraded for factors such as the Risk of Bias, Inconsistency, Indirectness, Imprecision, and Publication Bias, highlighting the need for further high-quality studies to confirm these findings.

Although this review highlights the promise of edaravone dexborneol as a neuroprotective agent, several important limitations must be acknowledged. First, the small number of available trials substantially limits the robustness and statistical power of the pooled findings, making definitive conclusions premature. Second, considerable heterogeneity existed across studies with respect to dosages, administration routes (intravenous versus sublingual), and comparator treatments. These differences may have influenced the observed variability in outcomes, particularly NIHSS scores, and reduced the certainty of the overall effect estimates. Determining the optimal dosage and administration route through standardized protocols should be a priority for future trials.

Another limitation is that most included studies assessed outcomes only up to 90 days post-treatment, leaving uncertainty about the long-term efficacy and safety of edaravone dexborneol. Extended follow-up durations (e.g., 6–12 months) are necessary to better understand its sustained benefits and risks. Additionally, subgroup analyses stratified by stroke severity, comorbidities, and demographic factors are needed to clarify which patient populations are most likely to benefit. The lack of head-to-head comparisons with other neuroprotective or adjunctive therapies also limits the ability to contextualize edaravone dexborneol within the broader stroke management landscape.

Perhaps the most significant limitation is the geographic concentration of all trials in China. This regional restriction raises concerns regarding external validity and generalizability to more diverse populations, health systems, and clinical practices. To strengthen confidence in these findings, large, high-quality, multicenter RCTs across different countries and ethnic groups are urgently required.

Finally, the overall quality of future studies must be improved. Rigorous randomization, proper blinding, and adherence to standardized reporting guidelines are essential to minimize bias and enhance reliability. Until such data are available, the current evidence should be interpreted with caution, recognizing both its promise and its significant limitations.