Description of studies
Three thousand four hundred thirty-five potentially eligible studies were obtained based on the search strategy. The flow chart describing the screening process is shown in Fig. 1. According to the predetermined criteria, a total of twenty-four RCTs[27–50] involving 2,522 ischemic stroke patients published between 2008 and 2021 were eligible for inclusion in the meta-analysis. The intervention of the control group was CT alone, which was provided by the above Chinese guidelines for the diagnosis and treatment of acute ischemic stroke 2018. The intervention of the trial group was the XNJ and CT. The main characteristics of those included studies are listed in Table 1.
Risk of bias assessment
Twenty-four RCTs were included in the meta-analysis, of which thirteen[27, 28, 30, 31, 34–36, 41–43, 47–49] mentioned the specific way of random sequence generation and judged that the selection bias was low-risk level. Eleven studies[29, 32, 33, 37–40, 44–46, 50] mentioned the word "random". However, they did not say the specific way of random sequence generation, and the selection bias was determined as an unclear risk level. All studies did not note the allocation concealment method and blind method, and the risk was defined as "uncertain". The outcome data of all RCTs were complete, and the risk of incomplete outcome was determined as low risk. One RCT[37] did not report pre-treatment TNF-α data, and IL-6 data, reporting bias was defined as high risk; One RCT[49] did not report NIHSS score and Barthel index data before treatment, and the report bias judgment was determined as high risk; Other studies did not report bias and were judged as low risk. No other bias was found, and other biases of all RCTs were regarded as low risk. The risk assessment of included literature is shown in Fig. 2.
Meta-analysis results
Overall response rate. A total of twenty-one RCTs[27–29, 31–33, 35, 37–50] reported the overall response rate. Heterogeneity analysis results (p = 0.15, I2 = 24.29%) suggested that twenty-one studies were homogeneous, so the fixed effect model was adopted for analysis. The statistical results manifested a higher clinical total effective rate of XNJ combined with CT than CT (RR = 1.15, 95%CI [1.12, 1.19], p = 0.00). The forest plot of the overall response rate is shown in Fig. 3.
NIHSS score. Eighteen RCTs[27–34, 36, 38–41, 44, 46–50] reported the differences in NIHSS score between the trial group and the control group. Heterogeneity analysis results (p = 0.00, I2 = 95.85%) indicated heterogeneity among 18 studies, so the random effect model was used for analysis. The statistical results suggested that XNJ could effectively improve neurological deficit (WMD=-3.38, 95%CI [-4.35, -2.40], p = 0.00). The forest plot of the NIHSS score is shown in Fig. 4. Meta-regression and subgroup analysis were used to explore the sources of heterogeneity.
Barthel index. Eight RCTs[29, 39, 40, 43, 44, 46, 49, 50] reported Barthel index differences between the trial and control groups. Heterogeneity analysis results (p = 0.01, I2 = 64.08%) indicated heterogeneity among 8 RCTs, so the random effect model was used for analysis. The statistical results suggested that the Barthel index of the trial group was higher than that of the control group (WMD = 8.19, 95%CI [5.70, 10.68], p = 0.00). The forest plot of Barthel index is presented in Fig. 5 (A).
TNF-α. Nine RCTs[27, 31, 32, 36, 37, 39, 41, 43, 44] reported the TNF-α. Heterogeneity analysis results (p = 0.00, I2 = 97.39%) indicated heterogeneity among 9 RCTs, so the random effect model was used for analysis. The statistical results show that the reduction of TNF-α level in the trial group was more significant than in the control group (SMD=-2.75, 95%CI [-3.67, -1.84], p = 0.00). The forest plot of total TNF-α was presented in Fig. 5 (B).
IL-6. Nine RCTs[27, 31, 32, 36, 37, 39, 43, 45, 48] reported the IL-6. Heterogeneity analysis results (p = 0.00, I2 = 96.74%) indicated heterogeneity among 9 RCTs, so the random effect model was used for analysis. The statistical results show that the reduction of IL-6 level in the trial group was more significant than that in the control group (SMD=-2.56, 95%CI [-3.34, -1.70], p = 0.00). The forest plot of IL-6 is presented in Fig. 5 (C).
CRP. Eight RCTs[27, 28, 31, 33, 36, 43, 44, 48] reported the CRP. Heterogeneity analysis results (p = 0.00, I2 = 74.87%) indicated heterogeneity among 8 RCTs, so the random effect model was used for analysis. The statistical results show that the reduction of CRP level in the trial group was more significant than in the control group (SMD=-1.30, 95%CI [-1.58, -1.01], p = 0.00). The forest plot of total CRP is presented in Fig. 5 (D).
Adverse reaction. Six RCTs[32, 34, 40, 46, 47, 50] reported adverse reactions. Heterogeneity analysis results (p = 0.47, I2 = 0.00%) indicated heterogeneity among 8 RCTs. The statistical result showed no significant difference in the incidence of adverse reactions between the two groups (RR = 0.59, 95%CI [0.26, 1.33], p = 0.21), so a descriptive discussion was taken. Two RCTs[34, 40] showed that there were no severe adverse reactions in the trial group and the control group; four RCTs[32, 46, 47, 50] reported specific adverse reaction symptoms but returned to normal without special treatment. The details of adverse reactions are shown in Table 2.
Meta-regression and subgroup analysis
For outcomes with high heterogeneity, meta-regression and subgroup analysis were used to explore the source of heterogeneity. In the meta-regression, the studies containing each outcome were divided into four categories, including the year of publication, the age of patients, the course of treatment and the producer of XNJ. Meta-regression found the year of publication (p = 0.007) may be the heterogeneity source of Barthel index. For NIHSS score, TNF-α, IL-6 and CRP, Meta-regression found no interactions between the four categories. The complete results of meta-regression are shown in S1 Table. Further, subgroup analysis was used to explore sources of heterogeneity. Subgroup analysis was made by type of the year of publication, the age of patients, the course of treatment and the producer of XNJ. Through subgroup analysis, the studies published between 2016 and 2018 (p = 0.000, I2 = 97.48%) and studies with patients aged 60 to 62 (p = 0.000, I2 = 97.86%) may be the heterogeneity sources of NIHSS score; patients age under 60 years old (p = 0.001, I2 = 82.99%), course of treatment less than 14 days and 14 days (p = 0.003, I2 = 72.26%) and XNJ of Dali pharmaceutical (p = 0.000, I2 = 93.19%) brought heterogeneity to Barthel index; patients age under 60 years old (p = 0.001, I2 = 82.99%) and between 60 and 62 years old (p = 0.000, I2 = 98.45%) may be the heterogeneity sources of IL-6; patients age between 60 and 62 years old (p = 0.000, I2 = 86.06%), and course of treatment less than 14 days and 14 days (p = 0.000, I2 = 81.27%) brought heterogeneity to CRP. The complete results of the subgroup analysis are shown in S2 Table.
Sensitivity analysis
Sensitivity analysis was used to assess the robustness of synthesized results.; the results are shown in Fig. 6. The sensitivity analysis of the overall response rate found that after excluding each study one by one, the study of Wang[38] was not within the 95%CI of the overall effect, suggesting that it may bring heterogeneity. The sensitivity analysis of NIHSS score, Barthel index, TNF-α, IL-6, and CRP showed that the combined impact was within the 95%CI of the overall impact after excluding each study, suggesting that the meta-analysis results were stable.
Publication bias
More than ten RCTs reported overall response rate and NIHSS score. Funnel plots were preliminarily judged that there was a publication bias in the overall response rate. Furthermore, Egger’s test (p = 0.0000) and Begg’s test (p = 0.0005) further provided significant evidence. Therefore, the trim and fill method explored publication bias's impact. The result(RR = 1.112, 95%CI [1.079, 1.145]) is consistent with the results before the trim and fill method(RR = 1.15, 95%CI [1.12, 1.19]). The value was changed a little using the trim and fill method. The publication bias was considered to have no directional effect on the result. The funnel plot of the NIHSS score seemed symmetrical. It showed no publication bias in NIHSS score; Egger’s test (p = 0.862) and Begg’s test (p = 0.112) further provide significant evidence. The funnel plots are shown in Fig. 7.