PubMed, EMBASE, Cochrane Central Register of Controlled Trials (CENTRAL), clinicaltrials.gov, Chinese Biomedical Literature database (CBM), China Academic Journals (CNKI), Chinese Science and Technology Journals (CQVIP), and Wanfang Database were searched systematically for all articles published from inception to July 19, 2020. The following search terms were used: ([gastr* OR stomach* OR digest* OR epigastr*] AND [carcin* OR cancer* OR neoplas* OR tumour* OR tumor* OR growth* OR adenocarcin* OR malig*]) AND (Astragalus OR radix astragali OR huang qi OR huangqi). All retrievals were implemented using MeSH and free words (the detailed search strategy is available in Supplementary 1 ). The languages were restricted to Chinese and English.

Inclusion criteria: (1) All studies were randomized controlled trials (RCTs) or quasi-RCTs. (2) Patients who had TNM stage III-IV AGC and were diagnosed using the histopathological and cytological diagnostic criteria. (3) The experimental group of patients was given Astragalus-based herbal therapy combined with PBC. Any form of Astragalus (Huang qi) preparation, including water decoction, extracts, granules, or injection, among other forms, regardless of administration route, were included. The control group of patients was given PBC alone. (4) Outcomes were identified as the tumor response, survival rate, QOL, ADRs, and peripheral blood lymphocytes levels, and at least one of these outcomes was reported; and (5) for repeated publication studies, we selected the data with the most comprehensive report and the longest follow-up.

Exclusion Criteria: (1) Patients who underwent radiotherapy, chemotherapy, or other antitumor therapy within 1 month before treatment; (2) patients with severe infection, other malignant tumors, and severe medical diseases; (3) the prescription of Astragalus-based herbal therapy was not fixed; (4) studies for which the data could not be extracted; and (5) the baseline data of patients in two groups were not comparable.

Studies were selected independently by two reviewers according to the above inclusion and exclusion criteria. Two reviewers independently extracted the data. Disagreements were discussed with and resolved by the third reviewer. The following data were extracted: first author (year of publication), sample sizes, gender, age, study arm, drug delivery, treatment duration, follow-up time, outcomes, and criteria. Data presented graphically were extracted using WebPlotDigitizer (https://automeris.io/WebPlotDigitizer).

Two reviewers independently assessed the quality of the selected studies according to the Cochrane Collaboration’s tool for RCTs. Items were sorted into three categories: low risk of bias, unclear bias, and high risk of bias. The following characteristics were evaluated: Random sequence generation (selection bias), allocation concealment (selection bias), blinding of participants and personnel (performance bias), incomplete outcome data (attrition bias), selective reporting (reporting bias), and other biases. Results from these evaluations were graphed and assessed using Review Manager 5.3.

Tumor response and the survival rate were the main outcomes. Tumor response, containing the objective response rate (ORR) and disease control rate (DCR), was based on the World Health Organization (WHO) criteria (12) or Response Evaluation Criteria in Solid Tumors (RECIST) (13). Complete response (CR), partial response (PR), stable disease (SD), and progressive disease (PD) were used as indicators. The CR plus PR rates were as equal to the ORR; the CR plus PR, and SD rates were equal to the DCR.

Quality of life (QOL), adverse drug reactions (ADRs), and peripheral blood lymphocytes levels were the second outcomes. QOL was considered improved when the Karnofsky Performance Status (KPS) score was 10 points higher after treatment than before treatment. ADRs were accessed by measuring hematotoxicity (neutropenia, anemia, and thrombocytopenia), gastrointestinal toxicity (nausea and vomiting, diarrhea), hepatic and renal dysfunction, neurotoxicity, alopecia, and stomatitis, based on WHO criteria (12) or NCI Common Terminology Criteria for Adverse Events (CTCAE) (14). The peripheral blood lymphocyte levels were evaluated by measuring the T-lymphocyte subsets such as the percentage of CD3⁺, CD3⁺CD4⁺, and CD3⁺CD8⁺ T cells, as well as the CD4⁺/CD8⁺ T cells ratio and the percentage of natural killer (NK) cells.

The Review Manager 5.3 and Stata V16.0 software were used in this study. The risk ratio (RR) and 95% confidence interval (CI) were used to assess dichotomous variables, while mean difference (MD) with a 95% CI was used to assess continuous variables. Statistical heterogeneity was measured by the I² statistic and the Chi-squared test. The summary RR, MD, and 95% CI were estimated by a random-effect model. P values were used to calculate outcomes, and P < 0.05 was considered statistically significant. The funnel plots and Egger’s tests were applied to examine potential publication bias when studies ≥10. In accordance with the drug delivery of Astragalus-containing TCM, PBC regimen, and treatment duration, subgroup analyses were performed to reveal the clinical heterogeneity and its influence on tumor response.

Trial sequential analysis software (TSA, version 0.9.5.10 beta) was used to determine the robustness of main outcomes and to calculate the required information size (RIS) in the meta-analysis (15). If the cumulative Z-curve stretched across the monitoring boundaries, a sufficient level of evidence for the intervention effect may have been reached and no further studies were needed. Type I and II errors were needed to perform the TSA. In this review, the RIS was estimated using α = 0.05 (two sided) with power equal to 80%.

Two reviewers independently assessed the quality of the evidence for each outcome using the GRADE approach (16). Disagreements were discussed with and resolved by the third reviewer. The quality of evidence was classified as being high, moderate, low, and very low. The quality was downgraded according to the following five domains: (1) risk of bias (if most trials showed an unclear risk, with or without high risk, but the results had good robustness, the evidence was rated down by only one level; if the results showed poor robustness, the evidence was downgraded by two levels); (2) inconsistency (statistical heterogeneity was present and the results of the sensitivity analysis had poor robustness); (3) indirectness (the participants, intervention, outcomes, or comparison of the study did not meet the objectives of this study); (4) imprecision (the sample size for each outcome was fewer than 300 cases); and (5) reporting bias (publication bias). Except for the risk of bias, evidence was downgraded by one level.

A total of 2,873 records were identified through database searching. All records were screened by two reviewers with a three-step process. Firstly, we screened the titles and removed duplicates, after which 1,829 records remained. Secondly, we included 227 full texts by reading the title or abstract. Thirdly, we evaluated the full texts and excluded 192 trials by following the inclusion and exclusion criteria. In the end, 35 eligible studies involving 2,670 stage III/IV AGC patients were included in our study ( Figure 1 ).

The characteristic information of the included studies is shown in Table 1 . All 35 studies were single-centered and conducted in China. There were 1,572 male and 1,098 female patients, and sample sizes ranged from 30 to 240. Eighteen studies used the oral TCM administration, and 17 studies used commercial TCM injections. All clinical trials were designed to use Astragalus-containing TCM as the principal drug together with PBC. Follow-up lasted from 4 weeks to 3 years. Twenty trials reported the tumor responses according to the WHO guidelines, and 12 reported them according to RECIST guidelines. Eighteen trials reported the ADRs using WHO criteria, three using NCI CTCAE criteria, and eight using unknown criteria.

All 35 studies referred to randomization; however, only eight studies (23, 25, 29, 30, 43, 47, 50, 52) used a random number table to generate a random sequence, one study used a lottery approach (24), and two studies (35, 49) used admission order, which was inappropriate. One study used an envelope method (44) to perform the allocation concealment, while the other studies did not report the method. Only one study mentioned single blinding (25), and none of the studies adequately reported the blinding of the investigators, patients, and outcome assessors. Selective reporting existed in one study (40) that failed to completely report the DCR. One study (20) had a high risk of other bias because it was funded by the industry. The risk of bias for the individual trials is summarized in Figure 2 .

Thirty-two trials with 2,454 patients reported the ORR following WHO or RECIST guidelines. The random-effects meta-analysis showed that Astragalus-based herbal therapy plus PBC enhanced the ORR, which showed a statistically significant difference (RR: 1.24, 95% CI: 1.15–1.34, P < 0.00001, I² = 0%, Figure 3 ).

Thirty-one trials with 2,358 cases reported the DCR. The meta-analysis using a random-effects model showed that Astragalus-based TCM plus PBC increased the DCR; the difference was statistically significant (RR: 1.10, 95% CI: 1.06–1.14, P < 0.00001, I² = 11%, Figure 4 ).

Six trials (19, 26, 27, 29, 44, 52) reported the survival rate ( Figure 5 ). Additionally, in two trials (47, 48), the survival rate was extracted from survival curves. Thus, eight trials containing 512 cases were included. The results demonstrated that there was no significant difference in the half-year survival rate between the two groups (RR: 1.14, 95% CI: 0.89–1.45, P = 0.31, I² = 90%). However, compared with the PBC-treated control group, the 1- and 2-year survival rates in the Astragalus-based TCM group were significantly improved (RR: 1.41, 95% CI: 1.09–1.82, P = 0.005, I² = 65%; RR: 3.13, 95% CI: 1.80–5.46, P = 0.84, I² = 0%).

In the included studies, two types of data were applied to report the QOL changes on KPS score. First, the number of patients who reported QOL improvement (KPS score 10 points higher after treatment). And second, the mean ± standard deviation (SD) of the KPS score before and after treatment. As shown in Figure 6 , the number of patients with improved QOL based on the KPS score were reported in 14 studies (RR: 2.03, 95% CI: 1.70–2.43, P < 0.00001, I² = 0%); and the mean ± SD of KPS score were reported in other six studies (MD: 12.39, 95% CI: 5.48–19.30, P = 0.0004, I² = 95%) ( Figure 7 ). Taken together, compared with PBC, the study results showed that Astragalus-containing TCM plus PBC significantly improved QOL.

Twenty-nine trials that included 2,053 patients reported ADRs principally according to the WHO criteria (12) ( Table 2 and Figures S1–10 ). The results of the random-effects meta-analysis demonstrated that, compared with PBC alone, Astragalus-containing TCM plus PBC significantly reduced the risk of neutropenia (RR: 0.65, 95% CI: 0.54–0.79, P < 0.00001), anemia (RR: 0.68, 95% CI: 0.52–0.89, P = 0.005), thrombocytopenia (RR: 0.65, 95% CI: 0.51–0.82, P = 0.0004), nausea and vomiting (RR: 0.73, 95% CI: 0.64–0.82, P < 0.00001), diarrhea (RR: 0.58, 95% CI: 0.45–0.75, P < 0.0001), hepatic dysfunction (RR: 0.55, 95% CI: 0.40–0.77, P = 0.0005), renal dysfunction (RR: 0.48, 95% CI: 0.26–0.88, P = 0.02), neurotoxicity (RR: 0.78, 95% CI: 0.65–0.92, P = 0.004), and alopecia (RR: 0.77, 95% CI: 0.61–0.97, P = 0.03), but not the stomatitis rate (RR: 0.73, 95% CI: 0.53–1.00, P = 0.05).

Ten trials that included 796 patients reported the peripheral blood lymphocyte levels. The results of CD3⁺ CD8⁺ T cell levels displayed poor clinical consistency. Therefore, we only evaluated the MD of the CD3⁺ T cells, CD3⁺ CD4⁺ T cells, the CD4⁺/CD8⁺ T cells ratio, and NK cells using a random-effects model. Our results indicated that Astragalus-containing TCM plus PBC significantly enhanced the ratio of CD3⁺ T cells (MD: 11.51, 95% CI: 5.94–17.08, P < 0.0001), CD3⁺ CD4⁺ T cells (MD: 6.44, 95% CI: 4.25–8.62, P < 0.00001), NK cells (MD: 4.58, 95% CI: 2.41–6.76, P < 0.0001) ( Figure 8 ), and the CD4⁺/CD8⁺ T cells ratio (MD: 0.41, 95% CI: 0.27–0.55, P < 0.00001) ( Figure 9 ) after treatment.

We performed subgroup analyses in accordance with the drug delivery of Astragalus-containing TCM, PBC regimen, and treatment duration to reveal the influence of the addition of Astragalus-containing TCM on the ORR and DCR ( Figures S11–16 ). The drug delivery of Astragalus-containing TCM was either orally or by injection. Subgroup analysis showed that Astragalus-containing TCM could improve the ORR and DCR regardless of whether it was administered orally or by injection, respectively ( Table 3 ). PBC regimens were cisplatin-based chemotherapy and oxaliplatin-based chemotherapy regimens. Subgroup analysis displayed that Astragalus-containing TCM plus oxaliplatin-based chemotherapy could enhance both the ORR and DCR; however, when combined with cisplatin-based chemotherapy, it could only enhance the ORR ( Table 3 ). Finally, the treatment durations were ≥8 weeks and <8 weeks. As shown in Table 3 , whether the treatment duration was ≥8 weeks or <8 weeks, the combination of Astragalus-containing TCM and PBC could improve the ORR and DCR.

We performed sensitivity analysis by omitting each study in turn to check the robustness of the main outcome, including ORR, DCR, and survival rate. This analysis showed that the pooled RR values of the ORR, DCR, and 1-year survival rate were stable, except for the half-year survival rate ( Table 4 ). Considering that the half-year survival rate was extracted from survival curves in the study of Pan, B.Y.2018 (48), we excluded this study, and the I² decreased from 90 to 0%.

As shown in Figures S17, 18 , the TSA for the ORR and DCR showed that the Z-curve (blue dashed line) crossed the conventional boundary (brown dotted line), the trial sequential monitoring boundary for benefit (lower red solid line), and the RIS of the TSA (vertical red line; 660 for ORR, 574 for DCR), which suggested that the findings of the meta-analysis were robust for the ORR and DCR.

The TSA for the half-year survival rate ( Figure S19 ) showed that the Z-curve did not cross either the conventional or trial sequential monitoring boundary, as well as the RIS (equal to 2920), which indicated that the evidence on the effect of half-year survival rate was insufficient. TSA for 1-year survival rate ( Figure S20 ) showed that the Z-curve crossed the conventional and trial sequential monitoring boundary for benefit, but not the RIS (equal to 611). This indicates that the combination of Astragalus-containing TCM and PBC might enhance the 1-year survival rate for AGC patients. However, we did not have enough power to confirm the conclusion before RIS of 611 participants. The TSA for the 2-year survival rate ( Figure S21 ) showed that the Z-curve crossed the conventional boundary and RIS (equal to 95), which indicated that the evidence on the effect of the 2-year survival rate was robust. Due to insufficient data, we failed to evaluate the effect on the 3-year survival rate.

The funnel plots ( Figures S22–33 ) were not strictly symmetrical in the meta-analysis of the ORR, DCR, QOL, neutropenia, anemia, thrombocytopenia, nausea and vomiting, diarrhea, hepatic dysfunction, renal dysfunction, neurotoxicity, and CD4⁺/CD8⁺ T cells ratio. However, as shown in Egger’s test ( Table 5 ), no significant publication bias existed among the meta-analysis of ORR (P = 0.3986), QOL (P = 0.7377), diarrhea (P = 0.3092), hepatic dysfunction (P = 0.5113), renal dysfunction (P = 0.9436), and CD4⁺/CD8⁺ T cells ratio (P=0.5250).

In summary, the quality was moderate for the ORR, 1-year survival rate, QOL, hepatic dysfunction, renal dysfunction, stomatitis, and levels of peripheral blood lymphocytes; low for DCR, neutropenia, anemia, thrombocytopenia, nausea and vomiting, diarrhea, and neurotoxicity; and very low for other results ( Table 6 ).