Based on the PICO framework, we selected people with depression as patients. The intervention was GKB, and it was compared with placebo or traditional antipsychotics. The outcome included the Hamilton Depression Scale (HAMD), the modified Barthel index (MBI), the modified Edinburgh-Scandinavian stroke scale (MESSS), brain-derived neurotrophic factor (BDNF), 5-hydroxytryptamine (5-HT), and adverse events.

The Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) flow diagram and checklist (Page et al., 2021), which were registered at the International Platform of Registered Systematic Review and Meta-analysis Protocols (INPLASY2023100052), served as the foundation for this study. To find all randomized controlled trials evaluating GKB for depression, we searched seven internet databases (PubMed, Embase, Scopus, Web of Sciences, Google Scholar, Cochrane Library, and China National Knowledge Infrastructure) for pertinent articles. After a thorough screening, 21 publications, released without language restriction between January 2002 and May 2023, were included.

In the Boolean phrase ((Ginkgo biloba) OR (G. biloba) OR GKB) AND ((depression) OR (depressive symptoms)), the medical topic terms and keywords utilized for the search were “Ginkgo biloba,” “G. biloba,” “GKB,” “depression,” and “Depressive Symptoms.” The references’ related articles were also investigated. Studies qualified if they included the following data: observational studies reporting clinical effectiveness, adverse events, and random control trials reporting depression assessments. Evaluation of GKB’s impact on depression was the main goal.

Two authors (Jingya Lin Xiaojing Sun) extracted the data using a standardized data form after screening the titles, abstracts, and full texts of the publications identified by our search and assessing the risk of bias. Any discrepancies were discussed with a third author (Lingli Yang).

After the initial selection of the studies, the pertinent texts were examined, and the studies needed to fulfill the following criteria:

(1) Studies that compared GKB and a control for the treatment of depression;

The following predetermined exclusion criteria were used to disqualify studies from consideration:

(1) Research on other drugs;

Two writers (Jingya Lin Xiaojing Sun) extracted the following information from each included study: the name of the first author, publication year, author’s country, groups based on intervention (GKB and control), research design, gender, average subject age, and research cycle. The outcome criteria for depression were GKB’s effectiveness and safety.

Using the ROB 2.0 scale (a revised Cochrane risk-of-bias instrument for randomized trials), two independent reviewers (Xiaojing Sun and Lingli Yang) assessed the caliber of the included studies.

We assessed heterogeneity using the Cochrane Q-statistic and Higgins and Thompson’s I². Depending on its value, I² was used to classify heterogeneity as low, moderate, or high: 25%, 50%, or 75%. When I² was less than 50%, the fixed-effect model was used for the meta-analyses. If I² was higher than 50%, the random-effect model was used in all other cases. For effect sizes of continuous outcomes, the mean difference (MD) with 95% CI (confidential interval) was precisely recalculated. Risk ratios (RRs) were computed for categorical outcomes. Our statistical analyses were conducted using R software version 4.2.1 (R Core Team, Vienna, Austria) and the R package meta (version 6.2.0). Statistical significance was defined as a p-value of 0.05 or less.

Publication bias was assessed for the Hamilton Depression Scale (HAMD) as the primary outcome using Egger’s regression test and a funnel plot. We ran a sensitivity analysis, deleting one included article at a time, to assess the robustness of the final results.

The systematic search of seven online databases identified 402 articles based on the PRISMA flow diagram and inclusion/exclusion criteria. Twenty-one publications (Lin, 2002; Fengqing Yu and Liu, 2004; Hartley et al., 2004; Kefeng Guo and Yan, 2006; Wei Chen, 2006; He, 2007; Binhua Chen and Su, 2011; Jia Yuan and Gan, 2011; Xiangdong Luo and Zhou, 2011; Xiangdong Luo and Zhou, 2012; Zhang, 2012; Gavrilova et al., 2014; Hao Song and Chen, 2014; Haixia Wu and Hao, 2016; Shichao and Chen, 2017; Dai et al., 2018; Ru Liu and Zhang, 2018; Junfeng Yin, 2019; Liang et al., 2019; Liu, 2020; Aijun Zhang and Yu, 2021) with 2074 patients were qualified for this meta-analysis after duplicate studies were excluded and eligible studies were screened [12–24]. The research selection flowchart is depicted in Figure 1, and the characteristics of the studies are listed in Table 1. The risk-of-bias evaluation for every included study is displayed in Supplementary Figures S1, S2. In addition, we list the chemical component or manufacturer of the GKB in Table 2 so we can make the evaluation of GKB clear.

Based on the ROB 2.0, it was determined that the overall quality of the included studies was sufficient in terms of selection bias, comparability quality, and outcome quality.

Of all the studies, 10 had more than 100 patients, while 11 had less than 100. The duration of the study varied from 1 to 4 years. All the research designs used in the included articles were randomized control trials (RCTs). The median ages of the patients studied in the included articles ranged from 51.95 to 69.37. There were nearly equal numbers of men and women (952 men and 1,020 women). The specific traits of the included articles are displayed in Table 1.

The Supplementary Figures S1, S2 show the ROB 2.0 summary and quality assessment details, respectively. More than 50% of the articles indicated a low risk of bias, while less than 20% exhibited a high risk. Only three articles had a high overall bias risk, six had a medium bias risk, and twelve had a low bias risk.

First, we evaluated the HAMD scale after giving GKB for 4 weeks, 6 weeks, and 8 weeks. Pooled estimates from 10 studies showed that patients receiving GKB had substantially lower 4-week HAMD scores than the control group (MD = −2.86, 95%CI [−4.27, −1.46], p < 0.01, I² = 91%) (Figure 2A). The 6-week HAMD score was documented in seven studies. The results showed that patients receiving GKB also had lower 6-week HAMD scores than the control group (MD = −3.36, 95%CI [−4.05, −2.67], p < 0.01, I² = 50%) (Figure 2B). After taking GKB for 8 weeks, the patients receiving GKB had lower HAMD scores than the control group, according to a meta-analysis of 11 publications (MD = −4.58, 95% CI [−6.11, −3.05], p < 0.01, I² = 95%) (Figure 2C).

In addition to GKB therapy, we further investigate the serum indices and other depression scales. Patients receiving GKB had a higher MBI than those in the control group (Figure 3A, MD = 14.86, 95%CI [12.07, 17.64], p < 0.01, I² = 42%). Meanwhile, patients receiving GKB had lower MESSS values than those in the control groups, according to a pooled analysis (Figure 3B, MD = −4.57, 95%CI [−6.34, −2.79], p < 0.01, I² = 77%). Serum levels of BDNF (MD = 16.35, 95%CI [7.34, 25.36], p < 0.01, I² = 82%, Figure 3D), and 5-HT revealed that patients receiving GKB had greater values than those in the control groups (MD = 4.57, 95%CI [3.08, 6.05], p < 0.01, I² = 0%, Figure 3C).

Finally, we assessed the clinical effectiveness and adverse events in the GKB and control groups. In a meta-analysis of clinical efficacy, the GKD groups outperformed the control groups in terms of clinical efficacy (RR = 1.22, 95%CI [1.11, 1.34], p < 0.01, I² = 67%) (Figure 4A), whereas there was no difference between the GKB and the control group in terms of adverse events (RR = 0.96, 95%CI [0.8, 1.14], p = 0.62, I² = 0%) (Figure 4B).

Begg funnel plots for the meta-analysis of adverse events were created to examine the publication bias. The funnel plots (Figure 5) demonstrated visual symmetry and revealed a low level of publication bias in this study. The Egger regression test for adverse events further indicated there was no publication bias (z = 0.23; p = 0.19).

Sensitivity analysis was done to verify that the results were reliable. We removed each article in turn from the sensitivity forest plot (Figure 6). Each included article in the adverse events meta-analysis exhibited a comparable RR in terms of robustness, with Lin 2002 having the greatest RR at −2.45 [−3.86, −1.03] and Chen 2011 having the lowest RR at −3.08, 95%CI [−4.52, −1.64]. These findings suggested that the conclusions were sound.