The systematic review was performed in accordance with the Cochrane Handbook for Systematic Reviews of Interventions (Higgins, 2011).

We enrolled randomized controlled trials (RCTs) that were published in formal English journals.

Patients diagnosed as primary diagnosis of major depressive disorder or bipolar I or II disorder were involved in this review. Interventions should be Vagus nerve stimulation (VNS) or VNS combined with treatment as usual. Comparators should be usual treatment as or sham VNS.

JGC and DYW systematically searched the PUBMED, EMBASE, Cochrane Library, Web of science, Chinese National Knowledge Infrastructure (CNKI) and Wan-fang databases from inception to 6th Sep, 2018 with MeSH terms and key words without language restrictions. The search terms were (“depressive disorder” OR depression OR melancholia OR dysthymic OR bipolar) AND (vagus OR vagus nerve stimulation) AND (randomized controlled trial OR controlled clinical trial OR randomized OR clinical trials). We also checked the reference lists of relevant reviews and included trials to identify further studies that meet the inclusion criteria for this systematic review.

Two reviewers (JGC and DYW) screened all the literature and extracted data independently using a standardized form. This was pre-designed for collecting information on trial characteristics such as first author, language, number of patients, mean age of the patients, diagnostic criteria, grades of hypertension, acupuncture treatment, control types, sessions of treatment, treatment course and outcome measures. Disagreements were resolved in consultation with the third reviewer (HL).

Suicide is the most serious consequence of depression. Therefore, suicide rate was considered as the primary outcome in this review.

Twenty-four item Hamilton Rating Scale for Depression (HAMD₂₄), Beck Depression Inventory self-rating score (BID), Montgomery-Asberg Depression Rating Scale (MADRS), Clinical Global Improvement ratings (CGI-I), 30 item Inventory of Depressive Symptomalogy-Self-Report (IDS-SR₃₀), response rate based on different scale score and adverse effect were listed as the secondary outcomes.

Two reviewers (YHZ and HC) assessed the risk of bias using the Cochrane Collaboration's tool for assessing risk of bias (Zeng et al., 2015). Each trial was scored as high, low or unclear risk for the following 7 domains: (WHO, 2018) random sequence generation (selection bias); (Fava and Davidson, 1996) allocation concealment (selection bias); (Rush et al., 2006) blinding of participants and personnel (performance bias); (Sackeim, 2001) blinding of outcome assessment (detection bias); (Kupfer et al., 2012) incomplete outcome data (attrition bias); (Schachter, 2002) selective reporting (reporting bias); (Berry et al., 2013) other bias. Disagreements were resolved in consultation with the third reviewer (HL).

Continuous data (such as score assessments) was presented as mean differences (MDs) with 95% confidence interval (CI), whereas, dichotomous data (such as response rate) was presented as relative risk (RR) with 95% CI. Statistical heterogeneity across trials was assessed by the Cochran Q-test (P < 0.1 for statistical significance) and quantified by the I² statistic. According to the Cochrane Handbook for Systematic Reviews of Interventions (Version 5.10), I² > 50% was defined as significant heterogeneity. Heterogeneous data was pooled using the random-effects model. Publication bias was evaluated by visually inspecting a funnel plot if more than 9 studies was included. Meta-analysis was performed in the case of more than three homogeneity studies by RevMan 5.3 software.

Figure 1 shows the flow chart of the study selection process based on PRISMA guidelines (Figure 1). One thousand five hundred and five records were identified during the initial search. After removing the duplicate records and screening the full text, 2 trials were included in this review (Rush et al., 2008; Hein et al., 2013).

These two trials were published in 2005 and 2013, respectively, both of which were randomized sham controlled add-on studies with two arms. Two hundred and fifty-five patients (134 in VNS group and 121 in control group) were included in the final analysis. One trial used a non-invasive method of stimulating the vagus nerve on the outer canal of the ear (Hein et al., 2013), while the other one used the implantation device (Rush et al., 2008).

The suicidal rate was not evaluated in these two trials. HRSD₂₄ was recoded in both two trials. It was presented as the sore in one trial. In the other trial, HRSD₂₄ response rate was presented. which was defined as ≥50%reductionafter 10 weeks treatment at the baseline (Rush et al., 2008).

Beck Depression Inventory self-rating score was considered as a secondary outcome in Hein's study (Hein et al., 2013), while Rush et al. took MADRS, CGI-I, and IDS-SR₃₀ as the secondary outcomes (Rush et al., 2008). Adverse effects were recorded in both two trials. More detail information was presented in Table 1.

As shown in Figure 2, the two trials were all double blind design. Rush et al. described the random procedure in detail and took Last Observation Carried Forward (LOCF) for the final statistical analysis (Rush et al., 2008), while Hein et al. did not (Hein et al., 2013). Selective reporting was unclear in both two studies for all the included studies as we have no access to the study protocol.

We did not pool the results of the two trials because of the potential heterogeneity between the two studies.

There was no significant difference between VNS and sham VNS on the score change of HAMD₂₄(MD: −2.40, 95% CI: −7.90 to 3.10) (Figures 3, 4). Similar findings were also reported on improvement percentage of HAMD₂₄ (MD:1.00, 95%CI: −6.06 to 8.06), MADRS(MD: 4.70, 95%CI: −2.98 to 12.38) and IDS–SR₃₀(MD: 4.9, 95%CI: −1.89 to 11.69). Significant difference of Beck Depression Inventory self-rating score was detected between the real and sham treatment (MD: 7.80, 95%CI: 0.34 to 15.26).

Only one trial focused on the response rate based on different scale score (Rush et al., 2008). Most of the response rate showed no significant difference between real VNS and sham treatment except for IDS-SR₃₀, which showed a better effect in real VNS group (RR: 2.33, 95% CI: 1.07 to 5.10)_.

Hein et al. reported that no adverse side effects were observed during the whole course of treatment and after the trial (Hein et al., 2013). Rush et al. reported that voice alteration was the most common adverse effect, which was occurred in 16 cases treated with VNS and 14 cased with sham treatment (Rush et al., 2008).

We did not perform publication bias evaluation as only two trials were included in this review. However, we inferred that publication bias would be exist as the small sample size and the unstable findings of the included studies.

Depression is a common mental disorder worldwide, which could lead to suicide (WHO, 2018). Thus, suicide rate may be the most important final endpoint for the effect evaluation of VNS in treating depression. Based on this clinical setting, suicidal rate was considered as the primary outcome. Unfortunately, only two RCTs were identified and suicide rate was not reported in both trials. We had to focus on the secondary outcomes such as HAMD24 sore, improvement percentage of HAMD₂₄, MADRS, IDS–SR₃₀ and response rate.

According to the findings, there may be no significant difference of main outcomes of depression between VNS and sham treatment. As lack of the primary outcome and the small sample size of the included studies, this review did not yield a stable and definitive evidence of efficacy of VNS in treating depression.

To identify the exact efficacy of VNS for depression, several systematic reviews (Daban et al., 2008; Martin and Martín-Sánchez, 2012; Berry et al., 2013; McGirr and Berlim, 2018) have been published on the same topic and one of them is an overview of review (McGirr and Berlim, 2018). Compared with the current review, these reviews only considered the Scale scores as the primary outcomes. In addition, some of the reviews involved RCTs and Non-RCTs which may lead to methodological heterogeneity in the analysis. Furthermore, most of the reviews were outdated and should be update (Daban et al., 2008; Martin and Martín-Sánchez, 2012; Berry et al., 2013). All of these may downgrade the quality and applicability of the reviews (Table 2).

Lack of sufficient primary trials and long-term follow-up data may narrow our findings and indicate high variability.

Although VNS is a novel approach for treating depression and has been approved by FDA, it still lacks sufficient evidence to confirm its efficacy and safety. Moreover, the potential mechanism and the cost-effectiveness is still unknown. Clinicians should take VNS as the treatment for depression according to the clinical settings.

VNS seems to be a new approach for depression treatment. However, its efficacy and safety needs to be further investigated. Suicidal rate is necessary to be evaluated in RCTs.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.