Efficacy of non-invasive brain stimulation combined with antidepressant medications for depression: a systematic review and meta-analysis of randomized controlled trials

Two of the authors independently electronically searched PubMed, Embase, Web of Science, the Cochrane Database of Randomized Controlled Trials and SinoMed using the following words and phrases: (1)"Antidepressive Agents"[Mesh] OR "Antidepressive Agents, Second-Generation"[Mesh] OR "Antidepressive Agents, Tricyclic"[Mesh] OR (Antidepressive Agents OR "antidepress*" OR Selective serotonin reuptake inhibitors OR SSRIs OR Tricyclic antidepressant OR Serotonin and noradrenaline reuptake inhibitors OR Noradrenergic and specific serotonergic antidepressant OR Norepinephrine and dopamine reuptake inhibitors OR Monoamine oxidase inhibitors OR Vortioxetine OR Vilazodone OR Agomelatine OR Serotonin OR amitriptyline OR bupropion OR citalopram OR desvenlafaxine OR duloxetine OR escitalopram OR fluoxetine OR fluvoxamine OR levomilnacipran OR milnacipran OR mirtazapine OR nefazodone OR paroxetine OR reboxetine OR sertraline OR venlafaxine OR vilazodone OR vortioxetine)[Title/Abstract]; (2) "Transcranial Magnetic Stimulation"[Mesh] OR "Transcranial Direct Current Stimulation"[Mesh] OR (noninvasive Brain Stimulation OR NIBS OR Transcranial Magnetic Stimulation OR Transcranial Magnetic Stimulations OR Repetitive Transcranial Electrical Stimulation OR rTMS OR Cathodal Stimulation OR Transcranial Direct Current Stimulation OR Cathodal Stimulation tDCS OR Cathodal Stimulation tDCSs OR Transcranial Random Noise Stimulation OR Transcranial Alternating Current Stimulation OR Transcranial Electrical Stimulation OR Anodal Stimulation tDCS OR Anodal Stimulation tDCSs)[Title/Abstract];(3) "Depression"[Mesh] OR "Depressive Disorder"[Mesh] OR (Depress* OR "dysthymi* OR mood disorder* OR affective disorder* [Title/Abstract]). The ultimate search method was "(1) AND (2) AND (3)". Further complementary access to the relevant literature can be gained by reading the references incorporated into the literature. The deadline for the search was October 6, 2023. In addition, ClinicalTrials.gov (https://​www.​clinicaltrials.​gov/​) and Google Scholar (www.​scholar. google.com.cn) were used as supplementary search. The specific search strategy is in (Additional file 1. Search strategies).

The population, intervention, comparison, outcome, and study designs (PICOS) framework [24] was the basis for the selection criteria. Studies meeting the following criteria were included in the meta-analysis:

Participants: adult individuals aged over 18 years who have been diagnosed with depression. The diagnosis of depression met DSM-IV, DSM-5, ICD-10 diagnostic criteria, or the depression disorder prevention guide.

Interventions: one of the NIBS techniques (rTMS or tDCS) was used in the interventions combined with antidepressant medications (the type and dose of medications were not restricted).

Comparison: the control group that received only medication did not receive the NIBS technique intervention.

Outcomes: the primary outcome was the depression scale score measured by the Hamilton Depression Rating Scale (HDRS), the Montgomery-Asberg Depression Rating Scale (MADRS), or the Beck Depression Inventory Rating Scale (BDI). The anxiety scale score measured was by State­Trait Anxiety Inventory (STAI). The secondary outcomes were clinical response rates, remission rates, drop-out rates, and changes in certain levels of neurotransmitters after intervention [i.e., dopamine (DA), gamma-aminobutyric acid (GABA), and 5-hydroxytryptamine (5-HT)]. The response rate was defined as a 50% or greater reduction in depression scores from baseline. The remission rate was defined by the criteria used in each trial (for example, an endpoint HDRS score ≤ 7 or MADRS score ≤ 10). If studies reported both the HDRS and MADRS scores, we analysed the scores from the scales used to define response and remission in their trials. The drop-out rate defined as the proportion of participants who prematurely discontinued their participation in the study for any cause.

Study designs: randomised controlled trials (RCTs), including parallel-group RCTs and crossover RCTs. We also considered quasi-randomised controlled trials (quasi-RCTs), in which the allocation was systematic but not random (e.g., based on hospitalisation number).

The exclusion criteria were as follows: (1) patients with other disorders (such as schizophreni, obsessive­compulsive disorder, substance use disorders, etc.); (2) non-simple depression patients (such as postpartum depression, bipolar disorder, geriatric depression, secondary depression, and vascular depression); (3) conference articles and case reports; (4) duplicate articles or duplicative datasets from the same trial; (5) articles lacking any of the primary outcomes; (6) articles not in Chinese or English.

All of the search results were imported into the Zotero software, and duplicates were removed. Screening, eligibility determination, and inclusion in this systematic review followed the same procedure. Two reviewers each individually evaluated one article, and a third author resolved any differences. A data extraction form was prepared in accordance with the Cochrane Handbook for Systematic Reviews of Interventions. The same method previously mentioned was used to collect data. The main contents were extracted as follows: (1) first author's name and year of article publication; (2) clinical characteristics of the included studies (age, sample size, and types of depression); (3) treatment/control group information, including forms, doses, duration of antidepressant use, and treatment stimulation parameters of NIBS; (4) primary outcome; and (5) secondary outcome. All data are expressed as mean and standard deviation (SD). As this meta-analysis compared the values of the data change between the experimental and control groups before and after the intervention, the collected data had to be converted. The difference between the pre- and post-intervention assessed values was the change in value. If the change was negative, the estimated value after the intervention was lower than that before the intervention; otherwise, a positive value indicated an increase. The following formulas were used to determine the mean value of change: \(\overline{{X }_{c}}=\overline{{X }_{a}}-\overline{{X }_{b}}\); and SD value change: \({S}_{c}=\sqrt{{S}_{a}^{2}+{S}_{b}^{2}-2\times corr\times {S}_{a}\times {S}_{b}}\), (corr = 0.50) [25]. If available, the intention-to-treat (ITT) or modified intention-to-treat (mITT) data were preferred to over data based only on completer.

We used the Risk of Bias Assessment Tool (ROB 2.0) [26] of the Cochrane Reviewers' Handbook 6.1 to assess the quality of the included studies. The five domains of ROB 2.0 are as follows: 1) the bias that is caused by the randomization method; 2) the bias that is caused by deviations from the interventions that were anticipated; 3) the bias that is caused by the absence of outcome data; 4) the bias that is generated by an evaluation of the outcome; and 5) the bias that is derived from the selection of the results that were presented. The risk of bias for each module was discussed and agreed upon by two researchers who composed each module. If a consensus judgement could not be reached, experts in evidence-based medicine, epidemiology, or health statistics were asked to assess it, and this conclusion was used to evaluate the total risk of bias in the article.

Odds ratios (OR) were used for dichotomous variables, and mean differences (MD) or standardised mean differences (SMD) were used for continuous variables. For each outcome, 95% confidence intervals (CI) were used as effect statistics. I² was used to measure the degree of study outcome heterogeneity. We interpreted I² following the Cochrane Handbook guidelines, considering the limitations of specific thresholds [27]: 0% to 40% may not be important; 30% to 60% may indicate moderate heterogeneity; 50% to 90% may indicate substantial heterogeneity; 75% to 100% is considered considerable heterogeneity. If there was significant statistical heterogeneity among the stuides, the source of heterogeneity was further analyzed, and a random-effects model was used for meta-analysis after excluding the effect of significant clinical heterogeneity. Significant clinical heterogeneity was addressed using methods such as subgroup, sensitivity, or only descriptive analyses. Test level ɑ = 0.05.

Analyses were conducted using the 'meta' and 'metafor' packages [28] for R software (version R × 64 4.2.2). All algorithms and scripts utilized within the software for conducting the statistical analysis are available in Additional file 3: algorithms and scripts. A funnel plot was used to assess publication bias, with asymmetry signifying potential publication bias, and the results were confirmed by Egger's test when bias was possible. We further visualised contour-enhanced funnel plots to assess whether the potential funnel asymmetry was likely to be due to a statistically significant publication bias. Sensitivity analysis was performed to test the robustness of the main and secondary outcomes.

The PRISMA flow chart of the study selection is shown in Fig. 1. Based on the retrieval strategy, 7602 articles were obtained from the database, and five articles were supplemented by tracing references. Duplicates of 2701 articles were removed using Zotero software, and 4819 articles were excluded after screening titles and abstracts. Based on the eligibility criteria, 87 articles were potentially relevant to our systematic review. After the full-text evaluation, 69 articles were excluded (Additional file 2.list of the excluded full-text).

Overall, 18 RCTs with 1357 patients were included. The included studies were published between 2005 and 2023. The characteristics of the studies are presented in Table 1, and the characteristics of the NIBS specific parameters are presented in Table 2. These studies were conducted in Germany (n = 4); Brazil (n = 2); France (n = 1); China (n = 7); Russia (n = 1); Turkey (n = 1); India (n = 1); and Italy (n = 1). Seven studies used tDCS + medication, and eleven studies used rTMS + medication. The number of NIBS sessions varied, whereas the duration of the therapy ranged from a single application to 8 weeks. The most common treatment period was two weeks. The parameters of the neural stimulation settings varied considerably between studies. The frequencies employed ranged from 5 to 20 Hz. The intensity can be expressed as a Tesla (MT; 80–120%) motor threshold.

Of the studies, only eight [15, 16, 21, 29, 32‐34, 37] provided clear descriptions of the methods employed for randomization and allocation concealment. Eleven studies [15, 16, 29, 30, 33, 34, 37, 39‐42] reported loss of follow-up, with the rate of incomplete data ranging from 3.0% to 35%. The primary reasons for losses to follow-up typically included patient mortality, intolerable pain, and refusal to persist with the intervention. Seven studies [15, 16, 29, 32‐34, 37] were at low risk of reporting bias, and the study protocol of these studies could be retrieved. A summary of this is shown in Fig. 2.

The meta-regression analysis indicated no significant association between the clinical parameters (such as the type of NIBS, the severity of depression, and the antidepressant class) and demographic factors (including the sample size, age, and percentage of females) with rates of remission or dropout. The details are shown in Additional file 6: Meta-regression analyses. A stepwise regression analysis was conducted to explore the relationship between the HAMD score and several independent variables (Table 3), revealing that the impact of the depression score was influenced by two factors: the use of tricyclic antidepressant medications (TCAs) and the sex of the participants (p = 0.01 and p = 0.03, respectively). Sample size demonstrated an influence on the response rate (p = 0.01). For other outcomes, the limited availability of studies precluded the application of meta-regression models.

The primary limitations of the current study are as follows: First, we only examined mean treatment effects and were unable to investigate potentially crucial clinical and demographic variables of response to therapy at the individual level (e.g., age, sex, degree of severity of symptoms, or the period of illness). In randomised trials, patients are typically rigorously screened, and patients with bipolar disorder and other comorbidities are excluded. Psychological disorders are often highly comorbid, which may limit the applicability of the findings to these clinical subgroups; however, this was a methodological advantage to ensure the study's transitivity. In addition, our studies were highly heterogeneous, perhaps because the NIBS parameters and antidepressant medications used varied widely across the studies; however, the limited number of eligible studies prevented us from assessing how these potential factors affected heterogeneity. Moreover, five studies included in the analysis displayed an uncertain risk of bias, and in six of them, the overall quality of the included studies was not high.

Notwithstanding these limitations, the findings of this meta-analysis represent the most comprehensive evidence base currently available that may guide clinical guidelines and aid in a shared decision-making process involving patients, caregivers, and physicians when selecting the most appropriate treatment for adult patients with depressive disorder in their daily practice.

Future research should strive to expand the scope of meta-analyses by including both aggregate and individual patient data from clinical trials, thereby doing what is commonly referred to as an individual-patient data meta-analysis. Additional high-quality RCT trials with larger sample sizes are needed to further validate the efficacy of NIBS.