**3. Results**

We used fairly broad search terms (see Section 2) to ensure the inclusion of all studies relevant to the use of DBS as a therapy for depression. Our search identified 1397 studies, and all were screened for inclusion (Figure 1). We excluded 964 studies at the abstract/title level because these studies either did not use DBS as the therapeutic modality, examined depressive symptomatology in the context of other diseases, or were non-human animal studies. Of the remaining relevant studies, 308 were excluded because they were case reports, non-systematic reviews, perspectives, commentaries, editorials, or opinions. The remaining 125 studies were included in our qualitative review. We then screened these studies for inclusion in our quantitative meta-analysis. Twenty-six studies were candidates for inclusion at the abstract level. Thirteen studies were excluded because they did not compare active to sham stimulation [31–43], and one study was excluded as it included only three patients [44]. Therefore, 12 studies [15–25] (186 unique patients) were included in the meta-analysis and meta-regression (Table 1). The Raymaekers et al. study was analyzed as two separate studies, because this study included two anatomically distinct stimulation targets, and both targets were evaluated with blinded stimulation periods.

**Figure 1.** Flowchart of studies selected for inclusion in the qualitative review and quantitative meta-analysis.


**Table 1.** Studies included in meta-analysis and meta-regression.

\* Only half of the patients crossed over. IC/BST: internal capsule/bed nucleus of the stria terminalis; ITP: inferior thalamic peduncle; MFB: medial forebrain bundle; SCC: subcallosal cingulate; vALIC: ventral anterior limb of the internal capsule; VC/VS: ventral capsule/ventral striatum.

The studies included in the meta-analysis had varied trial designs (Table 1). Due to our inclusion criteria, all studies contained a period of blinded sham stimulation and blinded active stimulation. The duration of the active and sham stimulation periods, however, was heterogeneous. The average blinded stimulation duration was 7.5 ± 6.6 weeks. All studies contained an open-label period of long-term active stimulation following the blinded phases. These long-term data were not included in the meta-analysis, since the goal of the present study was to compare blinded active stimulation to blinded sham stimulation. The majority of the trials (75%) were done in a crossover fashion (Table 1). Thus, all of the patients in these studies received both active and sham stimulation in a blinded fashion. Importantly, these study designs allow for within-subject comparisons and may enhance statistical power.

Using a random-e ffects model, our meta-analysis revealed that active stimulation results in a greater decline in HDRS/MADRS scores relative to sham stimulation (standardized mean di fference (SMD) = −0.75; −1.13 to −0.36, 95% confidence interval (CI); *p*-value = 0.0001; Figure 2). There was moderate heterogeneity across studies (I2 = 59%).

**Figure 2.** Meta-analysis forest plot depicting changes in HDRS/MADRS scores with active stimulation compared to sham stimulation. CI: confidence interval; IC: internal capsule; ITP: inferior thalamic peduncle; SMD: standardized mean di fference; PI: prediction interval.

In addition to differences in study design, the studies also investigated the efficacy of DBS for TRD using different stimulation targets (Table 1). The most common target was the SCC (50% of studies), followed by the internal capsule (IC, 25%), MFB (17%), and ITP (8%). While there were a limited number of studies, we utilized meta-regression to determine if the available data would reveal an optimal stimulation target. The meta-regression, however, did not demonstrate a statistically significant difference (*p* = 0.45) between stimulation targets (Figure 3).


**Figure 3.** Meta-regression forest plot comparing various stimulation targets. CI: confidence interval; IC: internal capsule; ITP: inferior thalamic peduncle; MFB: medial forebrain bundle; SCC: subcallosal cingulate; SMD: standardized mean difference; TE: treatment effect; seTE: standard error of treatment effect.

Since the duration of stimulation during the blinded phase varied between studies, we performed another meta-regression to determine if there was an association between the duration of active stimulation and SMD. Our analysis did not reveal a significant effect of stimulation duration on SMD outcomes (*p* = 0.20).

Publication bias is an important concern when conducting a meta-analysis. We investigated for possible publication bias by first generating a funnel plot (Figure 4). We then tested for asymmetry of the funnel plot with Egger's test. The test revealed that there was no statistically significant asymmetry in the plot (intercept −1.9; 95% CI −3.864–0.056; *p* = 0.07), thus arguing against publication bias. Given the strong trend of Egger's test and the fact that one study (Fenoy et al. 2018) was a clear outlier, as depicted in the funnel plot, we re-analyzed the data with this outlier study excluded. Using a random-effects model, a meta-analysis of the pared data confirmed that active stimulation results in a

greater decline in HDRS/MADRS scores relative to sham stimulation (SMD = −0.62; 95% CI −0.95 to −0.30; *p* = 0.0002). Removing the outlier study decreased study heterogeneity (I2 = 45%) and decreased the likelihood of publication bias as estimated by Egger's test (intercept −1.3; 95% CI −3.26–0.66; *p* = 0.21).

**Figure 4.** Funnel plot of studies included in the quantitative analysis.

We examined and compiled the adverse events reported in the studies included in the quantitative meta-analysis. The adverse events occurring in greater than 1% of patients are listed in Table 2, and the full list of adverse events in each study is detailed in Supplementary Table S1. The most common complaint was headache (26% of patients), followed by visual disturbances (21%), worsening depression (16%), sleep disturbances (16%), and anxiety (14%). All other adverse events were only seen in less than 10% of patients (Table 2). The authors of the original studies reported that the vast majority of adverse events were transient and were often resolved by stimulation parameter adjustment. The headaches were often postoperative and resolved a few days after surgery. A significant number of patients (*n* = 16, 8%) expressed suicidal ideation, and a similar number of patients (*n* = 15, 8%) attempted suicide. Completed suicides were rare. In two studies, one patient from each study who had no response to DBS committed suicide [15,16]. In one large study, there were two deaths by suicide in the control group during the open-label phase [17]. Finally, in another study, two patients committed suicide [23]. These suicides were deemed to be unrelated to DBS, because both patients had a history of suicide attempts and DBS did not appear to increase impulsivity [23].


**Table 2.** Adverse events.
