Neuromodulation Treatments of Pathological Anxiety in Anxiety Disorders, Stressor-Related Disorders, and Major Depressive Disorder: A Dimensional Systematic Review and Meta-Analysis

Background Pathological anxiety is responsible for major functional impairments and resistance to conventional treatments in anxiety disorders (ADs), posttraumatic stress disorder (PTSD) and major depressive disorder (MDD). Focal neuromodulation therapies such as transcranial magnetic stimulation (TMS), transcranial direct current stimulation (tDCS) and deep brain stimulation (DBS) are being developed to treat those disorders. Methods We performed a dimensional systematic review and meta-analysis to assess the evidence of the efficacy of TMS, tDCS and DBS in reducing anxiety symptoms across ADs, PTSD and MDD. Reports were identified through systematic searches in PubMed/Medline, Scopus and Cochrane library (inception to November 2020), followed by review according to the PRISMA guidelines. Controlled clinical trials examining the effectiveness of brain stimulation techniques on generic anxiety symptoms in patients with ADs, PTSD or MDD were selected. Results Nineteen studies (RCTs) met inclusion criteria, which included 589 participants. Overall, focal brain activity modulation interventions were associated with greater reduction of anxiety levels than controls [SMD: −0.56 (95% CI, −0.93 to−0.20, I2 = 77%]. Subgroup analyses revealed positive effects for TMS across disorders, and of focal neuromodulation in generalized anxiety disorder and PTSD. Rates of clinical responses and remission were higher in the active conditions. However, the risk of bias was high in most studies. Conclusions There is moderate quality evidence for the efficacy of neuromodulation in treating pathological anxiety. Systematic Review Registration https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=233084, identifier: PROSPERO CRD42021233084. It was submitted on January 29th, 2021, and registered on March 1st, 2021. No amendment was made to the recorded protocol. A change was applied for the subgroup analyses based on target brain regions, we added the putative nature (excitatory/inhibitory) of brain activity modulation.


INTRODUCTION
Pathological anxiety is a hallmark of mental disorders encompassing anxiety disorders, posttraumatic stress disorder and major depressive disorder. Anxiety disorders (ADs), including generalized anxiety disorder (GAD), panic disorder (PD), social anxiety disorder (SAD) and specific phobias (SP), are characterized by a functional impairment due to excessive anxiety. Anxiety is also a core symptom cluster in posttraumatic stress disorder [(PTSD; (1)]. Anxiety is a possible symptom in major depression, and anxious forms of depressive disorders are characterized by higher clinical severity (2)(3)(4). Moreover, ADs, PTSD and major depressive disorder (MDD) are highly comorbid internalizing disorders (5). They affect a large population, with lifetime prevalence of 15-20%, 2-7%, and 13-16%, respectively (6,7), and cause heavy individual and financial burden, summing up to over 150 billion e/year in Europe (8)(9)(10)(11). Current therapeutic strategies to manage these conditions rely on pharmacology (antidepressants) and psychotherapy (chiefly cognitive based therapy). However, more than one third of patients with ADs, PTSD or MDD are unresponsive to those conventional treatments (12)(13)(14)(15). The intensity of anxiety in MDD is itself predictive of poor therapeutic outcomes (16,17). As a result, effective treatment of pathological anxiety remains an unmet, pressing need. Key brain regions implicated in ADs, PTSD and MDD are largely overlapping, including the prefrontal cortex and the amygdala (18)(19)(20)(21), suggesting that interventions aiming at restoring physiological functioning in this brain network may represent complementary therapeutic options.
In line with this, heavy research efforts are being invested in the development of novel therapeutic tools to control brain activity in ADs, PTSD and MDD, including neuromodulation techniques. Available neuromodulation techniques modify the electrical activity of targeted brain areas with the application of external magnetic (Transcranial Magnetic Stimulation -TMS) or electrical fields (transcranial Direct Current Stimulation-tDCS), or direct intracerebral currents (Deep Brain Stimulation-DBS) (22)(23)(24). In TMS, magnetic pulses induce action potentials in local neuronal populations, and their repetition at low or high frequency inhibits or facilitates synaptic transmission and neuronal firing (23). In tDCS, the locally induced electrical field alters membrane polarization and excitability at subthreshold levels, resulting in neuronal firing facilitation dependent on ongoing synaptic activities in areas close to the anode, and inhibition near the cathode (24). The mechanisms of action of DBS remain poorly understood, potentially involving depolarization of axonal fibers, excitation of local neurons (principal neurons and interneurons), or modulation of firingpatterns of neuronal populations (22). In MDD, TMS and tDCS have demonstrated clinical efficacy (25)(26)(27) and DBS has shown promising results (28,29). ADs and PTSD may benefit from TMS and tDCS, as summarized in several recent reviews (30)(31)(32)(33)(34) and meta-analyses of the literature (35)(36)(37). These neuromodulatory techniques might also provide beneficial effects on MDD-related anxiety. In keeping with their clinical similarities and shared neurobiological substrates, brain stimulation protocols in ADs, PTSD and MDD are similar, with the most frequent procedures consisting in the activation of the left and/or inhibition of the right dorsolateral prefrontal cortex (dlPFC) (26,27,34,35).
Paradoxically, while ADs, PTSD and MDD are neurobiologically related and neurostimulation strategies applied in these disorders are similar, recent quantitative syntheses of the literature have assessed their efficacy on the basis of disorder-specific psychometric measurements (35)(36)(37). This categorial approach evaluates symptoms that belong to cognitive, affective, or behavioral domains, with items differing among disorders. Such heterogeneity precludes the global evaluation, and the synthesis of knowledge in neuroscience-based brain stimulation for pathological anxiety. In contrast, a dimensional, transdiagnostic evaluation is critical to evaluate and improve neuroscience-based treatments of pathological anxiety.
To tackle this limitation in the literature, we applied a dimensional approach to evaluate brain stimulation-based treatments for pathological anxiety across related disorders. Systematic review and meta-analysis were performed with the main objective to study the efficacy of spatially targeted brain stimulation techniques on the intensity of generic anxiety symptoms in ADs, PTSD and MDD. The secondary objective was to synthetize available evidence for anxiety reduction with a categorial approach for ADs and PTSD, based on disorderspecific rating scales. This will facilitate comparisons with previous, categorial studies, and will update available evidence.

Search Strategy
This systematic review was conducted following the PRISMA guidelines (38). Two independent investigators performed the review (FG, TB). A literature search was performed on Pubmed, Scopus, and the Cochrane library, from inception until 31 November 2020. The following terms were used on the three databases: ("brain stimulation" OR "TMS" OR "RTMS" OR "Transcranial Magnetic Stimulation" OR "Theta burst" OR "transcranial direct current stimulation" OR "TDCS" OR "DBS" OR "neuromodulation") AND ("Anxiety disorders" OR "agoraphobia" OR "social, Anxiety" OR "phobic disorders" OR "Panic disorder" OR "Phobia" OR "stress disorder, posttraumatic" OR "PTSD" OR "anxious symptoms" OR "HAMA"

Data Extraction
Included articles were reviewed to extract relevant data : article indexing information (authors, year of publication), study design (control, randomization, blinding), sample characteristics (diagnosis, number of patients in each arm of treatment, mean age, gender ratio), brain stimulation protocols (technique, target brain region, target identification method, stimulation and sham systems descriptions, stimulation settings, duration and number of sessions), concomitant therapeutic interventions, symptom measurement tools, dropout rates, and pre-and posttreatment data, including follow-up data (attrition rates, mean score and standard deviation of scale ratings, response and remission definitions and rates, adverse events). Study authors were contacted directly to retrieve data unavailable in original publications (twice in the absence of response to the first request). Data were collected in the form of an electronic spreadsheet (Excel).

Outcomes
The primary outcome was the difference in intensity of anxiety symptoms between the baseline and the end of the treatment period (pre-and post-treatment), measured using validated, generic, quantitative scales of anxiety. Pre-specified secondary outcomes were: (1) the rate of clinically significant response or remission (as defined by study authors), (2) the intensity of anxiety symptoms at the last available follow-up (using the same scales as in the primary outcome analysis), (3) the difference in the severity of stressor-related and anxiety disorders between baseline and the end of the treatment period measured using validated, disorder-specific quantitative scales, (4) the adverse events rates and nature.

Strategy for Data Synthesis (Meta-Analysis)
Analysis was performed using R software (R R 4.1.2). For the treatment and control groups of each study, we extracted (1) baseline/post-treatment mean differences in general anxiety rating scales and/or specific disorder rating scales, and (2) response and remission rates. The change-from-baseline SD was calculated following guidelines of the Cochrane Handbook for Systematic Reviews of Interventions (39).
Where Corr represent the correlation between pre-and posttreatment variance, and was set to 0.5 as suggested by Follman et al. (40). We calculated between-group standardized mean difference (SMD) and their variance for continuous variables, and Hedges' g to control for small sample bias, as suggested by Harrer et al. (41). Relative Risk and their variance were computed using the "escalc" function of the "epitools" package. The meta-analysis was conducted using a random model with the "rma.mv" function of the "metafor" package. A multi-level model was fitted to account for multiple active or control groups and to control for dependencies between effect sizes. Comparisons were made between active brain stimulation and control conditions. When two validated quantitative scales evaluating the same outcome were available (such as HARS and BAI), results from the scale most frequently found in the other studies were selected for the meta-analysis (see Supplementary Material for details).
Pre-specified subgroup analyses of primary and secondary outcomes were performed for specific disorders (GAD, MDD, PD, PTSD, SAD, SP), stimulation techniques (tDCS, TMS, DBS) and target brain regions. From the pre-registered protocol, we modified the latter subgroup definition by categorizing the putative nature (excitatory/inhibitory) of brain activity modulation. Neuromodulation was considered excitatory for TMS for frequencies ≥ 5 Hz, intermittent theta burst stimulation (iTBS) and for brain regions located below anodes in tDCS. Stimulation was considered inhibitory for low TMS frequencies (mainly 1 Hz) and cathodal tDCS. Multisite neuromodulation protocols were not included in the analysis of discrete target regions. Likewise, when studies were composed of three arms, of which two were active conditions targeting distinct brain regions, individual active arms were included in the corresponding brain target subgroup.
A meta-regression analysis was conducted to explore potential sources of heterogeneity, using the "rma.mv" package. We assessed the putative moderation effect of the number of sessions of stimulation, the total number of pulses and the motor threshold percentage in TMS studies, the difference in anxiety scores at baseline between active and sham groups (SMD) and the difference between active and sham groups for baseline to posttreatment changes in depression and PTSD scores (SMD). The model was assessed for each factor one by one.
Twenty-one RCTs were included for the secondary analysis, which was based on disorder-specific scales. Those 21 studies comprised 13 RCTs included in the primary outcome analysis [exclusion of one trial without specific scale measure (48) and five MDD trials (43,45,50,57,59)], and eight additional RCTs, in which only disorder-specific scores were reported (61)(62)(63)(64)(65)(66)(67)(68) (Tables 1, 2). This secondary analysis involved 712 subjects (mean age: 40.3 years; females: 49.9%). Of the 21 studies, 15 trials assessed TMS interventions (two in GAD, three in PD, nine in PTSD, one in SP), and six assessed tDCS interventions (three on GAD symptoms, three on PTSD symptoms). No controlled trial using DBS was found for anxiety or stressorrelated disorders. Although not excluded a priori, no article reporting continuous Theta Burst Stimulation (cTBS) protocol was found. Detailed reporting of patients' characteristics, study designs and neurostimulation protocols are provided in the Supplementary Material.

Rates of Clinically Significant Response and Remission
Meta-analysis was performed on response and remission rates, as defined by the authors. Criteria were alternatively defined on the basis of generic or disorder-specific scales (see Table 2).
Overall, neurostimulation was associated with higher response rates than control interventions  Figure 6].

Follow-Up Results
Data of follow-up visits were not submitted to meta-analysis because of a high heterogeneity in the length of follow-up periods, data reporting and attrition rates (see Table 4 for detailed reporting). For TMS, three studies stopped follow-up visits after 2-4 weeks (44, 51, 56), 7 after 1-3 months (43, 47-49, 52, 59, 60), and one at 6 months (54). Four studies showed persistent differences between active and sham groups at the end of follow-up in favor of neurostimulation (47,48,51,54) and two reported data suggesting the persistence of differences between groups, although statistics were not presented (52,59). There was no statistically significant difference between groups in two trials (43,49). Finally, one article reported incomplete data (60).
In tDCS studies, follow-up visits were also conducted at different time points: 1 week (53), 4 weeks (42), 2 months (58), and 3 months (55). Three studies described a persistently significant difference between active and sham groups (42,55,58). One trial presented within-group comparisons, but no statistics were provided for between-group comparisons at the end of follow-up (53).
In DBS studies, all subjects underwent an unblinded treatment phase following the endpoint evaluation. Thereafter, they received an active stimulation for six consecutive months (50), or until loss to follow-up (57). Values of HARS were similar between treatment groups at 6 months follow-up in both studies.

Adverse Events
Data regarding adverse events were mostly reported irrespective of intervention groups (active or control). They are therefore reported in a descriptive manner (see Supplementary Material for a more detailed reporting). For TMS studies, serious adverse events comprised three cases of generalized tonic-clonic seizures in active groups (43,48,62), two suicidal ideation episodes in the active group (52), one homicidal ideation emergence, one hospitalization for suicidality during follow-up [both receiving sham stimulation; (68)], two manic and one hypomanic episodes [all receiving active stimulation; (43,44)], and one chest pain occurrence in the active group, which was considered to be unrelated to the intervention (47). No serious adverse event was reported for tDCS. Other adverse events were mild, including: (i) TMS: headaches, pain at the stimulation point, neck, face (when specified, there was no significant difference between active and sham conditions for these side effects), dizziness/nausea, ipsilateral facial twitch, and ear discomfort; (ii) tDCS: headache, tingling, sleepiness, mild itching or burning sensations on the scalp and light skin redness. Only sleepiness, sensations on  the scalp and skin redness were more frequently reported in active groups. Side effects were reported in only one of the two DBS studies (50). Serious adverse events at 6 months were infections (six patients in the active stimulation group, one sham) related to the study device or surgery, hospitalization (one active) for unspecified reason, suicide attempt (one active), suicidal ideation (one active), seizures (one active, one sham), hearing or visual disturbances (one active), local erosion over the pulse generator (one active), serious postoperative discomfort (8). The most frequent adverse events were mild, including headache, acute pain/discomfort, persistent pain, anxiety, increase in depressive symptom intensity, nausea/vomiting, sleep disturbances, paresthesia, and dizziness (50).

Meta-Regression Analysis
In the meta-regression analysis, the difference in pre/posttreatment changes of depression scores between active and sham arms (Coefficient = 0.86, p = 0.0003) significantly accounted for the heterogeneity [heterogeneity test QE(df = 19) = 19.6, p = 0.42]. No statistically significant link was found for other covariates, namely: the baseline anxiety scores, the changes in PTSD scores, the gender ratio and the mean age of participants, the total number of sessions during the protocol and, in the TMS studies, the total number of pulses and the percentage of motor threshold did not explain results heterogeneity. Statistical characteristics for each factor are reported in the Table 5.

Publication Bias
Visual inspection of the funnel plot of studies included for primary outcome analysis revealed an asymmetric distribution of studies, indicating that publication bias may have occurred ( Figure 8A). However, Egger's test was not indicative of a statistically significant bias (p = 0.46). There was no evidence of publication bias for studies included for the analysis of disordersspecific scales (Figure 8B; p = 0.59).

Risk of Bias in Methodology
Detailed evaluation of the risk of bias is reported in Supplementary Contrary to what was initially planned in the registered protocol, no sensitivity analysis was conducted in the meta-analysis, because of the insufficient number of studies demonstrating a low risk of bias.
Limitations in blinding, especially neurostimulation blinding (10/19 and 11/21, respectively), and attrition bias (9/19 and 11/21, respectively), accounted for almost the totality of the high risk of bias. Conversely, outcome blinding or selective reporting of results were generally considered at low risk of bias. In most cases, methods of random sequence generation and randomization concealment were not reported. Only six of the 27  studies applied and reported appropriate randomization (48,50,52,57,63,66). Overall, according to the GRADE guidelines (39), evidence from this study was considered of moderate quality.

DISCUSSION
The treatment of pathological anxiety in ADs, PTSD and MDD is faced with high levels of resistance. Focal brain stimulation is a promising therapeutic avenue. However, there is currently a lack of synthesis of results from neuromodulatory interventions for anxiety across those disorders. This review and meta-analysis summarized available evidence regarding the efficacy of focal brain stimulation on pathological anxiety, with a dimensional perspective. Results are globally in favor of a positive effect of neurostimulation at the end of interventions across ADs, PTSD and MDD, with a medium effect size, and a moderate quality of evidence. Active neuromodulation protocols also resulted in greater response and/or remission rates than control conditions. These results suggest that focal brain stimulation may be an interesting option for the treatment of pathological anxiety. However, long-term efficacy data were insufficient to conduct a meta-analytic study, being scarce and heterogeneous. Nevertheless, follow-up data were qualitatively in favor of a prolonged favorable effect of neurostimulation. The meta-analysis purposely gathered results from heterogeneous conditions and protocols, which were investigated in subgroup analyses. Regarding stimulation techniques, both TMS and tDCS showed a tendency toward an improvement of anxiety symptom intensity and resulted in greater response and/or remission rates than control conditions immediately after treatment. Mild adverse events typical of the stimulation techniques were reported. DBS showed a tendency toward an improvement of anxiety symptom intensity immediately after treatment. However, the meta-analysis was limited to two studies on MDD, which included patients with moderate baseline anxiety levels. This analysis is not sufficient to determine the therapeutic potential of DBS. As a result, at present, only TMS and tDCS appear as potential neuromodulatory options for the treatment of pathological anxiety. As for categorically defined disorders, positive effects were found in PTSD and GAD subgroups, on the basis of psychometric scores measuring either generic anxiety or disorder-specific scores. Positive effects were also found for MDD-related anxiety. In contrast, there was no evidence for positive impact of brain stimulation in PD, and the only trial on patients with SP showed no positive effects. No study was found for SAD. Overall, our results indicate that in categorial terms, PTSD and GAD may benefit from non-invasive brain stimulation.
Notably, ADs, PTSD and MDD were pooled in the main analysis, as a consequence of the dimensional approach of the study. This strategy is in line with the National Institute of Mental Health research domain criteria initiative, which organizes mental disorders in constructs, e.g., the negative valence system, and allows a multi-level integration of neuroscientific findings to identify biological targets for treatment (69). The choice of the disorders included was guided by a line of arguments pointing toward similarities in neurobiological bases. From a phenomenological perspective, they are internalizing disorders that share the highest levels of comorbidities and a great number of reciprocal risk interactions (5). In agreement with this, their susceptibility profiles and potential genetic risk factors also overlap (5). Moreover, brain activation patterns identified in ADs, MDD and PTSD are similar, implicating a fronto-limbic network (18,21). On the contrary, obsessive-compulsive disorder was not included, as it is known to follow distinct, corticostriato-thalamo-cortical pathogenic mechanisms (70). Likewise, somatic and dissociative disorders (1) were considered too different in their expression and pathophysiology, and were not included. Interestingly, the finding of an association between anxious et depressive symptoms reduction in the meta-regression analysis may further suggest the existence of common biological underpinnings, or, at least, of close interactions between the two phenomena.
Regarding the relative efficacy of brain stimulation depending on target brain regions, both right dlPFC inhibition and excitation tended to alleviate anxiety symptoms, with a larger effect size for right dlPFC activation. Left dlPFC activation had no global effect. These findings are overall consistent with the negative correlation identified between pathological anxiety levels and the resting state connectivity strength between right dlPFC and amygdala (71). However, other authors reported similar correlations, but with left dlPFC (72,73). It should be emphasized that the classification of the net effect (excitatory vs. inhibitory) of brain stimulation is derived from simple observations of motor cortex responses, that may not directly translate to PFC. Moreover, the net effect of brain stimulation is dependent on a number of  factors, including stimulation parameters (duration, intensity etc.), variability in subjects' intrinsic cortical excitability, and ongoing pharmacological treatments (74). Thus, the "excitatory" or "inhibitory" classification should be interpreted with caution. Future preclinical studies should specifically investigate this issue in the PFC. Moreover, it should be reminded that our analysis pooled various stimulation techniques and protocols. Therapeutic effects of focal brain stimulation also depend on the accurate placement of stimulation devices above targeted brain regions. In most studies included in the meta-analysis, target area was determined by scalp distances anterior to regions where motor responses were obtained. This placement method likely resulted in imprecise, heterogenous targeting. Indeed, it has been demonstrated that results obtained with such positioning methods differ from those obtained with gold standard, brain imaging-based stimulation device positioning (75,76). Future clinical trials should aim to optimize stimulation parameters, and may primarily investigate the right dlPFC as a target, as our results would suggest. Results of the present study are generally in agreement with, and extend results from previous reviews of the literature, including narrative reviews (31,32,34) and meta-analyses (35-37, 77, 78). Among narrative reviews, the most closely related to our work systematically reviewed the effectiveness of non-invasive brain stimulation for the treatment of anxiety disorders across RCTs and open-label studies (34). Interestingly, no clear evidence was reported in this work for a positive effect of brain stimulation in SP, on a larger number of trials than reported in our meta-analysis [including open-label studies; (34)]. This suggests that current neuromodulation strategies have no efficacy in SP. Contrary to our results, Vicario et al. (34) suggested that neuromodulation may improve PD, and that left dlPFC stimulation is more efficacious. However, the majority of studies included in the review were not sham-controlled, possibly accounting for those differences with our results.
Three meta-analyses were recently conducted for noninvasive brain stimulation in PTSD (36), rTMS in GAD (37), and rTMS in ADs (35), reaching similar conclusions to the present review. Evaluating the efficacy of non-invasive brain stimulation in PTSD across RCTs revealed a strong positive therapeutic effect of TMS (36). Cui et al. reported a positive effect for rTMS on GAD, based on a much larger study sample size than reported here (n = 21 vs. 3). This discrepancy is due to the inclusion of 19/21 studies published in Chinese language by Cui et al. (37). Most consisted in inhibitory stimulation of the right dlPFC. Cirillo et al. reported a reduction in clinical severity after TMS, both in GAD and in PTSD, using disorder-specific scales, and suggested that right dlPFC excitatory stimulation may be superior to targeting left dlPFC (35). Despite reaching conclusions consistent with those obtained in our meta-analysis, these studies were restricted to single diagnostic categories, and/or stimulation techniques, and/or disorder-specific instruments.
Prior to our work, another transdiagnosis meta-analysis had been published. Trevizol and colleagues conducted a metaanalysis investigating the effects of TMS on generic anxiety scores across ADs, PTSD and OCD, and found no evidence for a positive effect, including PD, PTSD and OCD subgroups (77). Most RCTs (8/14) reported interventions on OCD, possibly accounting for differences between conclusions from this and our work. Four of those negative OCD studies targeted the dlPFC, further supporting the distinct pathophysiology of OCD, and its exclusion from neurobiologically-based dimensional studies of anxiety. Differences for the PTSD subgroup may be partly accounted for by the publication after 2016 of studies included in our review. During the conduct of our study -i.e., after the protocol was pre-registered-a closely related work has been released by Vergallito et al. (78). Using both disorder-specific and generic anxiety scores, they found that rTMS and tDCS reduced anxiety levels across ADs. This main finding is in agreement with our primary and secondary analyses. Depression scores were also lower in intervention vs. control groups (78). In addition to our analysis, one study was included for which we could not obtain original results despite contacting the authors (79). In contrast with Vergallito et al., we ranked risk of bias as high in a majority of studies, whenever, the sham stimulation induced sounds and/or bodily sensations that might have been readily identified by intervention administrators or subjects. Our risk of bias classification is conservative, attributing high risk of bias if intervention administrators were not blinded, or if participants could guess the intervention group. For instance, we conservatively ranked 90 • TMS coil placement as a high risk of performance bias. This classification may be mitigated by a metaanalysis reporting that blinding assessment was satisfactory in MDD participants receiving sham TMS with either sham coils or angled coil placement (80). However, in one study included here, while angled TMS placement was used and blinding quality was measured, patients guessed better than chance their allocation group (46). Notably, Vergallito et al. restricted the analysis to ADs as defined in DSM-5. As demonstrated by changes between the 4th and 5th versions of the manual, diagnostic categories are a moving target, and an obstacle for neurobiologically-informed research. Instead, similarities in the anxiety phenomenon across ADs, PTSD and MDD call for a dimensional evaluation, which constitutes the originality and relevance of the present study.

LIMITATIONS
First despite performing an extensive search of literature databases and applying PRISMA recommendations, a limited number of studies were included, with relatively small sample sizes [except for two studies measuring disorder-specific symptoms; (63,66)]. Second, the primary outcome of the meta-analysis (generic anxiety scales results) was the explicit primary outcome in only 5/19 of the studies included. Third, therapeutic efficacy was evaluated at the end of brain modulation trials, while long-lasting reduction of symptoms intensity is warranted. Fourth, as pharmacological treatments were often administered concomitantly with brain stimulation, the relative contribution of the two strategies and their interactions remain to be clarified. Fifth, although only RCTs were analyzed, the risk of bias was high in a majority of studies, including the three studies showing the strongest positive effects (42,48,55). This was mostly due to randomization methods (insufficiently described), attrition, and insufficient blinding. Moreover, we could not entirely rule out that publication bias had occurred.

FUTURE PERSPECTIVES
Larger-sized, multicentered studies with pre-specified primary outcomes are warranted to fully assess the potential efficacy of TMS, tDCS and DBS in pathological anxiety. Trials designed to demonstrate long-lasting therapeutic responses and remissions will be critical in order to support the application of brain stimulation strategies in daily practice. In this respect, the adoption of standard follow-up intervals will tremendously help reproducibility and generalizability of future findings. The relative contributions and interactions between medication, psychotherapy and neuromodulation will need to be clarified. In a dimensional view, generic anxiety scales should be used to assess outcomes, and inclusion criteria might even be defined in a dimensional manner. Care should also be taken to apply and report adequate randomization and blinding procedures, including measurements of blinding assessment. Moreover, sham conditions will need to be hardly distinguishable from active brain stimulation (81,82). Alternative brain targets should be investigated. It is likely that targeting dlPFC to treat pathological anxiety has been empirically carried forward -at least in part-from usual clinical practice in MDD [as indicated by (47)], and not necessarily informed by neuroscientific findings. Indeed, there is some, but limited evidence for the implication of dlPFC in pathological anxiety (71)(72)(73) and in the preclinical model of fear conditioning (83). In contrast, the anxiety circuit comprises key brain regions such as the medial PFC, the amygdala, and the insula (10,84) as demonstrated by a recent meta-analysis of fMRI data (18). Their neuromodulation might provide greater therapeutic outcomes than targeting dlPFC. However, those key nodes are located deeply within the brain. Their modulation will require the application of invasive or novel non-invasive DBS techniques [e.g., H-coil TMS: e.g., (62,85); interference-based deep tDCS: (86); transcranial ultrasounds: (87)].
Finally, several research lines aim to develop brain stimulation procedures in a precision medicine framework. In the spatial domain, neuromodulated regions could be targeted selectively in different patient "clusters" distinguished by symptoms and neurobiological (e.g., imaging) dimensions [e.g., (88)]. In the temporal domain, instead of delivering continuous patterns of regular pulses, stimulation paradigms could adapt to ongoing, pathological brain activities to restore physiological functioning [(i.e., "closed-loop" stimulation; e.g., (89)]. Overall, it is likely that empirical brain modulation will move toward precise, neurobiological data-based techniques.

CONCLUSION
Current evidence suggests that focal brain stimulation may be beneficial to alleviate pathological anxiety across ADs, PTSD and MDD. Additional, large-scale, RCTs measuring longterm effects of neuromodulation on anxiety are warranted. Future research will expand and refine our knowledge in the perspective of offering precision medicine neuromodulation tools to manage pathological anxiety in the context of high treatment resistance levels.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author/s.

AUTHOR CONTRIBUTIONS
FG, FS, BA, and TB: study design. FG and TB: data collection and manuscript drafting. FG, FS, AS, and TB: data analysis. FG, FS, AS, BA, and TB: revisions and approval. All authors contributed to the article and approved the submitted version.

FUNDING
This work was funded by a CCA Inserm Bettencourt grant to Thomas Bienvenu (R20005GS).