Transcranial magnetic stimulation in anxiety and trauma‐related disorders: A systematic review and meta‐analysis

Abstract Background Transcranial magnetic stimulation (TMS) has been evaluated as an effective treatment option for patients with major depressive disorder. However, there are limited studies that have evaluated the efficacy of TMS for other neuropsychiatric disorders such as anxiety and trauma‐related disorders. We reviewed the literature that has evaluated TMS as a treatment for anxiety and trauma‐related disorders. Methods We searched for articles published up to December 2017 in Embase, Medline, and ISI Web of Science databases, following the Preferred Items for Reporting of Systematic Reviews and Meta‐Analyses (PRISMA) statement. Articles (n = 520) evaluating TMS in anxiety and trauma‐related disorders were screened and a small subset of these that met the eligibility criteria (n = 17) were included in the systematic review, of which nine evaluated TMS in posttraumatic stress disorder (PTSD), four in generalized anxiety disorder (GAD), two in specific phobia (SP), and two in panic disorder (PD). The meta‐analysis was performed with PTSD and GAD since PD and SP had an insufficient number of studies and sample sizes. Results Among anxiety and trauma‐related disorders, TMS has been most widely studied as a treatment for PTSD. TMS demonstrated large overall treatment effect for both PTSD (ES = −0.88, 95% CI: −1.42, −0.34) and GAD (ES = −2.06, 95% CI: −2.64, −1.48), including applying high frequency over the right dorsolateral prefrontal cortex. Since few studies have evaluated TMS for SP and PD, few conclusions can be drawn. Conclusions Our meta‐analysis suggests that TMS may be an effective treatment for GAD and PTSD.

TMS is a biomedical application of Faraday's principle of electromagnetic induction, and it works by generating strong and rapidly changing electric currents in a circular coil that is placed on the surface of the skull. This primary current generates a magnetic field that travels unimpeded through the hair, soft tissue, skull, and cerebrospinal fluid (i.e., these structures are minimally affected by the magnetic field) until it reached the neurons of the cortex. At this level, the magnetic field converts back into a (secondary) electrical current able to depolarize neurons and force an action potential, which will then travel from synapse to synapse across an entire functional circuit of interest (Camprodon & Pascual-Leone, 2016). In a parameterdependent manner, TMS can induce long-lasting plastic changes and can cause either a long-term potentiation-like effect or a long-term depression-like effect on cortical neurons, and this can modulate the physiological dynamics across brain regions and networks (Huerta & Volpe, 2009). In this context, TMS has the potential to therapeutically modulate aberrant circuit properties across neuropsychiatric conditions with maladaptive circuit dynamics. Recent technical development has introduced variants of the traditional repetitive TMS (rTMS) protocols such as deep TMS (dTMS) or theta burst stimulation (TBS), both with current FDA-clearance for the treatment of OCD and MDD, respectively.
Anxiety and trauma-related disorders include conditions related to maladaptive fear processing and related behavioral changes (Marin, Camprodon, Dougherty, & Milad, 2014). Anxiety is a broad clinical concept and occurs with different features in each disorder and individual, like the anticipation of future, sudden periods of intense fear with somatic sensations, or worry of being judged. The most prevalent anxiety disorders in adults are generalized anxiety disorder (GAD), panic disorder (PD) and agoraphobia, specific phobia (SP), and social anxiety disorder (SAD) (Bandelow & Michaelis, 2015).
Before the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5), PTSD was also considered an anxiety disorder (Association, 2000).
The lifetime comorbidity rates of PTSD with other psychiatric disorders range from 62% to 92% (Perkonigg, Kessler, Storz, & Wittchen, 2000). Furthermore, there is evidence that PD, GAD, and PTSD may have a common genetic predisposition (Chantarujikapong et al., 2001). There is a significant percentage of patients who suffer from these disorders and show no improvement after several trials with pharmacotherapy and cognitive behavior therapy (Ballenger et al., 2004). This highlights the need to continue therapeutic development research for anxiety disorders, and to consider the role of device-based interventions such as TMS. The objective of this systematic review is to review and evaluate the existing literature on TMS for treating anxiety disorders and PTSD.

| Literature review
We screened Embase, PubMed, and ISI Web of Science (up to December 2017) following the recommendations of the Preferred Items for Reporting of Systematic Reviews and Meta-Analyses (PRISMA) statement (Moher, Liberati, Tetzlaff, & Altman, 2009). The search terms used were ("TMS" OR "Repetitive TMS" OR "Transcranial Magnetic Stimulation" OR "thetaburst") AND ("Anxiety Disorders" OR "Social Anxiety" OR "Generalized Anxiety Disorder" OR "Panic disorder" OR "stress disorder, post-traumatic" OR "Social, Phobia" OR "phobic disorder" OR "Phobia, Specific") NOT ("Obsessive-Compulsive Disorder" OR "Anxiety, Separation" OR "Neurocirculatory Asthenia" OR "Neurotic Disorders"). We also examined the reference lists from selected articles in search of papers that could be missing. Only original articles published in English were included. Studies with animals and duplicated references were excluded.

| Eligibility criteria and study selection
The eligibility criteria for the inclusion of studies in the present review were:

Highlights
• We reviewed TMS as a treatment for anxiety disorders and PTSD.
• TMS presented large effect sizes as a treatment for PTSD and GAD.
• Follow-up studies in GAD showed improvement of patients after TMS.
• Future studies should evaluate maintenance treatment. 1. Treatment of SP, SAD, GAD, PD, or PTSD diagnosed according to DSM-IV to DSM-5 or ICD-10 classifications.

2.
Intervention with any form of TMS with at least five sessions (except for SP), because this is the minimum number of sessions to induce plasticity and improve symptoms for long term, while in SP a short-term effect may be useful since the symptoms are more punctual (Racine, Chapman, Trepel, Teskey, & Milgram, 1995).
3. Report of response and remission rates, or score reduction on a validated scale of the investigated disorder.

Articles are written in English.
Controlled studies or open-label studies with or without randomization and retrospective studies were accepted. Two researchers evaluated titles and abstracts to select potentially eligible articles,  full papers were assessed to confirm eligibility whenever necessary, and divergences were solved by consensus.

| Quality assessment and data extraction
The assessment of the quality of the studies and risk of bias followed the Cochrane guidelines (Lundh & Gøtzsche, 2008). The pre-and posttreatment data extracted from each study consisted of study design, mean age, number of patients of each treatment group, TMS parameters (number of sessions, target and localization method, frequency, intensity, total pulses, type of coil), dropouts and reasons, scale scores mean and standard deviation (SD), response and remission rates, and period of follow-up. We contacted authors for additional data whenever necessary and we greatly appreciate the contributions of Dr. Zangen, Osuch, and Watts (Isserles et al., 2013;Osuch et al., 2009;Watts et al., 2012).

| Quantitative analysis
The analysis was performed with Stata 15. The primary outcome was the improvement of each disorder measured by a validated scale. The effect sizes of controlled studies were determined with the mean differences of sham versus active TMS using pretreatment and posttreatment score changes. In studies with one group, the effect sizes were estimated with standardized mean difference of pre-and postscores, in which the subject is its own control. The denotation of effect size is the same independent of the study design and can be analyzed together (Borenstein, Hedges, Higgins, & Rothstein, 2009). All effect sizes were weighted with Hedges' g, with a 95% confidence interval (CI) in a random effects model-which assumes variability across studies in terms of the effect size. In studies with three treatment groups, the active group with less effect was excluded. Heterogeneity between studies was assessed with the I-square test (I 2 ). In case of moderate or high heterogeneity (I 2 > 50%), a sensitivity analysis was done to determine the impact of each study on the results and a meta-regression was performed to evaluate the influence of each TMS parameter at a time. For studies without the SD of the total score of the primary outcome, the largest similar SD found in other studies was repeated, according to the Cochrane Handbook for Systematic Review (Higgins & Green, 2011). Publication bias was evaluated by funnel plots of effect size versus standard error and by Egger's test (Egger, Davey Smith, Schneider, & Minder, 1997).
The studies were analyzed in four groups: SP, GAD, PD, and PTSD since there were no articles about TMS in SAD. Furthermore, the meta-analysis was carried out only for GAD and PTSD since the other reviewed disorders do not have the minimum amount of studies and sample size needed to perform a meta-analysis.

| RE SULTS
A total of 643 references were found (165 in Embase, 360 in Medline, 113 in ISI Web of Science, and five through additional sources). Of those, 123 were duplicate references, and 37 were not in the English language. The remaining 483 references underwent a title and abstract analysis after which 419 were excluded. Finally, 64 articles were recovered for full-text reading. After this process, only 17 articles met the inclusion criteria of articles that assessed TMS as a treatment for anxiety disorders or PTSD (nine PTSD, four GAD, two SP, and two PD) ( Table 1). The meta-analysis of SP and PD was not performed because of the small number of studies and sample size. Figure 1 depicts a flow chart of the search results and selection of studies.
One study evaluated bilateral rTMS treatment in patients with comorbid GAD and MDD employing 1 Hz over the rDLPFC followed by 10 Hz over the left dorsolateral prefrontal cortex (lDLPFC) (White & Tavakoli, 2015). White and Tavakoli did not report the intensity applied on either side, nor the pulses delivered over the lDLPFC (White & Tavakoli, 2015). Last, one RCT applied 20 Hz, with 110% RMT over the rDLPFC (Dilkov et al., 2017). Figure 2 shows the weighted effect sizes of the studies.
The overall effect size was −2.06 (95%CI: −2.64, −1.48), widely favoring active rTMS treatment. There was low heterogeneity (I 2 = 11.6%, p = 0.335); therefore, the difference between studies is by chance. Possible causes of publication bias were tested with the funnel plot ( Figure 3), which showed no asymmetry (p = 0.705, Egger's test). Table 2 shows the reported dropouts and the number of dropouts due to side effects.

| TMS and posttraumatic stress disorder
The treatment of PTSD with TMS is the most studied among the conditions of interest. Nine studies were included in this meta-analysis  The funnel plot is symmetric (p = 0.992, Egger's test), suggesting that publication bias is unlikely. The reported dropouts and the amount of these that are due to side effects are in Tables 3-6.
All studies applied 1-20 Hz rTMS with traditional figure-ofeight coils to either the right or left DLPFC or both, with the exception of one study that evaluated the effect of dTMS to the medial PFC (mPFC) (Isserles et al., 2013). Six studies administered 10-15 sessions (Boggio et al., 2010;Cohen et al., 2004;Isserles et al., 2013;Nam et al., 2013;Rosenberg et al., 2002;Watts et al., 2012), two administered 20 sessions (Osuch et al., 2009;Oznur et al., 2014), and one 36 sessions (Philip et al., 2016). Concerning the sample characteristics, two studies assessed combat-related PTSD, and in one of these studies, all patients had a history of substance abuse (Oznur et al., 2014;Rosenberg et al., 2002). Also, four studies evaluated comorbid PTSD and MDD (Isserles et al., 2013;Osuch et al., 2009;Philip et al., 2016;Rosenberg et al., 2002). Three of the RCT consisted of three treatment groups (Boggio et al., 2010;Cohen et al., 2004;Isserles et al., 2013). One study compared 20 Hz rTMS over the right or left DLPFC against sham, and another study compared 1-10 Hz over the rDLPFC (Boggio et al., 2010;Cohen et al., 2004). F I G U R E 3 Funnel plot of the four studies that evaluated rTMS as a treatment for GAD nontraumatic events. Response was defined as an improvement of at least 50% in CAPS score. The response rate was 44% in the active-dTMS/traumatic images-group while in the active-dTMS/ nontraumatic images-group was 12.5% and, in the sham-dTMS/ traumatic images-group was 0% (Isserles et al., 2013). PTSD is characterized by intrusion or re-experiencing, avoidance, and hyperarousal clusters of symptoms (Ruggiero, Del Ben, Scotti, & Rabalais, 2003). In this study, they observed improvement of reexperiencing symptoms in the active-dTMS/traumatic imagesgroup (Isserles et al., 2013).
Three studies reported an improvement of all clusters of symptoms (Cohen et al., 2004;Philip et al., 2016;Watts et al., 2012), two studies reported an improvement only on the hyperarousal cluster (Osuch et al., 2009;Oznur et al., 2014), two studies reported an improvement only on the re-experiencing cluster (Isserles et al., 2013;Nam et al., 2013), and one study reported an improvement only on avoidance (Boggio et al., 2010). The two studies that applied rTMS over the lDLPFC in PTSD/MDD patients showed improvement of depressive symptoms as well (Philip et al., 2016;Rosenberg et al., 2002).
Four studies evaluated patients at follow-up intervals of 14 days (Cohen et al., 2004), 2 months (Rosenberg et al., 2002;Watts et al., 2012), or 3 months (Boggio et al., 2010). Three of these studies showed that there was a loss of improvement in PTSD symptoms at follow-up relative to the end of treatment despite the improvement from baseline (Boggio et al., 2010;Cohen et al., 2004;Watts et al., 2012). The one other study, which found that patients had improvements in MDD symptoms but not PTSD symptoms posttreatment, also found decreased depressive symptom improvement 2 months after the end of rTMS treatment (Rosenberg et al., 2002).
Tables 3, 4, 5, and 6 summarize studies that have evaluated the application of TMS in PTSD. Figure 5 depicts a forest plot for the meta-analysis evaluating TMS as a treatment for PTSD.

| TMS and specific phobia
We did not find any studies for SP with more than two treatment sessions. However, due to the peculiar features of the disorder with acute exacerbations that can be predicted in some situations, patients could benefit from short-lasting effects of stimulation. Two studies used single-session paradigms with a translational (not therapeutic) aim that are informative in the context of this review.
These studies evaluated rTMS or excitatory intermittent theta burst stimulation (iTBS) as a treatment for SP (Herrmann & Ebmeier, 2006;Notzon et al., 2015). Notzon et al. (2015)    interval of 8 s) with a pulse intensity of 80% of the resting motor threshold (RMT). One session of iTBS showed no improvement.
Previous studies showed the importance of the ventromedial prefrontal cortex (vmPFC) in fear extinction (Herrmann & Ebmeier, 2006). Since this brain area is too deep to be directly modulated by TMS, a research group used the strategy to indirectly stimulate this region through FPz, according to the electroencephalography (EEG) 10-20 system. This position had been identified as the center of the mPFC activation cluster by an increase of oxygenated hemoglobin during extinction of conditioned fear measured by NIRS in a prior study (Guhn et al., 2012). Herrmann and Ebmeier (2006)

| Side effects of TMS
Ten of the 17 studies (59%) included in this meta-analysis presented adverse events (Boggio et al., 2011;Cohen et al., 2004;Diefenbach et al., 2016;Dilkov et al., 2017;Herrmann & Ebmeier, 2006;Isserles et al., 2013;Mantovani et al., 2013;Nam et al., 2013;Notzon et al., 2015;Rosenberg et al., 2002). Most of the side effects were mild to moderate. However, two studies reported a single generalized tonicclonic seizure (Dilkov et al., 2017;Isserles et al., 2013). Both of these studies applied 20 Hz. One study used rTMS with 20 trains of 9 s, 51 s intertrain intervals, 110% RMT, 3,600 pulses/session, with a figureof-eight coil over the rDLPFC (Dilkov et al., 2017). A train of 9 s is long and may have contributed to the seizure. The other study used dTMS with 42 trains of 2 s, 20 s intertrain intervals, 120% RMT, 1680 pulses/session, with a H-coil over the mPFC (Isserles et al., 2013). This protocol parameters are in the upper limit of the parameters currently used for dTMS. Neither described clinical characteristics that could explain a higher risk of seizure.
Adverse events in patients who underwent active TMS were headache, neck pain, scalp pain, tingling, sleepiness, facial twitch, and impaired cognition during treatment. A PTSD study reported two patients with manic episodes: one patient in the 1 Hz-group and another in the 10 Hz-group (Cohen et al., 2004). Few studies reported the adverse events of the sham group separately, but these included neck and scalp pain, headache, impaired cognition, dizziness, sleepiness, and discomfort with treatment and the study schedule (Boggio et al., 2010;Diefenbach et al., 2016;Isserles et al., 2013;Mantovani et al., 2013;Nam et al., 2013). One PD study reported hearing impairment, mainly in the sham group (Mantovani et al., 2013). Adverse events are described in Table 9.
Another critical issue is to evaluate the percentage of patients who dropped out due to adverse events. A quarter of the studies TA B L E 8 Therapeutic use of TMS in specific phobia   Article reported the adverse events of both active groups together.
reported the reasons for dropouts: the minority of dropouts was due to adverse events and no studies reported treatment ineffectiveness as a reason for dropouts. The causes of dropouts varied from withdrawal or improvement of the disorder before starting treatment, to impossibility to determine the motor threshold, and technical error (Cohen et al., 2004;Dilkov et al., 2017;Rosenberg et al., 2002). Considering studies that evaluated TMS as a treatment for PTSD, one study reported two dropouts: one because of increased anxiety and one due to unease (Isserles et al., 2013), and another reported one dropout in a PTSD sample due to marked headache (Rosenberg et al., 2002). Therefore, there was no difference in the dropout rate due to adverse events between active and sham TMS treatments. However, only 24% of the studies reported in detail the reasons for dropouts per treatment group.

| D ISCUSS I ON
This review analyzes existing studies that evaluated TMS as a treatment for anxiety disorders or PTSD. Regarding GAD, the overall effect size largely favors TMS treatment (Bystritsky et al., 2008;Diefenbach et al., 2016;Dilkov et al., 2017;White & Tavakoli, 2015). Three of the four studies targeted the rDLPFC, two with 1 Hz inhibitory TMS and one with 20 Hz excitatory TMS (Bystritsky et al., 2009;Diefenbach et al., 2016;Dilkov et al., 2017). The other study associated 1Hz-rTMS over the rDLPFC and 10Hz-rTMS over the lDLPFC since the sample had comorbid GAD and MDD, and achieved high remission rates in both disorders (GAD: 84.6%, MDD: 76.9%) (White & Tavakoli, 2015).
The only study that used 20 Hz on the right side (as opposed to the usual 1 Hz) and 110% RMT presented the best response and remission rates, and highest effect size (Dilkov et al., 2017). Three GAD studies reported follow-ups from 1 to 6 months. The 6-month follow-up showed sustained improvement and the follow-ups of 1 and 3 months showed that patients were better when compared to the end of TMS treatment (Bystritsky et al., 2009;Diefenbach et al., 2016;Dilkov et al., 2017).
In relation to PTSD, the overall effect size was also large (Boggio et al., 2010;Cohen et al., 2004;Isserles et al., 2013;Nam et al., 2013;Osuch et al., 2009;Oznur et al., 2014;Philip et al.., 2017;Rosenberg et al., 2002;Watts et al., 2012). Considering the four PTSD studies that have larger effect sizes and small variability (all of these randomized, sham-controlled trials), there are indications that the rDLPFC is a better target to treat PTSD and anxiety symptoms when compared to the lDLPFC. Furthermore, two of these four studies applied high-frequency rTMS (10 and 20 Hz) over the rDLPFC and compared with low frequency over the rDLPFC or high frequency over the lDLPFC and, in both studies, high-frequency rTMS (10 and 20 Hz) over the rDLPFC showed greater improvement (Boggio et al., 2010;Cohen et al., 2004). The only trial that used dTMS could not demonstrate a substantial treatment effect of 12 sessions over the mPFC (Isserles et al., 2013). Therefore, further studies could assess the efficacy of dTMS with more sessions and over other cortical areas.
In three of the four studies that treated patients with comorbid MDD, which affects half of patients with PTSD, there was no significant improvement of depressive symptoms. The study that achieved response rates of 40% for PTSD and 50% for MDD applied 36 rTMS sessions while the other studies applied 10-20 sessions. The standard TMS course as a treatment for MDD consists of at least 30 sessions. Therefore, it is likely that a greater number of sessions could assign better results for both MDD and PTSD. The three PTSD studies that followed patients from 14 days to 3 months already found deterioration of PTSD improvement relative to the end of TMS treatment, despite remaining better when compared to baseline (Boggio et al., 2010;Cohen et al., 2004;Rosenberg et al., 2002;Watts et al., 2012).
In general, these results suggest that rDLPFC rTMS might have therapeutic activity in GAD and PTSD and that both high-and low-frequencies work. Therefore, despite the low-frequency Notably, these differences may be due to the pathophysiological differences of the two disorders, which would require unique approaches to induce therapeutic plasticity (Camprodon & Pascual-Leone, 2016). Appropriately powered randomized controlled trials should be considered to empirically confirm and validate these meta-analytical conclusions.
SP is still neglected, so almost no conclusions can be drawn except that treatments with more than one session should be used with intensities of at least 100% MT. Similarly, it is difficult to make assumptions on the use of TMS as a treatment for PD based on two small and heterogeneous trials. However, there are indications that 1 Hz over the rDLPFC may work with intensities higher than 100% RMT. On the other hand, future studies may clarify whether the failure of PD treatment on the left side was due to laterality or the iTBS technique.
TMS seems to be safe and well tolerated by patients with anxiety disorders or PTSD, although we found major gaps in the reports of these data. Two thirds of the studies in this meta-analysis reported the side effects but four of these studies just reported the types of side effects without mention of frequency or relation to treatment group. This is an important gap that highlights the need to systematically assess and report adverse events with validated questionnaires. This practice would allow for a comparison across treatment conditions and risk-benefit analysis.

| LI M ITATI O N S
One limitation of our meta-analysis is that 12 of the 17 studies were performed with small sample sizes of less than 20 subjects in each group. Moreover, across the reviewed studies, there is an absence of uniformity on the study design and how outcomes are measured and reported. These factors make it difficult to generalize the results, although meta-analytical approaches exist and were used. Furthermore, there may have been language bias since only English studies were included. However, it is unlikely that this bias would not interfere with the results of the meta-analysis.
Finally, the lack of reporting of adverse events restricts the evaluation of safety and tolerability.

| CON CLUS ION
While there are still limited data on the effectiveness of TMS in anxiety or trauma-related disorders (few studies, with small samples and diverse study designs and protocols), a number of trials have been published particularly for GAD and PTSD. Our metaanalysis concludes an overall positive therapeutic effect of TMS for these two conditions. These results suggest (but do not prove) an advantage of right over lDLPFC stimulation, and the possible therapeutic advantage of high-frequency stimulation to the rDLPFC. Based on the studies that reported side effects, TMS demonstrated to be safe and well tolerated in the treatment of anxiety disorders and PTSD but reports of side effects were inconsistent. In summary, the result of this meta-analysis confirms the therapeutic potential and safety of TMS for GAD and PTSD and generates some hypotheses for upcoming prospective, larger, and appropriately powered randomized controlled trials to confirm these results.

ACK N OWLED G M ENTS
The authors gratefully acknowledge Dr. Watts, Osuch, and Zangen for contributing with additional data. This research was partly supported by NIH grants (RO1 MH112737, R21 DA042271, R21 AG056958 and R21 MH113018) to JAC. Funding for this project was provided through a scholarship from CAPES for PC. This study was supported in part by the Dauten Family Center for Bipolar Treatment Innovation.

D I S C LO S U R E S TAT E M E N T
AKG receives research support from NIMH. AAN reports the following disclosures: Consultant -Abbott Laboratories, Alkermes,