A systematic search was carried out following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) Statement [29]. The protocol was registered prior to the start of the search process. We registered our protocol on INPLASY with the following registration number: INPLASY 202310054. PRISMA checklist is shown in Supplementary Material.

On May 16, 2023, we carried out a comprehensive search across the MEDLINE (via PubMed), EMBASE, and Web of Science databases. In our original protocol, we intended to incorporate PsycInfo as one of our primary databases for the literature search. However, due to the subsequent constraints on resources, we had to prioritize our database selection. To carry out our comprehensive research on the various databases, we employed the following search string: (“gambling disorder” OR “problem gambling” OR “pathological gambling” OR “compulsive gambling” OR “gambling addiction” OR “gambling addictions” OR “problematic gambling” OR “pathological gamblers” OR “problem gamblers” OR “pathological gamblers” OR “gamblers anonymous” OR “gambling addicts” OR gambling) AND (“transcranial magnetic stimulation” OR TMS OR rTMS OR “repetitive TMS” OR “theta burst stimulation” OR “theta burst” OR TBS OR cTBS OR iTBS).

The string was modified, when necessary, to accommodate the specific formatting and search parameters of each individual database. Both the process of selecting studies and extracting data were conducted by two pairs of authors (AC/CCh and ADF/GT) in a blinded manner. All discrepancies were resolved by a third expert author (CCo).

Information was collected using a data extraction form located on Airtable, encompassing the following details: author, publication year, country, study design, patient characteristics, stimulation protocol (including the total number of sessions and session frequency), stimulation frequency and intensity, stimulation area, comparator, primary outcomes, and follow-up information. Data extraction was performed independently by two pairs of authors (AC/CCh and ADF/GT) and conflicts were resolved with the involvement of a third, experienced author (CCo).

The evaluation of risk bias was carried out utilizing the Cochrane risk-of-bias tool 2 for RCTs, while for non-controlled studies, the National Institutes of Health Quality Assessment Tool was employed [30]. The assessment of bias was carried out independently by two authors (ADF and AR), and any discrepancies were resolved through the intervention of a third experienced author (CCo).

Three of the included studies were RCT [31, 32, 33], three were open label [34, 35, 36], one was a case series [37], and one was a case study [38]. The earliest publication date was 2013 and the most recent was 2022. In our systematic review, we analyzed and synthesized studies published from 2013 to 2022. Specifically, two articles were published in 2013–2016 [33, 36] one in 2017 [31], one in 2018 [32], two in 2019 [9, 37], one in 2020 [34], and one in 2022 [35].

The therapeutic protocols identified were high-frequency (HF)-rTMS and cTBS. The areas of stimulation were the dorsolateral prefrontal cortex (DLPFC), the medial prefrontal cortex (mPFC), and the pre-supplementary motor area (pre-SMA). Treatment duration ranged from one day (one single session) to several weeks. Two studies used rTMS in augmentation to pharmacological therapy [34, 37], while there was one article assessing the deep ‘H coil’ TMS [36].

The following gambling-related outcome assessment scales were used: Gambling-Symptom Assessment Scale (G-SAS), Yale-Brown Obsessive-Compulsive Scale adapted for Pathological Gambling (PG-YBOCS), 100-mm Visual Analogue Scale (VAS) for cue-induced craving, Numeric scale for desire, Numeric scale for control, Gambling Craving Scale (GACS) (only the first three questions), the Yale-Brown Obsessive-Compulsive Scale (YBOCS), Gambling Urges Questionnaire (GUQ), and the Dannon and Ainhold Gambling Scale (DAGS).

A detailed description of the included studies is summarized in Table 1 (Ref. [9, 31, 32, 33, 34, 35, 36, 37]).

The evaluation of the potential bias in RCTs was conducted employing the Cochrane Risk of Bias assessment tool 2. [39]. All studies [31, 32, 33] were rated as having “some concerns” in domain 1 (bias arising from the randomization process). Indeed, although all the studies had implemented a randomization process, none guaranteed allocation concealment through the involvement of an external entity. Moreover, both Gay et al. [31] and Zack et al. [33] lack information regarding the applied randomization method, stating exclusively that this process had been carried out. Regarding domain 2 (risk of bias due to deviations from the in- tended interventions), Gay et al. [31] and Zack et al. [33], were categorized as having some concerns risk of bias, whereas Sauvaget et al. [32] was considered as having low risk in the same domain. All studies [31, 32, 33] showed low risk of bias in domain 3 (bias due to missing outcome data), and 4 (bias in measurement of the outcome). However, for domain 5, all studies were categorized as having “some concerns” because they did not specifically report planning the data analysis methods before the availability of the data.

A detailed description of the risk of bias assessment for the RCTs is summarized in Table 2 (Ref. [31, 32, 33]).

The evaluation of the quality of non-randomized, uncontrolled studies was conducted using the National Institutes of Health (NIH) tool [30]. Rosenberg et al. [36] did not provide clear details regarding the inclusion and exclusion criteria. None of the studies, except for Rosenberg et al.’s, reported establishing the inclusion/exclusion criteria before enrolling patients. Regarding sample size, except for Cardullo et al.’s study [37], no other study conducted a calculation to estimate the number of patients needed for inclusion. In all the studies where this calculation was not made, the sample size was deemed insufficient. None of the studies mentioned any heterogeneity in the implemented interventions, except for Rosenberg et al.’s study, where one patient included in the analysis had twice as many sessions (30) compared with the other included patients. Furthermore, only Rosenberg et al.’s study had a loss of follow-up rate of at least 20%, as one out of the five analyzed patients was lost to follow-up.

All studies [35, 36, 37, 38] conducted a statistical analysis to examine changes in various outcomes before and after treatment, providing a p-value. The exception was Rosenberg et al.’s study [36], which only reported mean values and standard deviations before and after treatment. Finally, all studies evaluated the outcomes at multiple time points, except for Rosenberg et al.’s study.

A detailed description of the quality assessment for the non-controlled studies is summarized in Table 3 (Ref. [34, 35, 36, 37]).

Overall, among the studies that applied excitatory protocols of rTMS to the left DLPFC, an improvement in gambling symptomatology was observed as assessed through G-SAS or by clinical observation [31, 34, 37, 38]. However, the same studies did not show a corresponding improvement in this outcome when measured using PG-YBOCS. These conflicting findings might stem from differences in the characteristics of the two assessment scales. Specifically, a study by Kim et al. [41] demonstrated discrepancies in terms of reliability, internal consistency, and agreement with both physician and patient ratings between the two instruments. Among these domains, the PG-YBOCS exhibited superior performance.

Zack and colleagues [33] stated that a single active HF-rTMS session (10 Hz) targeting the left DLPFC was linked to a decreased post-game “desire to gamble” when compared with the sham stimulation. They hypothesized that this would have decreased cravings and delay discounting in a similar manner to the effects observed in non-treatment-seeking smokers [42]. This reduction in impulsive decision-making, as evidenced by a shift towards delayed options, has also been observed in healthy volunteers undergoing HF-rTMS over the mPFC [43]. Additionally, the administration of a single session of high-frequency rTMS has shown to reduce acute cravings in several groups of addicted individuals [44, 45, 46, 47, 48], including those cravings for food [49]. Gay et al. [31] reached a similar conclusion: the authors stated that a single session of HF-rTMS targeting the left DLPFC decreased cravings triggered by cues, although it did not affect the overall PG-YBOCS scale score. The choice to use a single session was supported by previous evidence from studies reporting a reduction in binge eating within 24 hours following active rTMS compared with sham stimulation [50], thus assuming a similar mechanism in behavioral addictions.

Among non-controlled studies, three of them employed several sessions of HF-rTMS [34, 37, 38]. In all these studies, patients underwent rTMS targeting the left DLPFC. This choice was centered on the concept that reduced activity in the prefrontal pathways may be associated with a progressive loss of control over gambling urges and behaviors, as highlighted by Moccia et al. [51] and Schluter et al. [52]. Pettorruso and colleagues [38] conducted SPECT examinations based on the hypothesis of dopamine dysregulation in addictive disorders [53]. As such, they aimed to investigate the role of rTMS in restoring dopaminergic activity by increasing DAT levels, as demonstrated in cocaine users [54]. To the best of our knowledge, this is the first study to combine rTMS with SPECT, thus opening future directions in the application of neuromodulatory techniques in the treatment of patients with GD, as previously suggested [55]. However, in one patient with GD, a bilateral reduction in the tracer uptake within the striatum was observed, indicating a decrease in DAT availability following rTMS. The authors suggested that rTMS may enhance dopaminergic transmission through a down-regulation of DAT, potentially mediated by gene expression modulation induced by rTMS over the left DLPFC. Comparable results have been documented among individuals diagnosed with alcohol use disorder [56]. In the case series conducted by Cardullo et al. [37], the included population consisted not only of individuals with gambling addiction, but also those with cocaine use disorder, given the high prevalence of comorbidity between these two conditions. It is known that there are common alterations in the mesolimbic reward system between substance-related disorders and GD [57]. Notably, the study found a decrease in cocaine craving and a reduction in Gambling-Symptom Assessment Scale (GSAS) score. In this context, much evidence has been reported regarding the reduction of cocaine craving after rTMS treatment, particularly after multiple sessions of HF stimulation targeting the left DLPFC. It has been suggested that the same neuromodulatory property could potentially have a role in the treatment of GD as well [58].

One non-randomized controlled trial only supported the positive impact of multiple sessions of HF-rTMS directed at the left DLFPC as a treatment for gambling-related symptoms, although noteworthy differences were not detected in the PG-YBOCS scores [34]. Previous studies [59] have demonstrated that HF-rTMS targeting the left DLPFC leads to the release of dopamine in the striatum, affecting both the mesolimbic and mesostriatal circuits.

This finding indicates that the positive outcomes detected in the context of rTMS treatment may be associated with dopamine regulation/plasticity, as observed in individuals with cocaine use disorder [58]. The choice to use multiple sessions of HF-rTMS was based on the inconclusive results of previous studies employing single-session HF-rTMS targeting the same area. The authors proposed that multiple sessions would play a more crucial role in maintenance treatment and be able to sustain positive outcomes on general symptoms.

Only one study conducted by Sauvaget et al. [32] assessed the efficacy of LF-rTMS over the right DLPFC in reducing cue-induced craving. The authors addressed cue-induced craving in their study using LF-rTMS in a single session over the right DLPFC. They ascribed their lack of success to a robust placebo effect and the specific parameters chosen for rTMS. This was grounded in earlier evidence that highlighted a connection between craving and heightened activity in the right DLPFC [60]. The authors induced craving by presenting visual cues and then asked the patients to complete a VAS assessment immediately after. Since visual-induced craving may contribute to this overactivation in pathological gamblers [61, 62], the authors indicated that the rTMS occurred during the peak level of craving, and it was anticipated that subsequent measurements would decrease at a later point in time.

Among the eight studies included, Rosenberg et al. [36] applied deep H-coil LF-rTMS over the left DLPFC. The rationale was substantiated by prior research that indicated heightened activation of the PFC in individuals when exposed to gambling-related stimuli in cue-exposure paradigms, as demonstrated by van Holst et al. in 2010 [63], and the possibility to reach dopaminergic subcortical areas with the use of an H-shaped coil [64]. The Deep H coil TMS was administered targeting the left DLPFC. Despite the patients reporting immediate improvement, this protocol showed ineffectiveness. Based on the initial effect, it was assumed that there might have been a transient positive, sham-like, effect, as it was unlikely a real effect of rTMS.

Zack et al. [33] assessed the efficacy of a single cTBS session directed over the right DLPFC. Regarding TBS, specifically continuous stimulation protocols have demonstrated a positive effect on impulsivity, decision-making, and delayed discounting by targeting the right DLPFC, which is a brain area known for its inhibitory control [43]. Zack and colleagues [33] administered a cTBS session over the right DLPFC compared with a sham stimulation. They did not observe any difference in the “desire to gamble”, as measured by a post-session VAS. Considering previous research by Ngetich et al. [65], who found that cTBS over the right DLPFC had mixed effects (resulting in impaired goal-directed behavior on one hand and reduced impulsivity on the other), the lack of significant findings in the desire to gamble could be explained by the interference caused by impaired goal-directed behavior on decision-making. Nevertheless, the study yielded significant findings in terms of secondary outcomes: these included a notable reduction in subjective psychostimulant-like arousal effects, in the Stroop interference test, and in diastolic blood pressure compared with the sham group. The authors attributed these outcomes to the enhancement of gamma-aminobutyric acid (GABA) levels, the involvement of which was demonstrated in studies involving cTBS over the primary motor cortex [66]. The mechanism behind this phenomenon suggests that cTBS may increase the inhibitory activities of interneurons, resulting in higher concentrations of GABA. Zack and colleagues explained the reduction in psychostimulant-like and arousing effects of the task, as well as the increase in the Stroop interference test, by proposing that stimulation of the prefrontal GABA neurons may impair the ability to shift attention away from a target stimulus [67].

Concerning cTBS, it has shown the ability to modulate cognitive control and motor inhibition in healthy participants by targeting the pre-SMA [68]. This modulation can be achieved either through the hyperactivation of pathways to the subthalamic nuclei or through its direct connections with the striatum. Building upon these promising findings, Salerno et al. [35] designed their study using cTBS as a “proof-of-concept” trial. Although cTBS over the pre-SMA was not originally part of the protocol, previous evidence highlighted it as a crucial region to target for ongoing response inhibition and conflict resolution, eventually resulting in a decrease in risky decisions and an enhancement in inhibitory control [69]. The study showed a progressive reduction in severity, measured by PG-YBOCS over the follow-up period, and an improvement in the CGI scores. Additionally, the study confirmed that cTBS was a safe treatment option, without any reported side effects.

The articles included in our systematic review have various limitations. Most of the included studies did not employ validated questionnaires to evaluate the sensation of craving, but different VASs, thus making it difficult to compare all these results. Additionally, not all articles mentioned the term ‘craving’, but rather defined symptoms such as ‘urge to gamble’ or ‘desire to gamble’. Craving, alongside the compromised capacity to manage impulses, stems from gradual alterations in synapses and circuits, brought about by prolonged exposure to addictive substances [24]. Interestingly, all these neural alterations may be targeted by TMS, although, to date, it remains challenging to implement. It would be advantageous to establish a common “craving network” that is consistent among individuals, regardless of whether they have substance use-related disorders. This would enable the identification and tracking of a neural pattern associated with craving or the drive for motivated behaviors [70]. Furthermore, the included studies had short follow-up periods, often limited to the time of the procedure, which resulted in an inadequate evaluation of the long-term effects.

Our systematic review also had some intrinsic limitations. The inclusion of diverse clinical populations, as well as the inclusion of case reports and case series, the high risk of bias in multiple domains of the included studies, and the concurrent use of other treatments introduce limitations and cofounding factors regarding treatment efficacy. In particular, the placebo effect may have influenced many of the results obtained, and randomized sham control trials are therefore needed in future.