Eligibility criteria

We will include randomized controlled trials involving participants diagnosed with AUD as defined in the Diagnostic and statistical manual of mental disorders (DSM), 5^th edition, alcohol dependence or abuse as defined in previous DSM editions, or alcohol dependence syndrome or abuse as defined by International Classification of Diseases [24,25]. All other types of trials will be excluded to avoid the risk of bias associated with non-randomized or non-controlled studies. To ensure the validity of the transitivity assumption, we will exclude studies that exclusively recruit participants with comorbid conditions, such as depression or post-traumatic stress disorder (PTSD). Including trials focused solely on comorbid populations could compromise the transitivity assumption by altering the distribution of disease severity across trials [21].

We will include trials comparing an active NIBS paradigm to a sham intervention, either as a stand-alone treatment or as an add-on therapy. This distinction will be accounted for when defining the nodes of our NMA. All types of NIBS paradigms will be considered, regardless of stimulation parameters such as cortical target, intensity, frequency, number of sessions, or number of pulses delivered. These broad inclusion criteria were defined in the aim of gathering all available evidence to construct a comprehensive network of interventions and determine their comparative effectiveness.

The comparator will consist of sham NIBS interventions. In cases involving combined treatment regimens, sham interventions paired with an active co-intervention will also be included, with this accounted for by splitting nodes in the NMA.

We will include studies in any language, from all geographical regions, with no restrictions on the year of publication. All types of randomized controlled trial reports will be considered, including journal articles, conference abstracts, theses, and book chapters. If necessary, we will contact authors to obtain supplementary information.

Outcomes

The primary interest outcomes of our review will concern alcohol consumption in AUD populations. In trials investigating NIBS as a strategy for alcohol use reduction, we will explore NIBS effects on the reduction in total alcohol consumption and the number of heavy drinking days among participants, assessed at immediate post-treatment and, if applicable, including the follow-up period. As follow-up durations may vary across trials, we will define short-term follow-up as less than 1 month, mid-term as 2–3 months, and long-term as more than 3 months after the end of treatment. Separate quantitative syntheses will be conducted for each follow-up period. These alcohol reduction outcomes are supported by international recommendations as clinically relevant [26,27]. Total alcohol consumption is defined as mean daily alcohol consumption in grams of ethanol per day, and heavy drinking day is defined as a day with more than 60 grams of ethanol consumed in men and more than 40 grams in women. In alcohol reduction trials, the mean difference and its standard deviation will be used as the measure of effect. In trials in recently detoxified AUD patients where NIBS potential to prevent relapse is explored, the primary outcome will concern the rate of relapse, and the risk ratio will be used as the measure of effect. Special attention will be given to the definition of relapse used by the authors, as relapse is often heterogeneously defined across trials [28]. The preferred working definition of relapse that we will use in our NMA is a return to the previous pattern of alcohol consumption. If a different definition is used in a given study, we will describe it in detail and contact the authors in an effort to harmonize relapse definitions for inclusion in the NMA. Relapse is expected to be defined according to one of the following criteria: return to the previous pattern of consumption; return to any alcohol consumption; percentage of heavy drinking days, as predefined by the study authors; and problems related to alcohol use. If relapse is defined as a return to any alcohol consumption in the majority of trials, we will contact the authors of studies using different definitions to obtain data for all individuals who resumed any alcohol consumption. Conversely, if only a minority of trials define relapse as any alcohol consumption, we will contact the authors of these studies to harmonize the definitions. If harmonization is not possible, the respective trials will be excluded from the quantitative synthesis.

Our analysis will include all available data, and in cases of missing data, we will contact the study authors for clarification.

Data on alcohol craving will also be investigated as part of secondary interest outcomes. To provide a comprehensive assessment of the interventions, we will also examine data on adherence, adverse effects, and acceptability. Definitions of these outcomes will be extracted from each study, and quantitative data will be summarized when appropriate to better characterize the efficacy and tolerance profiles of the interventions studied.

Information sources

The following databases will be searched: EMBASE, PubMed, PsycINFO, and The Cochrane Library. To ensure a comprehensive search strategy, we will also examine trial registries (ClinicalTrials.gov, clinicaltrialsregister.eu, WHO ICTRP), conference proceedings, and theses to identify unpublished studies. Additionally, we will screen the reference lists of included trials to identify any potentially eligible studies. All searches will be updated and re-run before the final analysis to ensure the inclusion of the most recent evidence.

Search strategy

We began our scoping searches in November 2024. Our literature search strategy utilized medical subject headings (MeSH) or their equivalents, as well as relevant text words related to NIBS. A draft Medline search strategy was developed by a member of the research team experienced in systematic review searches (AD). This strategy was pilot-tested and peer-reviewed by other team members to ensure accuracy and comprehensiveness. Cochrane filters for randomized controlled trials were applied where appropriate, but no additional filters were used. The complete search strategies for all databases are provided in the Supplemental file (S1 File).

Data management and study selection

After applying our comprehensive search strategy, all entries were retrieved and uploaded into the systematic review management tool, Rayyan QCRI [29]. The study eligibility form was created by AD and peer-reviewed by all research team members. It is available as a Supplemental file (S1 Fig). Two authors independently assessed all titles and abstracts using predefined selection criteria. Any disagreements were resolved through discussion with a third author (BP).

At the time of protocol submission, the second round of selection, based on the full texts of the previously retrieved entries, had not yet begun. The full texts will be reassessed independently by two authors after submission. Any disagreements will be resolved through discussion with a third author (BP). We will provide an explanation for all entries excluded at this stage. Additionally, two raters (AD and BT) will independently assess whether the entries should be included in the meta-analysis, with conflicts resolved through discussion with the entire research team.

Data collection process

Data extraction will be conducted independently by two authors using a data extraction form created by AD in consultation with all team members (Supplemental file, S2 File). This form will be pilot-tested on three randomly selected trials, and an adapted version will be applied to all selected trials. Any disagreements will be resolved through discussion with a third author. Patient characteristics (age, gender, disease severity) and study details (number of arms, blinding, randomization process, descriptions of interventions and controls) will be systematically extracted and presented in a tabulated form to facilitate evaluation of the transitivity assumption. Measures of effect will be extracted in the form of means, standard deviations, and sample sizes for alcohol reduction trials, and as number of participants who relapsed after treatment as well as a total number of participants for relapse prevention trials. Information on adverse events will also be extracted for safety evaluation.

In cases of missing data or the need for additional details, study authors will be contacted by institutional email, with up to three email attempts made over the course of one month. Data will be recorded using an Excel spreadsheet.

For studies with multiple reports, two authors (AD and BT) will independently compare the reports using author names, intervention characteristics, study locations, study dates and duration, and baseline sample sizes, as recommended by the Cochrane Handbook [30]. Only one report from duplicate trials will be included to avoid biases from including the same trial multiple times.

For crossover studies, only pre-crossover data will be used.

Risk of bias in the individual studies

Two authors (AD and BT), both experienced in risk of bias evaluation, will independently assess each trial at outcome level using the revised Cochrane risk-of-bias tool for randomized trials [31]. They will evaluate the following criteria: bias arising from the randomization process, deviations from intended interventions, incomplete outcome data management, measurement of the outcome, and selection of the reported result. Each domain will be categorized as “high risk,” “low risk,” or “some concerns” to determine the overall risk. Signaling questions will be used to guide the authors in making their judgments. If their independent evaluations do not align, consensus will be reached through discussion with a third author (BP).

The risk of bias evaluation will be visually represented next to the forest plots for each outcome, as recommended in the Cochrane Handbook [30]. This evaluation is a crucial component in assessing the internal validity of the included studies. It will provide essential information to critically appraise the review’s results regarding the true effect of the interventions. Importantly, this process enhances the credibility and transparency of the review by comparing individual studies to their registered protocols. We will discuss the results of the risk of bias evaluation and describe the implications for the validity of the review’s overall findings.

Data synthesis

We will quantitatively synthesize the extracted results using mean differences as the effect measure for alcohol use reduction trials and risk ratios for relapse prevention trials. First, a random-effects pairwise meta-analysis will be performed to compare treatment and control arms across different trials. The inverse variance method and random effects model will be used to account for clinical and methodological differences between studies. Sensitivity analyses will be conducted by comparing the random effects model to the fixed effects model to evaluate small sample bias. To estimate the effect size in alcohol use reduction trials, we will use Cohen’s d interpreted as small (d = 0.2), medium (d = 0.5), and large (d = 0.8) [32].

For the NMA, we will use graph-theoretical methods in R with the netmeta package [33]. This approach makes it possible to implement trials with more than two intervention arms. The netmeta package adjusts for the correlation of pairwise comparisons in multi-arm trials by appropriately reducing their weights.

Alcohol reduction trials and relapse prevention trials will be synthesized separately.

Network nodes will represent the technical characteristics of the interventions (tDCS, rTMS, or ultrasound neuromodulation). For tDCS and rTMS, we will further split the nodes by target characteristic (lateral prefrontal, medial prefrontal, other). For rTMS, the nodes will also be based on the pulse frequency (low or high). Sham interventions will serve as the reference group. We will be mindful of the need to split the sham node if the sham interventions are too heterogenous in the included trials, and we will thoroughly justify our choice of reference group.

Network geometry will be described alongside the network graph, analyzing the structure of the treatment comparisons. We will describe the interventions, their connectedness, the number of studies informing each comparison, and the sample sizes for each intervention node. Additionally, we will detail the direct comparisons and highlight the most well-supported ones.

Data on potential effect modifiers (e.g., sample characteristics and study methods) from each individual study will be extensively described to allow for an independent evaluation of the transitivity assumption.

The results of the NMA will be presented as a network graph showing treatment nodes and direct comparisons, a league table with both direct and NMA (indirect or mixed) estimates, and a net heat plot to assess inconsistency.

A sensitivity analysis will be conducted by merging the nodes in our meta-analysis based on the technical definition of each technique (e.g., all tDCS participants merged, all rTMS participants merged, etc.).

Between-study heterogeneity will first be explored using the clinical and methodological characteristics of the trials. For the statistical evaluation, we will use 𝜏² as a measure of heterogeneity magnitude. This will be expressed as a prediction interval, which will be compared with confidence intervals for each comparison to assess whether there are no concerns, some concerns, or major concerns related to heterogeneity [34]. To further explore heterogeneity, sensitivity analyses will be conducted by excluding outlier studies, studies with a high risk of bias, or studies with significant clinical or methodological differences.

Meta-biases

Selective reporting biases will be assessed by comparing the pre-published protocols of the included trials with their published results. If the protocol is unavailable, we will compare the Methods and Results sections of the published trials. To explore publication bias, we will use small study effects analysis as a proxy, analyzing the comparison-adjusted funnel plot and performing Egger’s test. The results of each evaluation will be described, interpreted, and critically discussed in the text of our review.

Overall confidence in the evidence

To evaluate the confidence in our results, we will use the CINeMA framework, which is based on the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach, with specific adaptations for NMA methodology [34]. CINeMA integrates the assessment of within-study bias by incorporating the conclusions from the risk-of-bias assessments of individual trials into its framework. It also integrates reporting bias, imprecision, indirectness, incoherence, and heterogeneity [35]. Contribution matrices will be presented and interpreted to analyze how each individual study informs the NMA results.