This systematic review followed the methodological standards of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and the Synthesis Without Meta-analysis (SWiM) guidelines (Page et al., 2021; Campbell et al., 2020). The protocol was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO CRD420251091407).

Studies were eligible if they included adult participants with neuropsychiatric disorders. Studies that examined healthy individuals or participants without such disorders were excluded. The intervention of interest was taVNS, irrespective of specific stimulation site or parameters. Studies using transcutaneous cervical vagus nerve stimulation were excluded. Studies with a sham-taVNS control condition were included, as well as non-controlled studies assessing cognitive functions before and after active taVNS treatment. The primary outcome was cognition operationalized by global cognitive functioning and domain-specific cognitive functions.

The review methodology was defined prior to data collection and adhered to PRISMA and SWiM recommendations. To ensure methodological quality and transparency, the review was evaluated using the AMSTAR-2 instrument (Shea et al., 2017). Of the 16 AMSTAR-2 items, 13 were applicable to this review, as no quantitative meta-analysis was conducted. Responses were rated as yes, partly, or no, and summarized in a traffic-light plot.

The systematic search was conducted on September 1, 2025. The search covered the period from database inception to September 1, 2025. Only articles published in English or German were eligible for inclusion. The following search string was applied:

(“cogniti*” OR “attenti*” OR “memor*” OR “language” OR “verbal” OR “visuo-spatial” OR “executi*” OR “dementia”) AND (“transcutaneous vagal nerve stimulation” OR “transcutaneous auricular vagal nerve stimulation” OR “transcutaneous vagus nerve stimulation” OR “transcutaneous auricular vagus nerve stimulation” OR “tVNS” OR “taVNS” OR “tcVNS” OR “t-VNS” OR “ta-VNS” OR “tc-VNS”) AND (“human” OR “humans”).

Electronic searches were conducted in MEDLINE/PubMed, EMBASE, and the Cochrane Central Register of Controlled Trials (CENTRAL). CENTRAL also includes studies indexed in clinicaltrials.gov. These databases were selected according to Cochrane Collaboration recommendations for evidence synthesis in medical research (Higgins et al., 2016). In addition, a manual search was carried out following Cochrane guidance (Lefebvre et al., 2020). References from all included papers were screened to identify further relevant studies not captured in the database search. Reference lists of newly identified publications were screened recursively.

Study selection and data extraction were performed using the commercial software rayyan.ai (Ouzzani et al., 2016). The systematic search, study selection, and risk of bias assessments were independently conducted by two reviewers (H. D. and F. H.). Discrepancies regarding study eligibility or extracted data were resolved by consensus discussion between the two reviewers. The selection process was visualized in a PRISMA flow diagram (see Figure 1), comprising the stages of identification, screening, eligibility assessment, and inclusion (Ziegler et al., 2011). Database and manual search results were tracked separately.

For each included study, PICO parameters (Population, Intervention, Comparator, and Outcome) were extracted. Additional extracted variables included stimulation parameters (site, frequency, pulse width, intensity, daily stimulation time, total duration) in line with the international consensus recommendations for reporting taVNS research (Farmer et al., 2020).

Cognitive outcome variables were categorized according to the cognitive domains defined by the Movement Disorder Society (Litvan et al., 2012). These domains comprise attention and working memory, executive function, language, memory, and visuospatial function. In addition, the domain of social cognition was considered (Simpson, 2014). For each domain, it was recorded whether taVNS led to significant improvement.

In line with SWiM recommendations for systematic reviews, because of the heterogeneity of study designs and stimulation protocols, we summarized quantitative data using mode and range statistics rather than mean-based estimates. For measures of global cognition, Cohen’s d was calculated as effect size (Morris, 2008). Risk of bias was evaluated using the Cochrane Risk-of-Bias tool for randomized trials, which evaluates potential bias across five domains: randomization process, deviations from intended interventions, missing data, measurement of the outcome, and selection of the reported result (Savovic et al., 2014). For non-randomized studies, the ROBINS-I tool was applied, which assesses bias due to confounding, participant selection, intervention classification, deviations from intended interventions, missing data, outcome measurement, and selective reporting (Sterne et al., 2016). Results were presented in a traffic-light plot (Nejadghaderi et al., 2024). Financial or non-financial conflicts of interest of study authors were also extracted.

Figure 1 illustrates the flow diagram of the literature search. There were no deviations from the protocol described in the Methods section. The systematic database and registry searches yielded the following results.

The PICO parameters (population, intervention, comparator, outcomes) of the 15 included studies are summarized in Table 1. Table 2 presents similarities and differences in stimulation parameters and treatment across studies. A synthesis of the cognitive outcomes associated with taVNS is summarized in Table 3.

Supplementary Table 1 summarizes adverse events across the included studies. Nine of the fifteen studies (60%) explicitly reported (serious) adverse events monitoring or occurrence, while six (40%) did not provide safety information. Reported adverse events were mild and transient, and no serious adverse events or adverse events with sequelae occurred in any study.

Using mode statistics across Tables 1–3, commonalities among the 15 included studies can be summarized as follows:

Between-study differences are quantifiable by the reported ranges: 1–270 days of stimulation, 6–240 min/day, and wide parameter spans (pulse width 0.25–1.00 ms; frequency 20–120 Hz; intensity 0.6–13.6 mA or 30–50 V). Target regions on the ear also varied, with stimulation most often at the left ear, typically the tragus or cymba conchae.

Across the 15 included studies, 8 reported statistically significant improvements in global cognition or at least one specific cognitive domain following taVNS, while 7 reported no significant cognitive benefit. Improvements were most frequently observed in attention and working memory (reported in 5 studies), followed by memory (4 studies), executive functions (3 studies), language (1 study), social cognition (1 study), and global cognition (1 study). For each cognitive domain, at least one study also reported no effect, underscoring the heterogeneity of findings across populations, outcome measures, and study designs.

For global cognition, three studies used the MoCA (Trevizol et al., 2016; Wang et al., 2022b; Yang H. et al., 2023). Cohen’s d was small in Trevizol et al. (2016), moderate in Wang et al. (2022b), and not computable for Yang H. et al. (2023), because results were presented only graphically without clear standard deviations.

Risk of bias was high in the majority of included studies (see Table 3). Only one study showed low risk of bias (Lench et al., 2023), and four studies were rated as having some concerns (Corrêa et al., 2022; Mertens et al., 2022; von Wrede et al., 2021; Yang H. et al., 2023).

The stimulation parameters observed here align with those summarized by Ridgewell et al. (2021) in a review of 19 studies in healthy adults, where 0.25 ms pulse width and 25 Hz frequency were also most frequent. In neuropsychiatric populations, the present review found attention to be the most commonly studied domain; by contrast, in healthy adults, executive functions were investigated most frequently, followed by attention (Ridgewell et al., 2021). Ridgewell et al. further noted a wide dispersion of outcome measures across studies in healthy adults, with no overlap of identical outcomes between studies. Their meta-analysis suggested that taVNS improves executive functions in healthy adults, while effects on attention and memory were not observed (Ridgewell et al., 2021). A subgroup analysis across nine publications indicated that left tragus stimulation yielded greater efficacy than left cymba conchae (g = 0.48; Ridgewell et al., 2021). They also reported a small but significant effect on global cognitive performance (g ≈ 0.21). The influence of baseline global and executive performance was not reported in that review.

Table 4 summarizes the methodological quality assessment of the present review using the AMSTAR-2 instrument. The 16 AMSTAR-2 items served as a structured checklist to systematically verify the methodological rigor and transparency of the review process. Each item was linked to the corresponding section in the manuscript. As the review did not include a quantitative meta-analysis, items 11, 12, and 15 were not applicable. Most domains (e.g., PICO definition, search strategy, study selection and data extraction, and risk-of-bias assessment) were rated positively, indicating good methodological transparency. The results are visualized in a traffic-light plot.

Study populations differed: 7 studies included patients with epilepsy, 3 with COVID-19/long-COVID, 2 with Parkinson’s disease, 2 with major depression, and 1 with MCI. The intervention in all included studies was taVNS. Of the included studies, two-thirds were sham-controlled, and one-third were uncontrolled. The largest study was randomized, controlled, and double-blind, with 76 patients with epilepsy in the active arm and 36 in the control arm (Yang H. et al., 2023). All other studies treated fewer than 30 patients with taVNS.

Interpretation is limited by heterogeneity of outcomes and scales. The most frequently assessed measure, the MoCA, was collected in only three studies; of these, only one study in MCI reported improvement with taVNS, whereas the other two showed no change. The study by Yang H. et al. (2023) with 76 active and 36 sham-treated epilepsy patients found no improvement in global cognition. Comparing MCI and epilepsy cohorts, baseline differences were evident in MoCA scores and age (Wang et al., 2022b, Yang H. et al., 2023). In MCI, the pre-treatment mean MoCA was approximately 19.5 points (graphically reported only), versus 20.6 in epilepsy; age differed more substantially (66.9 years in MCI versus 33.3 in epilepsy). Such clinical and demographic differences may confound and contribute to outcome heterogeneity.

A quantitative heterogeneity analysis is not informative given the small number of studies. Because cognitive domains were tested with different instruments, domain-specific comparisons are also restricted. A potential approach would convert study outcomes to norm-referenced metrics per test and then compare across cohorts; this could be pursued in future work, acknowledging possible trade-offs in interpretability and clinical utility.

Author honoraria indicating financial conflicts of interest were reported in 2 of 15 included publications (≈13%), which is not negligible. Nevertheless, with >85% investigator-initiated research, the field is not dominated by financial interests. To advance the field, public funding (e.g., from the German Federal Ministry of Education and Research, the German Research Foundation, the European Union’s Horizon Europe program, or the US National Institutes of Health) would be desirable but has not yet been provided for taVNS. Industry partnerships could also be beneficial but require financially strong partners and independent scientific steering groups. Major device manufacturers (Boston Scientific, Medtronic, Abbott) have not entered the taVNS market to date; in Germany and Europe, the market is dominated by tVNS Technologies, whose taVNS devices are CE-certified. With less than 10 million € in annual revenue and less than 20 employees (Duphorn, 2024), tVNS Technologies cannot fund large multicenter RCTs comparable to those supported by global leaders in deep brain stimulation (Deuschl et al., 2006; Schuepbach et al., 2013).

Potential mechanisms underlying the observed cognitive effects are discussed below. Building on the theoretical background, we distinguish direct neurophysiological mechanisms closely linked to neuroanatomy/physiology from general mechanisms with system-level or medication-related origins.

Across studies, 250 μs pulse width and 25 Hz frequency were most common, with stimulation intensity set just below pain threshold. fMRI data indicate that taVNS with these parameters suffices to modulate brainstem activity and distant but functionally connected regions, e.g., dorsolateral prefrontal cortex (Dietrich et al., 2008). Several technical issues warrant further research. The optimal stimulation site remains unsettled, and the review revealed substantial heterogeneity in outcomes, parameters, and sites. Future studies should examine how site and settings relate to cognitive effects. taVNS is commonly applied to the left ear—historically inherited from cervical VNS, where right-sided stimulation risks cardiac arrhythmias via efferent vagal fibers (Nemeroff et al., 2006). For taVNS, only afferent fibers projecting to the brainstem are stimulated (Kreuzer et al., 2012), and arrhythmias are not expected with right-sided stimulation; indeed, Trevizol et al. (2016) reported no arrhythmias even with bilateral taVNS.

Among the three studies assessing taVNS effects on global cognition (Trevizol et al., 2016; Wang et al., 2022b; Yang H. et al., 2023), only the MCI cohort showed a significant, moderate effect (Wang et al., 2022b). By contrast, a meta-analysis of cognitive training in MCI reported large effects with three studies and 199 participants and low heterogeneity (Mamayson and Lacanaria, 2024).

This review combined systematic database/registry searches with a manual search, including recursive reference screening and use of a personal reference database as described in Methods. A limitation arises from the nomenclature variability of taVNS. We searched multiple formulations (e.g., “transcutaneous vagus/vagal nerve stimulation,” with and without “auricular”), but numerous further variants are conceivable (e.g., “transauricular…,” “auricular non-invasive…”). To mitigate this, three strategies were applied: (1) acronym use (with/without hyphenation to “VNS”), (2) guideline-recommended terminology, and (3) a complementary manual search.

The AMSTAR-2 served as a practical framework for quality control across introduction, methods, results, and discussion. Moreover, the interpretation of the present findings is limited by the methodological quality of the available studies. The majority of included studies were characterized by small sample sizes, heterogeneous designs, and a high or unclear risk of bias. Only one study was rated as having low risk of bias, while most showed either some concerns or high risk across key domains. Such limitations substantially reduce statistical power, increase susceptibility to type I and type II errors, and limit the generalizability of reported effects. Consequently, the current evidence does not allow definitive conclusions regarding the clinical effectiveness of taVNS for cognitive deficits.

Further, insufficient studies were available to conduct a quantitative synthesis of findings by means of meta-analysis. Meta-regression on dose–response effects would be highly informative for the field, informing stimulation parameters and clinical dosing protocols.

This review indicates that taVNS can improve global cognition and specific domains (attention, memory, executive functions, language, and social cognition) in disorders such as major depression, epilepsy, Parkinson’s disease, MCI, and COVID-19. The range of clinical efficacy appears broad and may reflect demographic/clinical differences (age, disease duration, ethnicity, baseline cognitive severity). To increase impact and clarity, future research directions can be prioritized and grouped into four thematic areas: (i) patient stratification and biomarkers, (ii) harmonization of outcomes and study design, (iii) mechanistic and translational research, and (iv) long-term efficacy, technical optimization, and multimodal interventions.

For neurodegenerative or partially degenerative disorders, trials should extend observation windows to at least 1–2 years to capture medium- to long-term efficacy on cognition. Studies with durations of 5 years or longer are needed to evaluate sustained effectiveness, durability of response after parameter adjustments, and long-term safety, including adherence and device tolerability issues.

Technical advances are warranted to improve ear-conforming electrodes that deliver current more efficiently and reproducibly across auricular targets. Parameter optimization—particularly pulse width, frequency duty cycles, and daily dose—should be pursued alongside secure remote programming capabilities to enable supervised, telemedicine-based titration and monitoring.

Finally, taVNS should be evaluated within multimodal treatment strategies that pair neuromodulation with pharmacological and non-pharmacological interventions known to improve cognition, such as acetylcholinesterase inhibitors, structured cognitive training, or light therapy (Alves et al., 2013; Huang et al., 2024). Synergistic designs may enhance effect sizes and broaden clinical applicability across disease stages.