Sorting the Mind: Cognitive Functions in taVNS

An increasing number of studies are now anchoring their focus on specific aspects of cognitive functions, aiming to understand the role of taVNS within these domains [20]. However, cognitive functions are not an isolated puzzle piece; it is a vast and intricate Hogwarts LEGO castle, woven from countless brain regions, neural circuits, and cognitive processes. The slightest alteration anywhere within this network can trigger systemic effects [21]. To comprehend how taVNS operates within this framework, we cannot confine ourselves to a single cognitive dimension. Much like dividing Hogwarts into four houses during our LEGO construction, this study segments cognitive functions into four major modules (Fig 3).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Cognitive Functions and taVNS: A Four-Domain Framework.

https://doi.org/10.1371/journal.pone.0347849.g003

This four-domain taxonomy is theoretically grounded in converging evidence from cognitive neuroscience and is designed to match the specific neural targets of taVNS. (1) Executive Functions are anchored in the prefrontal cortex and its top-down control networks [22]; (2) Working Memory and Attention are jointly governed by prefrontal–parietal circuitry and the locus coeruleus–norepinephrine system [21], both of which are direct downstream targets of vagal afferents; (3) Social Cognition and Emotional Regulation share overlapping circuitry involving the amygdala, insula, and anterior cingulate cortex [23], which are regions that are selectively engaged by parasympathetic modulation; and (4) Cognitive Flexibility and Learning recruit frontostriatal loops and hippocampal plasticity mechanisms ([24];[25]), which have been shown to be sensitive to noradrenergic and cholinergic tone. The potential conceptual overlap between domains is addressed at the operational level: each cognitive measure will be assigned to a domain according to a pre-specified classification guide (available in Appendix II) that operationally distinguishes the domains by the primary cognitive process being assessed, rather than the neural substrate alone. Specifically, tasks requiring the active online updating, maintenance, or manipulation of information (e.g., N-back, Digit Span) are classified under Working Memory and Attention; tasks requiring the shifting of mental set in response to feedback or rule changes (e.g., Wisconsin Card Sorting Test, Intra-/Extra-Dimensional Shift) are classified under Cognitive Flexibility and Learning; tasks requiring the generation, monitoring, or inhibition of goal-directed actions under conflict (e.g., Stroop, Stop-Signal) are classified under Executive Functions; and tasks assessing the recognition, interpretation, or regulation of emotional or social information (e.g., Facial Emotion Recognition, Emotion Regulation Questionnaire) are classified under Social Cognition and Emotional Regulation. Any measure that plausibly spans two domains (e.g., the Trail Making Test, which involves both set-shifting and executive control) will be assigned by consensus between two raters using the primary cognitive demand criterion, and the assignment will be recorded for sensitivity analyses. This approach minimises classification ambiguity and ensures that boundary cases are handled consistently and transparently.

Executive functions

This module governs advanced cognitive activities such as decision-making, planning, problem-solving, and impulse control. Executive functions are not only important for individuals navigating complex environments but also represent the core capacity for flexible adaptation and adjustment to change [22]. They enable individuals to demonstrate qualities of “resolve” and ‘initiative’ when confronting intricate decisions. Within taVNS research, how these core cognitive functions are influenced—particularly self-regulation under high-pressure scenarios— emerges as a critical question. The present synthesis will quantify whether executive function outcomes improve following taVNS; however, we acknowledge that even a positive pooled effect would not resolve questions about the precise neural mechanisms (e.g., whether prefrontal modulation of impulse control is specifically responsible), and conclusions about mechanism will accordingly be drawn with appropriate caution.

Working memory and attention

Working memory represents the mental workspace that allows individuals to temporarily store, update, and manipulate information essential for reasoning, comprehension, and goal-directed behaviour [26]. It is not merely a passive holding system but a dynamic process that integrates incoming sensory input with existing knowledge to facilitate complex cognitive operations. Attention, on the other hand, functions as the selective gatekeeper of cognition—allocating limited neural resources toward task-relevant stimuli while inhibiting distractions [27]. From a neurocognitive perspective, these two mechanisms are deeply interwoven. The central executive component of working memory relies heavily on attentional control systems mediated by the prefrontal cortex and parietal networks ([21];[28]). This interaction enables individuals to maintain focus under conditions of high cognitive load and to flexibly shift attention when task demands change [29].

Social cognition and emotional regulation

Social cognition encompasses a constellation of mental processes that enable individuals to perceive, interpret, and predict others’ emotions, intentions, and social behaviours [30]. It forms the cognitive foundation for empathy, moral reasoning, and adaptive interpersonal functioning [30]. Emotional regulation, by contrast, refers to the capacity to monitor, evaluate, and modify one’s own emotional reactions to maintain psychological homeostasis and achieve social harmony [31]. Together, these systems constitute the core of affective intelligence—an essential element of prosocial behaviour and interpersonal sensitivity.

From a neurobiological perspective, both social cognition and emotional regulation are supported by overlapping neural circuits involving the prefrontal cortex, amygdala, insula, and anterior cingulate cortex [23]. These regions mediate the intricate balance between emotional reactivity and cognitive control [32]. taVNS, by modulating vagal afferent pathways and enhancing parasympathetic tone, may exert a dual influence: enhancing emotion recognition accuracy and improving self-regulatory capacity. The presumed direction of enhancement (rather than impairment) is grounded in evidence that increased parasympathetic tone, which taVNS reliably produces, is associated with improved vagally-mediated cardiac regulation, reduced amygdala reactivity, and better prefrontal inhibitory control over limbic responses [31]. These physiological shifts are mechanistically linked to more accurate affective processing rather than disruption of it. We nonetheless treat this as a directional hypothesis to be tested; the meta-analysis will examine effects in both directions and will report the full distribution of outcomes rather than assuming uniform enhancement. Such modulation could facilitate more adaptive social responses, attenuate maladaptive emotional arousal, and enhance empathy-driven interactions.

Cognitive flexibility and learning

Cognitive flexibility constitutes the executive faculty that enables individuals to modify their cognitive sets, shift attention, and adapt behavioural strategies when confronted with new or conflicting task demands [24]. It reflects an individual’s ability to disengage from habitual responses and adopt novel, contextually appropriate solutions—an essential mechanism for strategic reasoning and adaptive problem-solving. Complementing this, learning ability represents the progressive consolidation of knowledge and experience through iterative feedback and neural plasticity, ensuring that past encounters inform future decisions ([25]). Together, these capacities embody the essence of intellectual adaptability —the wisdom to evolve one’s strategy in pursuit of long-term goals. taVNS, by engaging vagal afferents and influencing the locus coeruleus–norepinephrine system, may enhance the neural mechanisms underlying cognitive exploration and adaptive learning. Such stimulation promotes heightened alertness, efficient error monitoring, and greater behavioural plasticity—key components of strategic cognition.

Current evidence landscape

The burgeoning interest in taVNS, representing a rapidly evolving field within cognitive enhancement research. However, the existing research landscape remains fragmented, with disparate findings across domains. Previous meta-analyses on taVNS only investigated its effects in other domains, such as insomnia [33], pupil dilation [34], heart rate variability (HRV) [35] or its overall safety [36]. Only one previous meta-analysis has specifically examined its efficacy on cognitive functions, and that study focused on healthy individuals. Ridgewell et al. [20] reported a small but significant weighted effect size of g = 0.21 for overall cognitive performance following taVNS, suggesting a modest enhancement of cognition. Critically, that meta-analysis was restricted to healthy participants, and a growing body of clinical evidence published since 2021 was not captured. The specific limitations of the Ridgewell et al. [20] analysis, and the rationale for the present synthesis, are detailed in the following section.

As detailed above, the existing evidence base for taVNS effects on cognition remains limited to one study [20] that excluded clinical populations. This synthesis, while valuable, has several important limitations that motivate the present analysis. First, it exclusively enrolled healthy participants (65 clinical-sample effect sizes were excluded), leaving open whether the g = 0.21 effect generalises to patient populations in whom taVNS is most commonly used. There is no a priori reason to assume that effects observed in healthy controls transfer to individuals with, for example, Alzheimer’s disease or ADHD, whose underlying neurobiology and baseline cognitive capacity differ substantially. Second, at least one study included in that meta-analysis has since been retracted owing to problematic data reporting; the retracted effect size was among the larger estimates in the pool (d > 0.60), and its removal from the original analysis would meaningfully reduce both the pooled estimate and its precision. Third, the five years since publication have produced a substantial new body of clinical evidence across multiple cognitive domains that was unavailable to Ridgewell et al. Based on their 19 eligible effect sizes and the subsequent publication rate in this field, we estimate that the present synthesis will incorporate approximately 30 independent effect sizes, nearly doubling the evidence base and affording substantially greater power to detect domain-specific and moderator effects. For these reasons, a new, updated, and clinically focused meta-analytic synthesis is warranted.

Protocol and registration

This systematic review and meta-analysis will follow the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Protocol (PRISMA-P) guidelines and is prospectively registered in the PROSPERO (ID: CRD 420251166774). The designation “systematic review and meta-analysis” reflects a deliberate methodological decision. The systematic review component serves functions that the quantitative synthesis alone cannot fulfil: (1) it will provide a structured narrative synthesis of studies that cannot be pooled quantitatively due to insufficient statistical reporting or unique outcome operationalisations; (2) it will document the stimulation parameter landscape, identifying gaps and inconsistencies in protocol reporting that are clinically important even when effect sizes are not extractable; (3) it will map study quality and methodological trends across the literature, informing both interpretation of the pooled estimates and recommendations for future research design. Case studies are excluded because they cannot contribute effect sizes and do not include control conditions; however, the systematic review component will note prominent single-subject findings where they inform understanding of inter-individual variability. This distinction between the qualitative and quantitative contributions of the review is now maintained throughout the manuscript.

Search strategy

A comprehensive and systematic search strategy has been developed in consultation with a specialist librarian to ensure broad coverage of biomedical, psychological, and interdisciplinary literature. The full search strings for each database, including all Boolean operators and controlled vocabulary terms (e.g., MeSH headings for PubMed), are provided in Appendix I. Briefly, the core search terms include: (“vagus nerve stimulation” OR “transcutaneous auricular vagus” OR “taVNS” OR “auricular vagal”) AND (“cognit*” OR “executive function” OR “attention” OR “memory” OR “learning” OR “emotion regulation” OR “social cognition” OR “cognitive flexibility”). We will search six electronic databases from their inception up to the date of final search execution (no end-date restriction will be applied in the search syntax; the search will be run immediately prior to PRISMA flow completion and the actual date will be reported in the final manuscript), including PubMed, EMBASE, PsycINFO, Web of Science Core Collection, Cochrane Central Register of Controlled Trials (CENTRAL), Scopus. To minimise publication bias and identify unpublished or ongoing studies, we will also systematically search trial registries, including ClinicalTrials.gov and the World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP).

In addition to database and registry searches, we will conduct extensive hand-searching procedures. The reference lists of all eligible full-text articles, as well as those of relevant systematic reviews and meta-analyses, will be examined to identify additional studies that may not have been captured through electronic searches. Grey literature will also be considered, including doctoral dissertations, and theses indexed in ProQuest Dissertations & Theses Global and WorldCat. Reports from scientific meetings and other unpublished materials will be examined where sufficient methodological detail and outcome data were available. All searches will be conducted without restrictions on languages or publication year.

Study selection and data extraction

Inclusion and exclusion criteria are listed in Table 1 and data extracted from articles can be found in Table 2.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Inclusion and exclusion criteria for study selection.

https://doi.org/10.1371/journal.pone.0347849.t001

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Study characteristics and data extraction framework.

https://doi.org/10.1371/journal.pone.0347849.t002

Effect size imputation and calculation

Missing data represents an inevitable challenge in meta-analysis, and we will address it systematically through a hierarchical approach. Our first step will be to contact corresponding authors of studies with incomplete reporting, allowing a two-week response window for data provision. When author contact proves unsuccessful or infeasible, we will employ statistical imputation methods. For graphically presented data, we will use digital plot digitisation software to extract numerical values from Figs. When studies report only inferential statistics such as F-values, t-values, or exact p-values, we will employ established conversion formulae to calculate standardised mean differences [39]. In cases where means are unreported, but medians and ranges are provided, we will use validated approximation methods to estimate means and standard deviations, noting these approximations in our dataset and conducting sensitivity analyses to assess their impact.

Our primary metric will be the standardised mean difference using Hedges’ g, which is preferred over Cohen’ s d for its correction for small sample bias (when sample sizes <20).

When means and standard deviations are available for both intervention and control groups, we will calculate effect sizes directly from these values. For studies reporting pre-post change scores, we will utilise these alongside correlation coefficients when available. Critically, for each extracted effect size we will record in our coding sheet whether the standardisation denominator is based on (a) the pooled post-intervention standard deviation (between-group variability) or (b) the standard deviation of change scores (within-subject variability). Effect sizes standardised on change-score variability can be substantially inflated relative to those standardised on between-group variability, because the denominator conflates the magnitude of the treatment effect with its homogeneity across participants. We will therefore conduct sensitivity analyses comparing pooled estimates computed from each class of effect sizes separately and will report these two sets of estimates alongside the overall pooled estimate so that readers can assess the impact of standardisation choice on the conclusions. In cases where only inferential statistics are provided, we will extract or convert t-statistics, F-statistics, or p-values to derive standardised mean differences. All reported effect sizes will be converted to our common metric to ensure comparability across studies.

When studies designate a primary outcome measure, we will prioritise this selection. In the absence of such designation, we will favor well-validated, standardised tests over custom or ad-hoc measures. For studies reporting multiple measures within a single cognitive domain, we will calculate composite scores by averaging effect sizes while accounting for the correlation between measures, assuming a correlation of 0.5 when specific correlation data are unreported. This default of r = 0.5 is consistent with conventional practice in psychological and cognitive meta-analyses ([40];[41]) and represents a conservative mid-point estimate of the inter-measure correlation when empirical values are unavailable. Importantly, we acknowledge that this assumption can meaningfully influence pooled composite estimates. We therefore commit to conducting sensitivity analyses at r = 0.3 and r = 0.7 to evaluate the robustness of composite effect sizes to this assumption; results across all three values will be reported transparently in the supplementary materials.

RO1: Model selection and justification

Given the expected heterogeneity across populations, stimulation protocols, and outcome measures, we will employ a random-effects model for all primary analyses. We will use the restricted maximum likelihood estimator (REML) as our default τ² estimator for all primary analyses. REML is preferred over the method-of-moments DerSimonian–Laird (DL) estimator because simulation studies consistently show that REML produces less biased estimates of between-study variance and yields confidence intervals with better coverage, particularly in the small-to-moderate sample sizes typical of this literature ([42];[43]). We retain DL only in the sensitivity analysis comparing estimators (see Statistical Model Sensitivity below), where its results will be presented alongside REML for transparency.

The overarching analysis will pool all eligible effect sizes comparing taVNS to control conditions across all cognitive outcomes. This analysis will yield an overall Hedges’ g with its 95% confidence interval, accompanied by z-test statistics and associated p-values. Importantly, we will report prediction intervals alongside confidence intervals when feasible, as these provide important information about the expected range of true effects in future similar studies, accounting for between-study heterogeneity.

RCTs constitute the primary evidence base for this synthesis. Non-randomised interventional studies (single-arm before–after designs) are retained as a supplementary stratum and will be analysed separately using ROBINS-I for risk-of-bias assessment and a distinct GRADE certainty rating (as described in the Quality Assessment section above); they will not be pooled with the RCT stratum in the primary analysis but will be reported alongside the primary results for contextual comparison.

RO2: clinical population stratification

To address our second RO concerning taVNS effects in clinical populations, we will stratify studies into distinct population subgroups. These include healthy control participants, individuals with depression, patients diagnosed with ADHD, those with Alzheimer’s disease or mild cognitive impairment, participants with other neuropsychological conditions. We will conduct separate meta-analyses for each clinical population that includes at least three independent studies, ensuring sufficient power for meaningful estimation. Direct comparison between clinical and healthy populations will be performed using mixed-effects models, which allow us to test whether population type significantly moderates the effect of taVNS on cognitive functions. For clinical populations where sufficient data exist, we will calculate the number needed to treat, providing a clinically interpretable metric of intervention efficacy that complements standardised mean differences.

Heterogeneity assessment

RO3: moderator analyses

To pre-empt concerns about multiplicity and interpretive ambiguity, the moderator analyses are organised into two tiers: (1) a small number of pre-specified primary moderators with strong theoretical or empirical rationale, and (2) a larger set of exploratory moderators to be interpreted with appropriate caution and as hypothesis-generating rather than confirmatory. Primary moderators are: (a) stimulation frequency (Hz) and intensity (mA), given that dose-response is the central mechanistic question; (b) clinical versus healthy population status, given that RO2 directly concerns this distinction; and (c) overall risk of bias, to assess whether methodological quality inflates effect estimates. All remaining moderators listed below are designated exploratory and will be reported with explicit acknowledgement of increased Type I error risk.

Sensitivity analyses

Publication bias analyses

Additional exploratory analysis

Interpretation framework

Statistical software and packages

All statistical analyses will be conducted using RStudio (version 2025.03.1, R version 4.4.1). The primary meta-analytic procedures will be implemented through the metafor (v4.6-0) and meta (v6.3-2) packages for random- and mixed-effects modelling, effect size computation, and heterogeneity assessment. For robust variance estimation (RVE) and cluster-robust sensitivity analyses, we will employ the clubSandwich (v0.6.0) and robumeta (v3.1) packages. Moderator and meta-regression analyses will be performed using the metafor framework with REML estimation. For generating contour-enhanced funnel plots and forest plots, we will use ggplot2 (v3.5.1) and ggpubr (v0.6.0).

Potential theoretical and mechanistic implications

Within the conceptual “Sorting Hat” framework proposed in this protocol; cognitive functions are understood as an integrated system composed of interdependent functional domains. taVNS, by virtue of its widespread neural projections, may exert differentiated yet convergent influences across these domains, strengthening working memory and attention, enhancing emotional regulation and social cognition), promoting adaptive flexibility and learning, and optimising executive functioning and decision-making. This integrative approach transcends the reductionist tendency of prior research, recognising that cognitive processes operate as an orchestrated network rather than isolated faculties.

This protocol adopts several potential methodological advancements to enhance the validity and interpretability of results. First, by including both healthy and clinical populations, it aims to bridge the gap between mechanistic and translational research and enable the examination of whether taVNS exerts baseline-dependent effects—enhancing cognition in impaired populations while fine-tuning neural efficiency in healthy individuals.

Second, the comprehensive moderator analyses proposed, including stimulation parameters (frequency, intensity, duration), participant demographics, and study design variables, will provide critical insights into the boundary conditions of taVNS efficacy. Such analyses can inform the standardisation of stimulation protocols, a current limitation in the field.

Furthermore, the inclusion of dose–response and time-course analyses will represent a significant step toward identifying optimal stimulation “dosage” and understanding the temporal dynamics of taVNS effects. The use of the GRADE framework to assess the overall certainty of evidence will also provide clinicians and researchers with a clear interpretative benchmark. If the forthcoming meta-analytic synthesis identifies consistent and clinically meaningful improvements, taVNS could be positioned as a cost-effective, safe, and accessible adjunct for cognitive rehabilitation.

At the same time, exploring its effects in healthy and subclinical populations can shed light on its potential as a cognitive enhancer—a topic with growing ethical and societal relevance. Establishing empirical boundaries between therapeutic benefit and enhancement will inform future policy, clinical guidelines, and research ethics.

Patient and public involvement statement

Patients or members of the public were not involved in the design, conduct, reporting, or dissemination plans of this research. Because this study is a systematic review and meta-analysis protocol based solely on data extracted from previously published studies, no direct patient recruitment or public participation was required. Nevertheless, the review is designed with clear translational intention to inform future clinical applications of taVNS for cognitive enhancement and rehabilitation in both healthy and clinical populations. To ensure that future stages of this research remain patient-centred, the findings from the completed review will be disseminated in accessible formats through public-facing summaries and infographics and shared with patient advocacy and neuromodulation interest groups. This approach aims to bridge the gap between mechanistic research and real-world clinical understanding, ultimately contributing to evidence-based and ethically grounded use of taVNS in cognitive health interventions.