Somatosensory training: a systematic review and meta-analysis with methodological considerations and clinical insights

Abstract

Somatosensory training, which involves repetitive somatosensory stimulation, has been employed to enhance somatosensory performance by modulating excitability in the primary somatosensory cortex. This process, known as perceptual learning, can benefit stroke patients with somatosensory deficits. However, its effectiveness in both healthy individuals and stroke patients has not been thoroughly investigated. This systematic review and meta-analysis aimed to evaluate the effectiveness of somatosensory training in these groups. However, no eligible data on stroke patients were identified, excluding them from the analysis. In healthy participants, somatosensory training improved performance in 61.2% datasets, but this effect was observed only at the stimulated site. Additionally, it increased early somatosensory-evoked potential amplitudes in 76.9% of datasets at the stimulated site, with no effect on the non-stimulated site. Despite these moderate improvements, the risk of bias assessment revealed methodological concerns including randomization process, proper control conditions, blinding information, and missing data. The meta-analysis focused on the impact of somatosensory training on tactile two-point discrimination (TPD) in various factors, including different age groups, stimulus durations, stimulus frequencies, and stimulus types. A marked reduction in TPD threshold was observed at the stimulated finger post-training compared to pre-training, though there was a noticeable heterogeneity across studies. In contrast, no significant changes occurred at the non-stimulated fingers, and the subgroup analysis found no specific factors influencing TPD improvements. Although somatosensory training benefits healthy individuals, the variability and methodological concerns highlight the need for further high-quality research to optimize its use in treating somatosensory deficits in stroke patients.

Introduction

Somatosensory training, which applies repetitive somatosensory stimulation (RSS), has been shown to passively enhance somatosensory performance [1,2,3]. It is often employed to treat stroke patients experiencing deficits in various somatosensory modalities [4]. This approach, known as perceptual learning, involves simple exposure to somatosensory stimulation for a few hours [5]. While the term “perceptual” encompasses various senses, including vision and hearing, the term “somatosensory training” is used here to specifically denote methods focused on somatosensory stimulation. Somatosensory training typically does not require participants to actively focus on external stimuli [2, 3, 6], making it easily applicable in neurorehabilitation fields. The effects of this training are thought to stem from neuroplastic mechanisms in the primary somatosensory cortex (S1) [2, 6], which is primarily responsible for somatosensory processing [7]. Pharmacological studies suggest that these neuroplastic changes may be mediated by the activity of N-methyl-D-aspartate (NMDA) [2], dopaminergic [8], and cholinergic receptors [9]. Additionally, a magnetic resonance spectroscopy study has shown that gamma-aminobutyric acid (GABA) concentration in the sensorimotor area is involved in neuroplastic changes [10]. In addition to functional changes, somatosensory training induces structural reorganization in the S1 [11] and preferentially influences the cortical representation corresponding to the stimulated site [12].

Other stimulation techniques targeting brain regions on the scalp, known as non-invasive brain stimulation (NIBS), have attracted considerable attention as tools for neuroplastic induction, similar to RSS. Examples of NIBS include transcranial direct current stimulation [13], repetitive transcranial magnetic stimulation [14], and theta-burst stimulation [15]. These interventions are primarily applied to the primary motor cortex (M1). However, their effectiveness has recently become controversial due to remarkable variability both between and within participants [16,17,18,19]. Similarly, our previous systematic review, which focused on the S1, revealed that NIBS interventions targeting this region have minimal impact on somatosensory performance and S1 excitability [20]. Despite these recent trends, the effectiveness of somatosensory training has not yet been thoroughly investigated. Consequently, it remains uncertain whether somatosensory training consistently improves somatosensory performance and modulates S1 excitability. If it does not, future research should explore alternative strategies or develop optimized RSS parameters for stroke patients with somatosensory deficits.

We therefore aimed to systematically summarize the effectiveness of somatosensory training in enhancing somatosensory performance and S1 excitability by statistically analyzing relevant studies in healthy individuals and stroke patients. To assess S1 excitability, we focused on electroencephalography (EEG) and magnetoencephalography (MEG), which capture electrical and magnetic neural signals at a high temporal resolution, respectively. These neuroimaging techniques assess S1 excitability via somatosensory-evoked potential (SEP) and somatosensory-evoked magnetic field (SEF), respectively. Additionally, we included various types of somatosensory performance assessments, such as electrical, vibratory, and mechanical stimulations. Alongside this systematic review, we conducted a meta-analysis of tactile two-point discrimination (TPD) tasks, which are commonly used to assess the effects of somatosensory training in healthy individuals [2, 21]. The meta-analysis aimed to investigate the overall effect and heterogeneity of somatosensory training on performance across studies using mean effect size and prediction interval (PI). Additionally, subgroup analyses were conducted to determine whether aging and RSS parameters influence the effectiveness of somatosensory training. Aging has been shown to impair neuroplasticity [22]. The effectiveness of neuroplastic changes can also be affected by certain factors related to stimulus parameters [1, 23]. We hypothesized that somatosensory training improves somatosensory performance, but its effects may vary between studies, with aging and stimulus parameters contributing to this variability. Furthermore, we explored novel perspectives on somatosensory training, shedding light on stimulus strategies as well as addressing methodological considerations and clinical applications for stroke patients. These insights aim to deepen our understanding of somatosensory training and provide guidance for optimizing its efficiency and ensuring high-quality research in future studies.

Methods

The study protocol has been registered at the PROSPERO (https://www.crd.york.ac.uk/prospero/; registration number: CRD42023487471). Initially, TPD and grating orientation discrimination (GOD) were planned as the main outcomes; however, due to the small sample size for GOD, only TPD was included in the meta-analysis. After submitting the initial draft, we considered including stroke patients in this review. The current review followed the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) 2020 statement [24]. The PRISMA 2020 main and abstract checklists were completed online (https://www.prisma-statement.org/prisma-2020-checklist) and exported as a Word file. While the main checklist met all requirements, the abstract checklist only partially met the requirements (Additional file 1: Tables S1 and S2), as summarizing all the necessary information within the 250-word limit was challenging. However, these details were provided in the main text. This study did not require ethical approval as it involved a thorough analysis of previous published data.

Search strategy and study selections

An initial literature search covering the period from January 1, 2000, to June 24, 2024, was conducted in PubMed and Web of Science databases on June 24, 2024, using the filters “Humans” and “English.” The search for healthy individuals employed the following search strings: ([“somatosensory cortex” OR “somatosensory system” OR “somatosensory function” OR “somatosensory-evoked potential” OR “somatosensory-evoked magnetic field”] AND [“perceptual learning” OR “somatosensory learning”]) (Additional file 1: appendix 1). A subsequent literature search covering the period from January 1, 2000, to December 24, 2024, was conducted in PubMed and Web of Science databases on December 24, 2024, using the filters “Humans” and “English.” The search for stroke patients employed the following search strings: ([“somatosensory cortex” OR “somatosensory system” OR “somatosensory function” OR “somatosensory-evoked potential” OR “somatosensory-evoked magnetic field”] AND [“perceptual learning” OR “somatosensory learning”] AND [“stroke”]) (Additional file 1: appendix 2). Additionally, a manual search of the reference sections of the retrieved studies was conducted when necessary. All article information was then exported to EndNote Version X8.

Eligibility criteria

The selected studies had to meet the following eligibility criteria: (1) Peer-reviewed original research published in English; (2) Healthy adult participants (aged > 18 years) or stroke patients with somatosensory deficits; (3) Somatosensory performance or SEP/SEF measurements involving the upper limbs; (4) Evaluation of pure somatosensory performance using electrical, vibratory, or mechanical stimulation (excluding pain and itch sensations) both before and after RSS; (5) assessment of S1 excitability via SEPs or SEFs before and after RSS; (6) Applying RSS to the upper limb without additional interventions, such as medication or NIBS; (7) Post-intervention measurements taken within 24 h with statistical data. Duplicate articles were automatically excluded using EndNote software. The titles and abstracts of the articles were initially screened by a single reviewer (R.S.), and eligibility for inclusion was independently assessed by two reviewers (R.S. and S.K.), with full-text evaluation. Discrepancies were resolved by a third reviewer (K.S.). In stroke patients, none of the available studies met the inclusion criteria, and no eligible data were identified. Consequently, stroke patients were excluded from the review, and their potential inclusion was addressed solely in the discussion section. The systematic review and meta-analysis were therefore concluded at this state.

Data extraction and analysis

The following sections describe the analysis process conducted exclusively on healthy adults. In this systematic review, somatosensory performance and SEP/SEF amplitudes were regarded as the primary outcomes. Following a full-text assessment, we categorized the effects of RSS on somatosensory performance as improved, unchanged, or declined, and SEP/SEF amplitudes as increased, unchanged, or decreased. These categorizations were based on the results of ANOVA or t-tests comparing baseline and post-training measurements within 24 h. P values of less than 0.05 between time points were deemed statistically significant.

In the meta-analysis, the TPD threshold (mm) was selected as the primary outcome, as this task was widely used across the included studies. The effect of RSS on S1 excitability was not assessed because of methodological variations, including differences in reference EEG electrodes, targeted SEP/SEF components, and analysis methods. Outcomes were extracted from the articles included in the systematic review by a single reviewer (R.S.). After full-text assessment, it was noted that many studies used one or more fingers on the right hand as the test (stimulated) fingers, while one or more fingers on the left hand often served as the control (non-stimulated) fingers when assessing TPD within the same session. To minimize the influence of site-specific differences [25], only these particular test and control settings were included in the meta-analysis. In contrast, a few studies selected a test finger on the right hand with another finger on the same hand as the control. While two studies reported no transfer effect to the control finger [1, 26], one study reported such an effect [23]. Due to this inconsistency, this control setting was excluded from the meta-analysis. Furthermore, due to the limited number of studies that implemented a control RSS to the same finger in a separate session [1, 27,28,29,30], the meta-analysis focused on the test and control fingers within the same session. When multiple fingers were assessed for TPD, the thresholds were averaged across fingers for each hand. Similarly, when multiple time points were assessed for TPD within 24 h, the thresholds were averaged across those time points for each hand. The mean and standard deviation (SD) values measured between the pre- and post-RSS time points were extracted from the main text, figures, tables, or supplementary data. If graphs were used instead of original data reporting, the data were digitized using GetData Graph Digitizer (v. 2.26). When studies reported the standard error of the mean instead of the SD, the SD was calculated. If median data were provided instead of the mean, the mean and SD were estimated using the sample size, median value, and first and third quartiles [31]. For studies that only reported t-values or p-values from paired t-tests comparing pre- and post-training, these values were used in the meta-analysis. For all missing data, we contacted the corresponding authors to obtain the relevant information.

Meta-analysis

All statistical analyses were performed by a single reviewer (R.S.) using Comprehensive Meta-Analysis software (version 4.0). Two separate datasets were developed from the studies included in the meta-analysis. One dataset focused on the TPD threshold for the test finger, while the other dataset focused on the TPD threshold for the control finger. The meta-analysis compared pre- and post-training TPD thresholds across studies to estimate the overall effect of RSS using a random-effects model. For the test finger, subgroup analyses were further conducted based on different age groups (young and older adults [≥ 51 years]), stimulus durations (short [≤ 1 h] and long [˃ 1 h] RSS), stimulus frequencies (low-frequency [1–2 Hz] and high-frequency [20 Hz] RSS), and stimulus type (tactile and electrical RSS). The classification of stimulation frequency and duration was subjectively determined after reviewing the collected data results. A mixed-effects model was used for these analyses, which assumes that studies within a subgroup are random while subgroups themselves are fixed [32]. The mean, SD, and sample size were primarily used as inputs to compute the standardized mean differences (SMDs) using Hedges’ adjusted g, which includes an adjustment for small sample bias [33]. The sample size of each study influenced the weight assigned to its mean and SD in the analysis. Z-statistics and corresponding P-values were calculated to determine whether the mean effect size for the TPD threshold differed significantly between pre- and post-training in the main analysis. In the subgroup analysis, heterogeneity statistics and P-values were computed to assess whether the mean effect sizes varied significantly across fixed effects. Statistical significance was set at P < 0.05.

Heterogeneity analysis

A 95% PI was calculated to estimate the degree of heterogeneity in the main analyses. The PI reflects the range where true effects are likely to fall around the mean, while the confidence interval (CI) indicates the precision of the mean effect size estimate [34]. In line with recent recommendations [34], a 95% PI was used to assess the variability in mean effect sizes, providing a clear representation of the dispersion of true effects rather than relying on Cochran's Q, P-value, and I-square test. The Q-statistic was employed to test the null hypothesis that all studies included in the analysis share a common effect size [34]. Additionally, the I-squared statistic was used to quantify the proportion of variance in the observed effects that can be attributed to true differences in effect sizes rather than to sampling error [34].

Risk of bias

The risk of bias (RoB) in the included studies was assessed at the outcome level by a single reviewer (R.S.) using the Cochrane Collaboration tool for randomized trials (RoB 2) [35]. This tool evaluates internal validity across five distinct domains: (1) bias arising from the randomization process, S) bias arising from period and carryover effects, (2) bias due to the deviations from intended interventions, (3) bias due to missing outcome data, (4) bias in the measurement of the outcome, and (5) bias in the selection of the reported results. The RoB is categorized as low, some concerns, or high [35]. Although this tool is widely applicable and recommended in meta-analyses, it was not fully applicable to the current meta-analysis due to the unique study design of the included studies. Specifically, most studies utilized an allocation strategy where each participant served as their own control, with test and control measurements taken from right and left index fingers, respectively, within the same session. This design does not align with a traditional crossover model, as each participant underwent only one intervention. Given this context, the signaling questions in the domains related to the randomization process and period/carryover effects (domains 1 and S) were adjusted. Specifically, the test and control fingers were treated as distinct test and control groups, respectively.

The RoB 2 tool includes an integrated algorithm that automatically calculates the level of bias for each domain and the overall bias using an Excel sheet (https://www.riskofbias.info/welcome/rob-2-0-tool/rob-2-for-cluster-randomized-trials). The reviewer evaluated responses to each signaling question across the domains using options such as ‘yes,’ ‘probably yes,’ ‘probably no,’ ‘no,’ and ‘no information.’ Detailed instructions on how to classify each response were reviewed in advance, accessible at the following link (https://sites.google.com/site/riskofbiastool/welcome/rob-2-0-tool/current-version-of-rob-2). If any uncertainty arose during the evaluation, the reviewer referred to the guidelines to ensure accurate classification. Once all responses were recorded, the tool automatically determined the RoB for each domain. The overall RoB was then determined based on the outcomes from all domains. If any domain was deemed to have a high risk of bias, the overall RoB 2 classification was labeled as “high risk.” Conversely, if there were concerns in at least one domain, the overall RoB 2 for the study was categorized as having “some concerns.”

Publication bias analysis

Publication bias was evaluated using a funnel plot with the trim-and-fill method. This approach aimed to evaluate the visual symmetry of the distribution to identify potentially missing studies and estimate the adjusted effect size [36, 37]. Similarly, Begg’s adjusted rank correlation test [38] and Egger’s regression test [39] were employed in the main analysis to evaluate the potential for publication bias based on P values.

Study information

After conducting a full-text assessment of somatosensory performance, various characteristics were extracted. These included sample size, type of RSS, RSS parameters [such as duration, inter-trial interval (ITI), frequency, intensity, and target area], participants’ state during RSS, the performance task, and the effect of RSS on somatosensory performance. Similarly, for SEP/SEF measurements, characteristics related to participants, RSS, and EEG/MEG were collected, including sample size, type of RSS, RSS parameters, participants’ state during RSS, SEP/SEF components, electrode location, and the effect of RSS on SEP/SEF amplitudes.

Results

Selection of studies

A flow chart illustrating the study selection process for healthy adults is presented in Fig. 1. The literature search for healthy individuals identified a total of 3,233 studies, of which 2,695 were retained after removing duplicates. Titles and abstracts were screened based on the predefined selection criteria. In total, 59 full-text articles were eventually assessed for eligibility, and 21 articles were excluded for the following reasons: (1) Outcomes were not assessed before and after RSS (N = 19), (2) Statistical data were not available (N = 1), and (3) Nociceptive stimulation was applied instead of tactile stimulation (N = 1). This left a total of 38 articles focusing on somatosensory performance, of which 8 specifically focused on SEP/SEF recordings, to be included in the systematic review. In the meta-analysis, 13 articles were excluded from the systematic review for the following reasons: (1) TPD was not assessed (N = 11); (2) TPD data were not available (N = 1); (3) Test and control fingers were on the same hand (N = 1). Consequently, a total of 25 articles were included in the meta-analysis.

Fig. 1

PRISMA flow diagram. This diagram illustrates the process of conducting a literature search and selecting studies in healthy adults. Abbreviations: RSS, repetitive somatosensory stimulation; SEP, somatosensory-evoked potential; SEF, somatosensory-evoked magnetic field; TPD, two-point discrimination

A flow chart illustrating the study selection process for stroke patients is presented in Additional file 1: Figure S1. The literature search for stroke patients identified a total of 476 studies, of which 443 were retained after removing duplicates. Titles and abstracts were then screened based on the predefined selection criteria. In total, 4 full-text articles were eventually assessed for eligibility, but all were excluded for the following reasons: (1) somatosensory performance and early SEPs were not evaluated (N = 2), (2) the study was based on single-case experiments (N = 1), and (3) the somatosensory training involved multiple modalities of somatosensory functions (N = 1).

Participant and methodological characteristics

Table 1 summarizes the number of participants and the methodological characteristics related to somatosensory performance in the retrieved studies. The RSS was primarily applied to any of the fingers on the right hand as the test finger and any of the fingers on the left hand as the control finger, utilizing either tactile or electrical stimulation. Stimulus parameters varied widely across the studies, including stimulus duration (20 min–6 h), ITI, frequency (1–25 Hz), and intensity (> sensory threshold). Most studies required participants to passively receive the RSS, while a few required them to actively pay attention to the intervention [30, 40, 41]. The tasks included in these studies were as follows: (1) TPD: distinguishing stimulation by a single tip versus two closely spaced tips (84 datasets), (2) GOD: discriminating the orientation of hemispherical dome grooves (orthogonal or parallel) at different groove widths (11 datasets), (3) Somatosensory temporal discrimination (STD): discriminating between two temporally separate electrical stimuli (20 datasets), (4) Frequency discrimination (FD): detecting differences in tactile stimulation frequency (14 datasets), (5) Roughness discrimination (RD): discriminating between two different surface roughnesses using sandpaper (6 datasets), (6) Touch detection (TD): detecting a mechanical stimulus, such as a von Frey Filament (14 datasets), (7) Pattern discrimination (PD): discriminating different patterns of tactile stimulation in the spatial domain (3 datasets), and (8) Tactile threshold (TT): discriminating between different amplitude ranges of tactile stimulation (2 datasets).

Table 1 Studies investigating the effects of somatosensory training on somatosensory performance

Table 2 summarizes the number of participants and the methodological characteristics of the SEP/SEF measurements across the included studies. As with somatosensory performance studies, the RSS was primarily applied to any of the fingers on the right hand as the test finger and any of the fingers on the left hand as the control finger, utilizing either tactile or electrical stimulation. Notably, the stimulus parameters displayed considerable variation between studies, including stimulus duration (45 min or 3 h), ITI, frequency (1 or 20 Hz), and intensity (> sensory threshold). All studies required participants to passively receive the RSS. SEP was generally elicited through either tactile or electrical stimulation, while SEF was exclusively evoked by tactile stimulation in a single study [26]. Early SEP components, such as P14, N20, and P25, were analyzed. The amplitudes of each SEP component were assessed as baseline-to-peak or peak-to-peak. Some studies have measured paired-pulse inhibition using these components with various inter-stimulus intervals (ISIs) ranging from 5 to 100 ms. The most commonly utilized ISIs were 5 ms, 20 ms, and 40 ms, each represented in eight datasets. Notably, most studies employed only a few EEG electrodes rather than multichannel arrays. For right-hand stimulation, the electrodes were typically placed at C3’ or CP3 to measure S1 excitability, with Fz or an earlobe electrode serving as a reference electrode. For left-hand stimulation, C4’ was used to measure S1 excitability, while Fz was used as a reference electrode. Three studies conducted source-level analysis of SEP/SEF data [2, 12, 26], while the remainder performed sensor-level analyses.

Table 2 Studies investigating the effects of somatosensory training on SEP/SEF amplitude

The effects of somatosensory training on somatosensory performance

The selected articles investigating the effects of somatosensory training on somatosensory performance are outlined in Table 1. A total of 38 studies included 155 datasets, with 85 datasets for the test finger and 71 for the control finger. Notably, one dataset was included in both fingers, as the effect of test and control fingers was collapsed in the statistics [41]. For the test finger, 52 datasets (61.2%) demonstrated an improvement in somatosensory performance following RSS, while 28 datasets (32.9%) exhibited no significant change, and 5 datasets (5.9%) showed a decrease. For the control finger, 6 datasets (8.8%) demonstrated an improvement in somatosensory performance, while 61 datasets (89.7%) exhibited no change, and 1 dataset (1.5%) showed a decrease. Three datasets were unavailable for the control finger, and no results were reported.

The effects of somatosensory training on SEP/SEF

The selected articles investigating the effects of RSS on SEP are summarized in Table 2. A total of 8 studies included 26 datasets, with 13 datasets for the test finger and 13 datasets for the control finger, in single-pulse SEP. For the test finger, 10 datasets (76.9%) demonstrated an increase in SEP amplitude post-RSS, while 1 dataset (7.7%) exhibited no change, and 2 datasets (15.4%) showed a decrease. In contrast, all 13 datasets (100.0%) for the control finger displayed no change in SEP/SEF amplitudes. Five studies analyzed paired-pulse SEP, contributing 28 datasets, with 26 datasets for the test finger and 2 datasets for the control finger. For the test finger, 18 datasets (69.2%) showed an increase in inhibition following RSS, while 1 dataset (3.8%) exhibited no change, and 7 datasets (26.9%) showed a decrease. For the control finger, both datasets (100.0%) showed no changes in inhibition.

Risk of bias

The RoB in somatosensory performance is presented in Fig. 2A. Nearly half (47.4%) of the studies exhibited a high overall RoB. The primary reasons for these high-risk ratings were as follows: Domain (1) The absence of randomized allocation (i.e., inclusion of only one group without a control group or no information); Domain (3) Lack of detailed information regarding missing outcome data; Domain (4) Potential assessors’ awareness of the intervention, inherent to the experimental design; and Domain (5) Potential for selective reporting of results.

Fig. 2

Risk of bias of the included studies. A The risk of bias in somatosensory performance; B The risk of bias in SEP/SEF. Abbreviations: SEF, somatosensory-evoked magnetic field; SEP, somatosensory-evoked potential

The RoB in SEP/SEF measurements is presented in Fig. 2B. One-quarter of the studies (25%) exhibited a high overall RoB. The primary reason for these high-risk ratings was Domain 5 “Potential for selective reporting of results.”

Meta-analysis

Figures 3 and 4 provide a summary of the effect sizes across all studies for the test (39 datasets) and control (20 datasets) fingers, respectively. The effect of RSS for the test finger was significant between pre- and post-intervention (SMD = -0.71, 95% CI [ − 0.94, − 0.48], 95% PI [ − 2.03, 0.60], Z-value = − 6.067, P < 0.001), suggesting that RSS resulted in a lower TPD threshold. The Q-value was 201.0 with 38 degrees of freedom, and the P-value was less than 0.001. The I-squared statistic was 81%. The effects of RSS for the control finger were not significant between pre- and post-intervention (SMD = 0, 95% CI [ − 0.12, 0.12], 95% PI [ − 0.23, 0.24], Z-value = 0.04, P = 0.97). The Q-value was 21.8, with 19 degrees of freedom, and the P-value was 0.29. The I-squared statistic was 13%.

Fig. 3

Forest plot for the effect of somatosensory training on TPD threshold at the stimulated site. The analysis was based on 39 studies. The random-effects model was employed for the analysis. The mean effect size was − 0.71, with a 95% CI ranging from − 0.94 to − 0.48. This interval represents the possible range for the mean effect size in comparable studies. The Z-value tests the null hypothesis that the mean effect size equals zero. With a Z-value of − 6.07 and P < 0.001, the null hypothesis at the alpha level of 0.05 was rejected. This suggests that the mean effect size in comparable populations is significantly different from zero; somatosensory training significantly improved TPD threshold at the stimulated site. The PI is − 2.02 to 0.60. The true effect size in 95% of all comparable populations falls in this interval. The lowercase letters following the year of each study indicate that different datasets are referenced within the same study. Abbreviations: CI, confidence interval; PI, prediction interval; SMD, standard mean difference; TPD, two-point discrimination

Fig. 4

Forest plot for the effect of somatosensory training on TPD threshold at the non-stimulated site. The analysis is based on 20 studies. The random-effects model was employed for the analysis. The mean effect size was 0, with a 95% CI of − 0.12 to 0.12. The Z-value tests the null hypothesis that the mean effect size is zero. The Z-value is 0.04, with P = 0.97. Using a criterion alpha of 0.05, this null hypothesis was not rejected. This suggests that the mean effect size in comparable populations is not significantly different from zero; somatosensory training did not significantly improve the TPD threshold at the non-stimulated site. The PI is − 0.23 to 0.24. The true effect size in 95% of all comparable populations falls in this interval. The lowercase letters following the year of each study indicate that different datasets are referenced within the same study. Abbreviations: CI, confidence interval; PI, prediction interval; SMD, standard mean difference; TPD, two-point discrimination

Publication bias

Figure 5 summarizes publication bias using funnel plots. The trim-and-fill method estimated the overall effect sizes as -0.71 for the test finger and 0 for the control finger, both matching the original values without imputed data. For the test finger, neither Begg’s adjusted rank correlation test (Tau = − 0.05, P = 0.64) nor Egger’s regression test (intercept = 0.21, t = 0.18, P = 0.86) demonstrated statistical significance. For the control finger, Begg’s adjusted rank correlation test (Tau = 0.05, P = 0.75) also yielded no significant results; however, Egger’s regression was statistically significant (intercept = 2.63, t = 2.30, P = 0.03).

Fig. 5

Funnel plots for somatosensory training on TPD thresholds. A Funnel plot for the test finger; B Funnel plot for the control finger. Blue diamonds represent the original SMD value, while red diamonds represent the adjusted SMD value. The trim-and-fill method estimated the overall effect sizes to be -0.71 for the test finger and 0 for the control finger, consistent with the original values obtained without imputing data. This suggests that there is little to no publication bias or data skewness affecting the analysis. Abbreviations: SMD, standard mean difference; TPD, two-point discrimination

Subgroup analysis

For the test finger, a subgroup analysis was performed to identify factors contributing to the large variability across studies in the main meta-analysis. Based on heterogeneity statistics, no significant differences were observed between the young (36 datasets) and old (3 datasets) groups (Q = 0.66, df (Q) = 1, P = 0.42), short (15 datasets) and long (24 datasets) stimulus durations (Q = 2.81, df (Q) = 1, P = 0.09), low (26 datasets) and high (12 datasets) frequency stimulation (Q = 0.24, df (Q) = 1, P = 0.62), or electrical (6 datasets) and tactile (31 datasets) stimulation (Q = 0.06, df (Q) = 1, P = 0.80).

(Additional file 1: Figures S2–S5).

Discussion

This systematic review and meta-analysis investigated the effects of somatosensory training on both somatosensory performance and S1 excitability across a variety of published studies involving healthy adults and stroke patients. However, due to insufficient studies meeting the inclusion criteria for stroke patients, the review was limited to healthy adults. The systematic review revealed that somatosensory training improved somatosensory performance in various modalities in more than half of the datasets (61.2%), with this effect being limited to the stimulated sites. Similarly, this training increased early SEP amplitudes in most datasets for the stimulated sites (76.9%), with no observed changes in SEP amplitudes for the non-stimulated sites. However, the RoB assessment indicated a high RoB in the overall results of several studies (25.0–47.4%). Additionally, the meta-analysis indicated that somatosensory training improves the TPD threshold at the stimulated sites with minimal publication bias detected. However, this effect largely varied between studies, and the subgroup analysis could not identify the contributing factors.

Effects of somatosensory training on various somatosensory performance

The RSS has primarily been applied to the upper limbs to improve somatosensory performance. This intervention moderately improved somatosensory performance at the stimulated site, while the effects at the non-stimulated site were small, with varying RSS parameters used across studies. Despite the widespread use of RSS, the optimal RSS parameters for achieving consistent and robust improvements remain largely unknown. While several studies have explored the influence of RSS parameters on somatosensory training, definitive conclusions are yet to be established. For instance, Ragert, et al. [23] investigated the effect of low- and high-frequency RSS on somatosensory performance, revealing that high-frequency (20 Hz) RSS improved somatosensory performance, whereas low-frequency (1 Hz) RSS disrupted it. Similarly, Godde, et al. [1] investigated the impact of different RSS durations (i.e., 0.5, 2, and 6 h), demonstrating that 2- and 6-h RSS improved somatosensory performance. Furthermore, Schlieper and Dinse [50] reported that high-intensity RSS (below the pain threshold) produced greater improvements in somatosensory performance compared to low-intensity RSS. Although these studies have deepened our understanding of how these RSS parameters influence improvements, consistent replication of findings remains a challenge across investigations exploring various RSS frequencies [12, 26], durations [23, 51, 61], and intensities [3, 52]. Future studies should therefore prioritize developing a standardized protocol for RSS application to achieve consistent and robust somatosensory improvements. Notably, our analysis identified two frequently used RSS parameter sets across studies. The first set involves a 3-h stimulation duration at a frequency of approximately 1 Hz using tactile stimulation [12, 43, 45]. The second set involves a shorter stimulation duration of 20–45 min with a higher frequency of 20 Hz using electrical stimulation [3, 23, 29]. Although direct comparisons between these two sets have not been made, the second set's shorter duration makes it potentially more practical for clinical applications, offering a more feasible option for therapeutic use. However, the generalizability of findings remains limited, highlighting the need for a more standardized protocol that can be applied to a broader clinical population. While these two parameter sets have been widely adopted in healthy adults, the rationale for their predominant use could not be discerned from the background of previous studies. It is possible that these parameters were traditionally adopted because they demonstrated effectiveness in the initial studies where they were employed. Addressing this issue in future research could further enhance the reliability and applicability of RSS in diverse clinical settings.

Somatosensory training typically does not demand active attention from participants, making it easily applicable to patients with somatosensory deficits. However, attention to RSS may play a pivotal role in enhancing somatosensory performance. Notably, focusing attention on a stimulus has been shown to amplify both tactile and pain perceptions [62, 63], indicating that attention modulates somatosensory function in a top-down manner. This is further supported by MEG studies demonstrating that attention to tactile stimuli influences specific brain oscillations in the S1 [64,65,66], potentially enhancing somatosensory perception. Furthermore, attention-deficit hyperactivity disorder is characterized by altered tactile processing [67]. In the systematic review, three studies incorporated RSS that required participants' attention: two studies had participants judge differences in tactile stimulation [40, 41], while one study instructed participants to observe the stimulation of another person’s hand or an object via a screen with real stimulation [30]. These studies combined somatosensory training with attentional engagement but did not specifically examine the effects of attention. Future studies should explore how attention influences somatosensory training outcomes and its potential role in standardizing RSS protocols.

We found that various types of somatosensory performance were employed to assess the effectiveness of RSS. However, a critical question persists: can the effects of RSS be generalized to other tasks? While it seems that S1 is primarily involved in somatosensory perception [7], it is essential to recognize that multiple brain regions concurrently process somatosensory information depending on the task demands. For instance, our studies have shown that TPD threshold is associated with functional connectivity between the S1-superior parietal lobule, S1-angular gyrus, and S1-superior temporal lobule [68]. In contrast, GOD is associated with functional connectivity between the S1-superior parietal lobule and S1-parieto-occipital sulcus [69]. RSS has been shown to improve performance in specific tasks while failing to do so in others [3, 27, 57], indicating that functional differences between tasks may influence the observed performance improvements. However, it remains unclear whether this difference arises from variations in the neural mechanisms required for somatosensory processing across tasks or from methodological differences. Although the effects of RSS may not be universally applicable to all somatosensory assessments, investigating the effects of RSS on several somatosensory performances with various RSS parameters is unfeasible and time-consuming. A more efficient approach would be to first establish a standardized RSS protocol tailored for a specific somatosensory task and then explore the potential generalizability of these parameters across various tasks in healthy adults. Ultimately, it is desirable to apply the established RSS protocol to stroke patients and evaluate its effectiveness in this population.

Effects of somatosensory training on S1 excitability

We identified only eight studies examining the effects of S1 excitability using EEG/MEG. While most studies showed an increase in early SEP amplitudes, two studies reported either no change [26] or a decrease [58] in the SEP/SEF amplitudes. Notably, Erro, et al. [58] applied a low-frequency 1 Hz RSS aiming to inhibit S1 excitability, which resulted in a decrease in somatosensory performance. However, this inhibitory effect was not observed in other studies that utilized low-frequency RSS (mean frequency = 1 Hz) [1, 12, 26]. It is necessary to investigate further whether such bi-directional effects truly exist, as direct comparisons between low- and high-frequency RSS in SEP/SEF studies have not yet been conducted.

Based on the present findings and previous studies, we discussed the potential mechanisms underlying somatosensory training. Similar RSS parameters were employed in both the SEP/SEF and somatosensory performance measurements. Among these studies, only one utilized MEG, while the remainder focused on assessing changes in early SEP components, including P14, N20, and P25, using EEG. The application of RSS preferentially enhances these components at the stimulated site [3, 54, 59]. The subcortical SEP component, P14, is thought to be generated around the foramen magnum and is recorded as a far-field potential from the scalp [70]. In contrast, N20 and P25 originate from area 3b of S1 [71, 72]. Notably, the sources of N20 and P25 within S1 exhibit distinct characteristics: N20 reflects excitatory postsynaptic potentials projecting onto area 3b from the thalamus, while P25 reflects inhibitory postsynaptic potentials occurring within area 3b [71, 72]. Such changes in SEP are thought to reflect underlying functional neuroplasticity within the somatosensory pathway. Importantly, pharmacological and magnetic resonance spectroscopy studies have identified potential contributors to these neuroplastic changes in S1, including NMDA [2], dopaminergic [8], cholinergic [9], and GABAergic [10] activities. Our previous systematic review, which employed various NIBS methods over S1, demonstrated that the amplitudes of N20 and P25 serve as predictors for improvements in somatosensory performance [20]. This supports the idea that improvements in somatosensory performance are, at least in part, driven by changes in S1 excitability. However, robust evidence directly linking somatosensory improvements to changes in S1 excitability remains limited. Other brain regions, such as the superior parietal lobule, angular gyrus [68], and secondary somatosensory cortex [73], also contribute to somatosensory performance but have not been thoroughly investigated. In other words, most studies have focused primarily on investigating the effects of RSS on S1 excitability [3, 6, 59]. Exploratory studies investigating the entire brain may be essential to elucidate the mechanisms underlying the improvements in somatosensory performance. To achieve this in EEG, source analysis is a robust method utilizing co-registration between individual structural magnetic resonance imaging and multiple EEG sensor locations (> 64 channels) generalizability. This approach enables a comprehensive assessment of brain activity by estimating brain activity from thousands of voxels or vertices [74, 75]. A theoretical framework for how somatosensory training enhances somatosensory performance is summarized in Fig. 6.

Fig. 6

Theoretical framework of somatosensory training. This framework illustrates the potential mechanisms through which somatosensory training enhances somatosensory performance. The improvement may be driven by: (1) Neuroplastic changes in S1, (2) Neuroplastic changes in other regions involved in somatosensory processing, or (3) Changes in cortico-cortical networks involved in somatosensory processing. These brain regions involved in somatosensory processing were highlighted for the purpose of visual illustration; therefore, other brain regions, such as the secondary somatosensory cortex, might also be involved. These brain images were created using Brainstorm for visual illustration purposes [76]. These changes may be influenced by neurotransmitters such as NMDA [2], dopaminergic [8], cholinergic [9], and GABAergic [10] activities. Abbreviations: AG, angular gyrus; S1, primary somatosensory cortex; SMG, supramarginal gyrus; SPL, superior parietal lobule

Investigating the effects of inhibitory circuits is crucial given the substantial role of GABAergic activity in mediating neuroplastic changes [77, 78]. The paired-pulse paradigm has frequently been employed to assess these inhibitory circuits in the S1 [68, 79]. In this paradigm, the first stimulus attenuates the cortical response to the second stimulus with interstimulus intervals of 3–300 ms [80]. While GABAergic activity appears to attenuate the second cortical response, the administration of a GABA_A receptor agonist does not modify this response [72, 81]. This paradigm is widely used with various ISIs, ranging from 5 to 100 ms in our systematic review [6, 54, 59]. Nevertheless, the primary mechanisms responsible for the observed attenuation remain unclear. In the systematic review, 69.2% of datasets showed increased inhibition following RSS. This change was observed in only 3 out of 5 studies, while the remainder showed the opposite effect. Such discrepancies highlight a lack of consistent evidence regarding the effects of RSS on inhibitory circuits. Direct comparisons are not feasible due to variations in the paired-pulse methodologies and experimental setups of RSS employed across these studies. Further investigation into the effects of RSS on inhibitory circuits using various ISIs may provide valuable insights into the underlying mechanisms.

Effects of somatosensory training on TPD

This meta-analysis provides for the first time a comprehensive evaluation of the overall effects of RSS on TPD across multiple studies. The findings indicated that RSS, similar to the initial study [1], preferentially improves the TPD threshold at the stimulated finger. However, these results warrant cautious interpretation. Although the null hypothesis was rejected, the 95% PI encompasses both positive and negative values. The 95% PI, which reflects the variability in effect sizes between studies [34], indicates that while RSS improves the threshold in some studies, it decreases or has no effect on performance in others. Various factors may influence the efficacy of RSS. The meta-analysis included a subgroup analysis based on several independent variables such as age, RSS duration, frequency, and stimulus type. Contrary to our expectations, none of the variables exhibited significant differences in their effects on TPD. In prior studies, however, the effects were found to depend on factors such as stimulation frequency [23] and duration [1].　This lack of significant findings implies that the influence of RSS on TPD may be affected by another independent variable or a combination of these variables.

Publication bias was assessed multidirectionally using a funnel plot with the trim-and-fill method, Begg’s adjusted rank correlation test, and Egger’s regression test. The funnel plot provides a subjective assessment of publication bias based on its shape (i.e., symmetric or asymmetric) [37]. The trim-and-fill method estimates missing studies to achieve a symmetrical shape and provides a corrected effect size [37]. In our analysis, we found no significant differences between the true and corrected effect sizes without imputed studies, indicating that the effect of publication bias was minimal. However, it is important to note that when there is substantial heterogeneity between studies, the trim-and-fill method can underestimate the true positive effect, even in the absence of publication bias [82]. To further assess publication bias, we utilized additional methods, with only one of the four additional tests yielding a P-value of < 0.05. This finding suggests that the likelihood of publication bias in our analysis was relatively low, as indicated by the trim-and-fill method.

Most datasets in the systematic review and meta-analysis indicated that RSS specifically influences the stimulated site. This effect was particularly evident when the control condition was established at the contralateral site of the RSS application [2, 12, 43]. However, studies examining the effects of RSS on neighboring fingers have yielded conflicting results. While Harris, et al. [40] reported transfer effects to non-stimulated neighboring fingers [40], other studies did not [1, 26, 52]. Additionally, a few studies demonstrated that RSS applied to the index finger not only affects the stimulated site but also affects the lips [25, 26]. However, this transfer effect was limited to the lips and did not extend to the forearm [25]. These suggest that there is a lack of consensus among studies, highlighting the need for investigations using standardized experimental methods to assess the effects on these remote regions.

The risk of bias and future directions

In several studies, the RoB in assessing somatosensory performance and SEP/SEF was categorized as high. In this field of research, most studies focused on “perceptual learning” OR “somatosensory learning,” often establishing a control condition by evaluating the non-stimulated left hand within the same session, or, in some cases, not including a control condition at all. Studies that used the left hand as a control primarily evaluated somatosensory performance in both hands without disclosing the order of evaluation. These setups were subsequently classified as either “some concerns” or “high risk” in the assessment. Consequently, most studies did not incorporate randomized allocation and frequently involved only a single group without including a separate control group. This approach introduces several issues: (1) Sequential order may influence the outcome, (2) Handiness may influence the outcome, and (3) Participants are easily aware of which fingers are being tested and which are controlled. In summary, the high RoB observed in these studies raises remarkable concerns regarding the reliability and validity of their findings. However, establishing appropriate control conditions in the context of RSS is challenging. Participants can perceive RSS, which is typically applied to a target area above the sensory threshold [12, 29, 61]. If participants received a control condition that did not stimulate a target site, as in previous studies [1, 27,28,29, 51], they could easily discern which condition was applied. If an RSS below the sensory threshold efficiently improves somatosensory performance, establishing an appropriate control condition would be straightforward. However, to the best of our knowledge, no study has explored the use of RSS below the sensory threshold, leaving its practicality uncertain. Another solution is to blind the experimenter, as blinding participants is infeasible at this point. While this does not completely resolve the issue, it can partially reduce the potential bias.

In the somatosensory performance assessment, Domain 3, which concerns missing outcome data, was classified as “high-risk” in five studies. The primary reason for this classification was that these studies excluded several participants due to low performance prior to the application of RSS. However, some studies lacked a clear definition of what constituted "low performance." This ambiguity suggests that low performance may reflect participant characteristics that may affect true values, leading to a “high risk” designation in the assessment. One study defined low performance using a goodness-of-fit estimation based on a logistic regression response [30]. However, this method resulted in the exclusion of 53 out of 101 participants during the initial assessment. Due to the high exclusion rate, this study was also classified as “high-risk,” as the remaining sample may not accurately represent the initial population. To enhance the quality of future research, it is recommended to clearly define the criteria for participant exclusion and minimize exclusions, thereby ensuring a more representative sample of the initial population.

High RoBs were observed in a few studies concerning both the measurement of outcomes and the selection of reported results. First, bias in outcome measurement was attributed to the nature of the experimental assessments. In the 2000s, manual evaluations were commonly used to assess the somatosensory performance [1, 12, 40]. Additionally, given that blinding of examiners was often not implemented in RSS studies, manual evaluations by the examiner could lead to potential manipulation of the results. Although it is unlikely that such manipulations occurred in practice, the potential for bias remains, leading to a high-risk assessment in these cases. Therefore, the blinding of examiners and the use of automated performance assessments are critical for ensuring the reliability of RSS effects. Regarding the selection of reported results, the potential for bias in reporting only favorable outcomes was also noted. Such selective reporting undermines the overall reliability of the study, highlighting the need for comprehensive reporting that includes both negative and unexpected results.

Potential clinical applications for stroke patients

Although stroke patients were included in our literature search, none of the studies met the inclusion criteria, contrary to our expectations. During the full-text assessment, we found that all studies focused on stroke patients with somatosensory deficits. However, these studies did not exclusively evaluate somatosensory performance or conduct purely somatosensory training. Consequently, the effects of somatosensory training on somatosensory performance remain underexplored in stroke patients. This gap likely stems from the inherent characteristics of the somatosensory system, which is less observable compared to the motor system [83]. Therapists may tend to prioritize motor deficits caused by stroke, as these are more easily detectable and directly impact patients' daily lives. In contrast, somatosensory deficits—despite affecting 50% to 80% of stroke patients and involving multiple somatosensory modalities—are often overlooked [83]. Loss of somatosensory function has profound implications for daily life [84, 85], as this function plays a critical role in perceiving and interacting with the external environment. Impairments in tactile sensation hinder the ability to differentiate textures, identify objects through touch, or detect potentially harmful stimuli, such as extreme temperatures or sharp objects [83]. Similarly, proprioceptive deficits disrupt the perception of body position and movement [86]. Moreover, the absence of somatosensory feedback adversely affects fine motor skills [86]. These findings underscore the critical importance of addressing somatosensory deficits in rehabilitation programs for stroke patients. However, specific rehabilitation for somatosensory deficits is lacking [85]. In this review, no studies were identified that implemented purely somatosensory training in stroke patients. Therefore, it is not possible to specify what types of somatosensory training should be applied to this population. As a first step, it is necessary to adapt somatosensory training methods currently used in healthy individuals for stroke patients and evaluate their effectiveness in future studies.

Conclusion

Somatosensory training is a non-invasive approach that enhances somatosensory performance through changes in the excitability of the S1. This systematic review revealed that somatosensory training improved somatosensory performance and increased S1 excitability in more than half of the datasets in healthy adults. However, despite these positive findings, the RoB assessment revealed high bias risks in several studies, necessitating cautious interpretation. Future studies should address the identified methodological issues through more rigorous experimental designs, including the use of randomized controlled trials and appropriate strategies for blinding, handling missing data, and reporting results accurately and transparently. The meta-analysis showed that somatosensory training markedly improved the TPD of the stimulated finger, but this effect exhibited considerable variability across studies. Subgroup analyses did not identify the factors contributing to this variability. Moving forward, to establish RSS as a therapeutic tool in neurorehabilitation, it is essential to address the methodological concerns identified and validate its effects in more rigorously designed studies. Additionally, the next step should involve evaluating its efficacy in stroke patients with somatosensory impairments.

Limitations

This review has three key limitations. First, while we aimed to discuss the effects of somatosensory training on stroke patients, this was not possible due to a lack of clinical research. Consequently, it remains challenging to conclude whether somatosensory training can improve somatosensory deficits in stroke patients. Given that moderate effects have been observed in healthy individuals, we hypothesize that somatosensory training may address somatosensory deficits caused by stroke. Second, the aftereffects of somatosensory training reported in the review are limited to 24 h. This makes it difficult to determine whether the observed effects are temporary or long-lasting. One study has explored the aftereffects of somatosensory training over 24 h, reporting sustained effects within the first 2 h [1]. However, it remains unclear whether these aftereffects are clinically meaningful for improving somatosensory deficits. Further research is needed to assess the long-term impact of somatosensory training and its potential clinical relevance for addressing somatosensory impairments. Third, in the systematic review and meta-analysis, an electronic search was conducted using the terms “perceptual learning” or “somatosensory learning.” The terms “electrical stimulation” and “mechanical stimulation” were not included as search keywords. Given that thousands of studies using these terms were identified in the search, a comprehensive investigation including these terms could potentially yield new insights. However, screening all of them was not feasible, so this was not pursued in the current review.