Effects of anodal transcranial direct current stimulation on short-term memory and delayed recall in patients with memory impairment due to progressive and non-progressive brain diseases: a meta-analysis

Introduction

Memory disorders are among the most disabling problems in patients with neurological diseases. Memory deficit is a hallmark symptom of Alzheimer’s disease (AD), with a dominating memory-loss profile in ~70% of patients (Reed et al. 2007). The frequency of various degrees of memory deficits is around 70% after traumatic brain injury (TBI) (Hux et al. 2009;), ~45% in mild cognitive impairment (MCI) (Manly et al. 2008), ~40% in encephalitis (Pewter et al. 2007), 13-24% in patients with stroke (Middleton et al. 2014), and ~10% in patients with Parkinson disease (PD) (Dujardin et al. 2015).
Since the currently used therapeutic methods have established limited effectiveness (Caramelli et al. 2022; Cicerone et al. 2019; Petersen et al. 2018), it has been postulated that modulation of activity of memory-involved neural networks may be an alternative or complementary approach. Among all transcranial electrical stimulation (tES) techniques, transcranial direct current stimulation (tDCS) is the most frequently used method of neuromodulation (Antal et al. 2017) due to its safety, tolerability, ease of application, and low cost (Paulus 2011). tDCS originates from the research of Nitsche and Paulus (2000), who observed that the application of a weak (up to 2 mA), direct current polarizing tissue over the cerebral cortex of animals caused a moderate change in cortical excitability. The direction of this change appeared to be dependent on the polarity of the electrodes; anodal stimulation increased the level of excitability, while cathodal stimulation decreased the excitability of the neural circuits located under the electrode.
Studies on healthy volunteers have shown that tDCS was associated with changes in motor, sensory, and cognitive functions (Stagg et al. 2009; Stagg and Nitsche 2011). The presence of changes in excitability and their functional/behavioural consequences was confirmed not only during the procedure (intra-stimulation effect), but also after its completion (post-stimulation effect) when the stimulation lasted at least three minutes (Nitsche et al. 2008; Nitsche and Paulus 2000). Increasing the intensity of the current (up to 1 mA) and the time of its delivery (up to 13 minutes) resulted in the extension of the post-stimulation effects, even to over an hour (Nitsche and Fregni 2007). Increased cortical excitability by anodal tDCS (a-tDCS) is usually explained by the phenomenon of subthreshold neuronal depolarization, which augments the likelihood of spontaneous neuronal firing. Cathodal tDCS (c-tDCS) seems to have the opposite hyperpolarization effect, leading to excitability reduction (Stagg and Nitsche 2011).
Currently, either unilateral two-channel stimulation is utilised (with the active electrode positioned above the target of stimulation and the other – cephalic or non-cephalic – functioning as the reference electrode), or multi-channel stimulation is employed (involving one active electrode and several reference electrodes), or bilateral tDCS is implemented for research purposes (involving two active electrodes and one or two reference electrodes). Multi-channel tDCS additionally enables more focused stimulation in terms of electric charge concentration in the cerebral cortex (Bikson et al. 2019). However, such concentration is not always desired, especially when the target function has a widespread neuronal base, as in the case of the memory system. Also, the choice of stimulation site depends on researchers’ assumptions regarding the specific effects of tDCS on particular memory processes. The most frequently targeted area is the dorsolateral prefrontal cortex (DLPFC) (Ladenbauer et al. 2017), but the temporal cortex (TC) (Kazuta et al. 2017) and the parietal cortex (PC) (Ferrucci et al. 2008) have also been chosen by experimenters.
At the neurophysiological level, it appears that tDCS alters the way the fronto-parietal networks respond to increased cognitive load, similarly to repetitive transcranial magnetic stimulation (rTMS; another popular neuromodulation method). The P300 amplitude was found to be increased when high-frequency rTMS or a-tDCS was applied, and reduced when low-frequency rTMS and c-tDCS were applied (Mendes et al. 2022; Liu et al. 2021). Cespón et al. (2019) found positive correlations between increased frontal P300 amplitude in healthy elderly and enhanced accuracy in n-back performance after a-tDCS. Taken together, recent evidence suggests that tES over DLPFC can enhance working memory (WM) efficiency in humans, and that this enhancement is likely due to neurophysiological effects (including increased excitability and stochastic resonance), which may mediate other processes involved in formation of memory traces.
The effects of stimulation are specific to its polarity, as evidenced by the results of the study by Javadi et al. (2012), in which an improvement in accuracy and reaction time in a memory task immediately after the use of a-tDCS and a worsening of performance after the use of c-tDCS were noted. Behavioural changes were also accompanied by changes in the activity of the cerebral cortex during memory tasks. As shown in the study by Zaehle et al. (2011) on a group of healthy participants, a-tDCS and c-tDCS differently affected such parameters as the latency of cognitive evoked potentials, N100 and P300.
Although the effects of tDCS on memory performance in individual studies are encouraging, meta-analyses of experimental studies provide mixed results. No effects of multi-channel (so-called high-definition, HD) tDCS on healthy participants’ WM were found in the meta-analysis by Müller et al. (2022). On the other hand, Hill et al. (2016) found subtle (Cohen’s d = 0.15) but significant effects on WM in favour of a-tDCS, as compared to the sham in healthy participants, as well as moderate effects in neuropsychiatric populations, but only for online protocols. However, the data came from only two studies.
In a meta-analysis of 24 studies, Huo et al. (2021) found moderate effects of a-tDCS on episodic memory, both in cognitively healthy and impaired older adults. Cruz Gonzalez et al. (2018), who performed a meta-analysis of four trials in patients with memory impairment in dementia, found a statistically significant improvement with a-tDCS over DLPFC, but the effect size (ES) was small. Chen et al. (2022), in their recent meta-analysis, reported a significant ES on general cognitive functioning in patients with AD and MCI, but did not find any effects of a-tDCS on verbal memory. Similarly, Šimko et al. (2022) found a small ES on overall cognition, and only speculated that WM improvement contributed to this result.
As regards non-progressive brain diseases, two recent meta-analyses (Galimberti et al. 2024; Ahorsu et al. 2021) revealed small to moderate effects of rTMS and a-tDCS on general cognitive performance in people with TBI. However, it is not clear how a-tDCS alone contributed to the overall ES, and which specific functions improved. Similarly, Yan et al. (2020) found a large ES of a-tDCS on the general cognitive level, but did not find any significant improvement in memory (the results were based on three RCT studies).
There are several potential reasons for these inconclusive results. Mixing patients with varying types and severities of memory impairment may have obscured the therapeutic effects. Additionally, the duration and number of sessions are often not considered in meta-analyses or are included as dichotomous variables (e.g., single vs multiple sessions). Current intensity and density are frequently the only measures of the electrical stimulus, which does not properly account for the cumulative nature of tES (Antal et al. 2017). Another issue is the excessive diversity of assessment tools used in individual studies, including experimental tasks, standardized tests, and clinical scales. Mixed measures of memory (verbal vs. non-verbal, immediate vs. delayed recall) add further confusion to this picture.
In summary, while a-tDCS shows limited evidence for improving general cognition in various brain disease populations, its specific effects on memory impairment remain largely unknown. Consequently, the research objective of this project was to conduct a meta-analysis of randomized, controlled trials (RCTs) investigating the effects of a-tDCS on memory improvement in individuals with memory impairment. We hypothesized that patients with both progressive (i.e. neurodegenerative) and non-progressive conditions benefit from this type of intervention.

Material and methods

Study design and protocol registration

The protocol for this meta-analysis was registered with the International Prospective Register of Systematic Reviews (PROSPERO, registration number: CRD42023460773).

Literature search

An extensive literature search was conducted using the following databases: PubMed, MEDLINE, Cochrane Library, and NeuroBITE up to May the 1^st, 2024. After all relevant studies were retrieved, their title and abstract were screened against the inclusion and exclusion criteria. If the title and abstract alone provided insufficient information to determine whether the study could be included, the full-text version of the article was screened. One investigator selected the studies, and another verified whether all inclusion/exclusion criteria were met. In case of any disagreements, the third author was involved in the decision-making process.

Study selection

We defined the following as the inclusion criteria for the selected studies: 1) RCT study either with parallel groups or crossover design, 2) study involving adult human participants, 3) article written in English and published in a peer-reviewed scientific journal, 4) study that used a-tDCS as a means of treating cognitive impairments, at least as part of the experimental intervention, 5) study involving patients with acquired memory dysfunction due to non-progressive (stroke, TBI, or encephalitis) or progressive (MCI, AD, PD, or vascular dementia – VD) brain disease, 6) study using memory tasks as outcome measures, and 7) study using sham stimulation as a control condition.
We included studies that used standardized neuropsychological tests for the assessment of verbal or visuo-spatial memory, such as Rey’s Auditory Verbal Learning Test (RAVLT), the California Verbal Learning Test (CVLT), and the Wechsler Memory Scale (WMS). Additionally, studies that employed computerized tasks measuring episodic memory or working memory (e.g., n-back) included. These neuropsychological measures had to be able to provide either an index of short-term memory and/or an index of delayed recall. WM tasks were considered measures of short-term memory, along with typical word list or picture set learning tasks, since they all refer to similar processes of brief memory retention.
We included studies that had used both single and multiple session stimulation regimens with a variety of stimulation parameters in terms of current intensity, electrode montage (active electrode over DLPFC, TC or PC), and electrode size; however, if the authors used more than one level of intensity, only the highest level was considered in this analysis.

Data extraction

The following data were derived from the included studies: basic information about the original study, sample size, patient diagnosis and characteristics, age of participants, electrode positions, stimulation parameters (electrode size, intensity, and duration), the number of sessions, evaluation methods, whether cognitive training was introduced concurrently, and adverse events. For the quantitative analysis of a-tDCS effects on both short-term memory and delayed recall, measures were chosen separately. If the study used multiple memory measures of one type of memory, the primary outcome measure was selected. For studies that did not specify the primary outcome, we chose the outcome that was reported first, or had the greatest importance. Consequently, only one measure of short-term memory and one measure of delayed recall from a single experiment could be used in the analysis. This decision was based on consideration of the cumulative effect of potential methodological bias on the overall ES. If the study design assumed more than one experiment, these experiments were analysed separately. Also, only short-term post-stimulation effects were taken into account, which means that follow-up measurements were not considered in this meta-analysis.
To calculate ES for an experiment, the following data were extracted for an experimental and a control group: means and standard deviations (calculated from a standard error if needed) from both pre-treatment and post-treatment assessments and the mean score change and its standard deviation. If the latter was not reported, a pooled standard deviation was derived from F or t statistics following the guidelines of Higgins et al. (2023). In the case of missing numerical data, a free online tool (Plot Digitizer: https://plotdigitizer.com/app) was used for data extraction from graphs, if they were available. If only the ES (Cohen’s d or η²) or z score was reported, they were adjusted in a way that ensured consistency between studies. In cases where the data were insufficient to establish the ES, the reviewers attempted to contact the study’s corresponding authors to obtain the missing information.

Quality assessment

Methodological quality assessment was conducted for the included studies, according to Higgins et al. (2023). The method used in Cochrane reviews takes into consideration the following five bias risks: 1) bias arising from the randomization process, 2) bias due to deviations from intended interventions, 3) bias due to missing outcome data, 4) bias in measurement of the outcomes, and 5) bias in selection of the reported results. In each case, the risk was described as “high”, “low”, or “some concerns”. The assessment was performed by two independent reviewers.

Data synthesis and analysis

All data were analysed in R (metafor module; Viechtbauer 2021) and in Python programming language. Effect sizes were calculated to express the difference between active and sham stimulation conditions using the standardized mean difference corrected for bias (Hedges’ g; Higgins et al. 2023), with positive values indicating an increase in memory accuracy following a-tDCS. The Q statistic and the I² index were used to assess the variance of ESs across studies. Given significant heterogeneity, random-effects models were applied to calculate the overall ES in the meta-analysis. To address the probability of publication bias, both qualitative and quantitative techniques, including the funnel plot, Egger’s regression intercept test, and the rank correlation test, were employed. The trim and fill nonparametric data augmentation method (Duval and Tweedie 2000) was used to further explore publication bias.
Given the fact that substantial heterogeneity existed in the current and similar studies (Müller et al. 2022), additional moderator analyses were conducted. The latter included: 1) population: non-progressive (stroke, TBI) vs. progressive (MCI, AD, PD, VD) disease, 2) stimulation target area: prefrontal (left or right DLPFC) vs. posterior (left or right TC or PC), 3) stimulation side: bilateral vs. left vs. right, 4) current density (mA/cm²): < 0.06 vs. ≥ 0.06, 5) charge density (C/m²): as a continuous moderator, reflecting the cumulative effect of a-tDCS resulting from the number of sessions, their duration, and electric field intensity taken together, 6) concurrent cognitive training – present vs. absent, 7) study design: parallel groups vs. crossover study, and 8) experimental (active) group size: as continuous moderator. Also, separate meta-analyses were conducted for two major populations, i.e., non-progressive and progressive diseases.

Results

The study selection process is illustrated in Figure 1. From the initially retrieved 161 publications, 22 studies (23 experiments) were included in the meta-analysis. An overview of the included studies can be found in Table 1. One study (Yun et al. 2015) was represented twice in the database, since it reported two separate experiments. Twenty-one experiments provided short-term memory indices, and twelve experiments were used in the analysis of delayed recall. The former included nine experiments with non-progressive brain disease groups (no study on patients with encephalitis was found), and twelve experiments with progressive brain disease groups. The delayed recall analysis included four experiments on patients with non-progressive diseases, and eight experiments on patients with progressive diseases.

Risk of bias

A detailed analysis of the risks of biases for each domain and for each study can be seen in Table 2. Overall, there were some concerns of a risk of bias. This risk was primarily associated with the absence of a double-blinding procedure – six (27%) of the RCTs were single-blinded. Post-hoc power estimation (using G*Power software; Faul et al. 2007) revealed that, on average, studies focusing on both short-term memory and delayed recall were underpowered (0.33 ±0.28 and 0.27 ±0.19, respectively).

Analysis of effect of a-tDCS on short-term memory

The results indicated that a-tDCS resulted in significant (p < 0.001) improvement in short-term memory (Fig. 2), with an overall Hedges’ g = 0.58 (95% CI = 0.27-0.88). The heterogeneity of studies was high: Q(20) = 58.508, p < 0.0001, I² = 69.78%. The analysis of moderators revealed that only bilateral stimulation was significantly (p = 0.045) associated with the overall ES (Hedges’ g = 0.71). For the target area (prefrontal cortex) the contribution to this effect was not significant (p = 0.057). Other moderators did not contribute to the above-mentioned heterogeneity.
Both Egger’s regression (z = 3.031, p = 0.002) and the rank correlation test (τ = 0.562, p < 0.001) suggested an increased risk of publication bias. The visual inspection of the funnel plot (Fig. 3A) confirmed asymmetry of the data. However, the application of the trim and fill technique did not impute any additional studies, nor did it change the overall effect.
The analysis of subgroups revealed a statistically significant (p < 0.001) effect in groups with progressive brain diseases: Hedges’ g = 0.4 (95% CI = 0.18-0.63) with non-significant heterogeneity: Q(11) = 17.66, p = .09, I² = 6.15% (Fig. 2). Effect size for non-progressive disease groups was also significant, but with a wider confidence interval (Hedges’ g = 0.72, p = 0.024, 95% CI = 0.1-1.35) and significant heterogeneity [Q(8) = 39.513, p < 0.001; Fig. 2].

Analysis of effect of a-tDCS on delayed recall

The delay in memory tasks across analysed studies ranged from 20 minutes to 48 hours. The results indicated that ESs of a-tDCS for all populations were statistically significant (Hedges’ g = 0.45, p < 0.001, 95% CI = 0.23-0.67), and the heterogeneity was low: Q(11) = 6.814, p = 0.814; I² = 0.0% (Fig. 4). None of the moderators proved to be significant. Neither the funnel plot (Fig. 3B) inspection nor Egger’s regression analysis indicated significant asymmetry of data (z = –0.021, p = 0.983). Similarly, the rank correlation test was found to be non-significant (τ = –0.091, p = 0.737).
The analysis of subgroups revealed a statistically significant (p < 0.001) ES in groups with progressive brain diseases (Fig. 4): Hedges’ g = 0.45 (95% CI = 0.19-0.72). The studies demonstrated homogeneity: Q(7) = 5.195, p = 0.656; I² = 0.0%. The effect for non-progressive disease groups was also significant (Hedges’ g = 0.44, p = 0.032, 95% CI = 0.04-0.85), and the heterogeneity of studies was low: Q(3) = 1.616, p = 0.656; I² = 0.0% (Fig. 4).

Discussion

This meta-analysis focused specifically on the efficacy of a-tDCS for the enhancement of memory in clinical populations. The effects of the applied stimulation on both short-term and delayed memory were analysed in several populations of patients with acquired memory disorders and grouped into two categories: non-progressive brain diseases (stroke and TBI), and progressive (neurodegenerative) brain diseases (MCI, PD, AD, and VD).
Overall, when all populations were analysed together, a small-to-medium positive ES was found in short-term memory tasks. While effects on delayed memory were small, they were also significant. Both categories of patients benefited from the a-tDCS, but the positive effects were more consistent in groups with progressive brain diseases. Improvements in non-progressive groups were more variable. Moreover, the results should be interpreted cautiously due to a significant risk of publication bias found among the included studies, and their low statistical power. What makes the results credible is the fact that effects of a-tDCS on short-term memory tended to be greater in studies which implemented the protocol with bilateral stimulation and the prefrontal cortex as a target area. Given that DLPFC plays a crucial role in WM and memory encoding (Blumenfeld et al. 2011; Israel et al. 2010), this result is a logical consequence of the functional organisation of the human brain. However, other factors which are expected to modulate a-tDCS effects, such as current density, charge density, and simultaneous cognitive training, were not associated with the overall effect.
The results of this study are in line with those from the meta-analysis by Huo et al. (2021), which revealed significant effects of a-tDCS on episodic memory in patients with MCI, AD, and PD. The present meta-analysis also showed some memory benefits of a-tDCS in patients with non-progressive brain diseases. This outcome contrasts with the results of Yan et al. (2020), which indicated a lack of positive effects. However, it is important to note that our sample of studies included not only those involving patients with stroke, but also individuals with TBI. The results are consistent with the findings of Galimberti et al. (2024) and Ahorsu et al. (2021), indicating that patients with non-progressive cognitive impairments appear to benefit from a-tDCS. However, the mentioned studies did not specifically analyse the effects of a-tDCS on memory. It seems, though, that these effects are weaker than that observed in neurodegenerative groups.
The various degrees of effects in the analysed populations can be attributed to the limited number of studies, but also to differential pathological changes in cortical excitability, which may alter the way non-invasive brain stimulation influences neurophysiology. The weaker and more variable overall ES observed in non-progressive groups may be explained by a phenomenon elucidated by Li et al. (2019). According to their findings, disruptions in white matter connections within a stimulated brain network, a common occurrence in focal brain injuries such as stroke and TBI, result in a reduced behavioural response to cortical stimulation. In patients with progressive diseases such as AD, typically, pathologically increased excitability is observed, including the DLPFC (Joseph et al. 2021), which is associated with symptom severity and is probably caused by compensatory mechanisms (Bagattini et al. 2019). If that is the case, according to the results obtained by Cespón et al. (2019), who observed polarity-specific effects of tDCS in different populations of elderly participants, patients with the hyper-excitability typical for neurodegenerative diseases should benefit from c-tDCS, which decreases excitability. If the positive effects of a-tDCS applied to neural networks engaged in memory functions are true, it is possible that mechanisms other than excitability enhancement may play a role in the improvement process. These mechanisms may include noise induction leading to stochastic resonance, and consequently, an increased signal-to-noise ratio (Fertonani and Miniussi 2017). For example, the study by Murphy et al. (2020) found large beneficial effects of transcranial random noise stimulation of the left DLPFC on WM. These behavioural effects were accompanied by changes in cortical event-related responses, and were greater compared to both the sham and a-tDCS.
The observed effects of a-tDCS may also be ambiguous due to publication bias, which was highly probable in this meta-analysis, especially when short-term memory measures were considered. It is important to note that a large (n = 379) recent study on an elderly population did not find any positive effects on cognitive performance when a-tDCS was paired with cognitive training (Hausman et al. 2023). It is crucial to note that in the current meta-analysis, only five out of the 22 studies had experimental groups comprising 20 participants or more. Thus, given the overall ES, most studies were underpowered.
This study is not free of limitations. One of them is that it included a small number of studies. There are still not enough good quality studies specifically aimed at memory performance enhancement by a-tDCS in clinical populations. This consequently led to the decision to perform the analysis on a heterogeneous population.
Also, to avoid cumulative effects of methodological issues, only one (primary) outcome measure referring to each memory type was considered from each experiment. Unfortunately, not all authors explicitly identified one main memory measure. Also, some authors tended to report in detail only significantly improved results. This led to situations where the authors of this meta-analysis decided to use a measure which was described first, and not the one that was considered the main result of an experiment.
Another potential issue is associated with the chosen outcome measures. Specifically, mixing tasks of WM and episodic memory can be problematic from the perspective of memory theories. However, the fact that both constructs, particularly when involved in new information acquisition, are overlapping and often difficult to distinguish (Aben et al. 2012) supports the idea of combining these measures when considering more general effects. Moreover, we exclusively analysed performance accuracy, omitting measures such as reaction time. Our objective was to identify functionally significant improvements rather than focusing on subtle changes. A point worth mentioning in this context is the diversity of memory measures used in the analysed studies. We identified 20 different assessment tools (both standardized tests and experimental tasks) in just 22 studies. This diversity may confound the overall effect of tDCS; therefore, it would be beneficial to establish a commonly used set of measures in future research.
Another potential concern is the mixing of stimulation protocols. Different research teams investigate various parameters that are more suitable for their participant groups, and that reflect their specific hypotheses. This variability in protocols is likely to contribute to substantial heterogeneity in ES. To address this issue, in the current meta-analysis we conducted a moderator analysis to explore the potential influence of stimulation parameters.
Finally, categorizing MCI as a progressive/neurodegenerative brain disease can be questioned. While MCI may be the prodromal stage of dementia, it can also result from other pathological processes (Petersen et al. 2018). However, this was not the case in the populations studied in the included publications, as such patients were excluded by the authors of these RCTs.
This study only investigated the immediate effects of a-tDCS on memory and did not include any follow-up assessments. Future research should explore whether this intervention provides long-term benefits for patients with memory impairment due to brain damage.

Conclusions

This meta-analysis revealed that a-tDCS may have a small-to-medium positive effect on short-term memory and a small effect on delayed recall. Patients with both progressive and non-progressive brain disorders benefited from this type of stimulation. Bilateral stimulation and targeting the prefrontal cortex appears to be a more effective strategy than concentrating on posterior cortical areas when seeking to enhance short-term memory. The results should be interpreted with caution due to increased risk of publication bias and low statistical power. Thus, more high-quality, large-scale studies are needed to substantiate the therapeutic effects of a-tDCS in memory disorders.

Disclosures

This work was supported by the Faculty of Psychology, University of Warsaw, from the funds awarded by the Ministry of Science and Higher Education in the form of a subsidy for the maintenance and development of research potential in 2023 (501-D125-01-1250000 zlec. 5011000246).
Institutional review board statement: Not applicable.
The authors declare no conflict of interest.

References