Introduction

Memory for episodes of experience (episodic memory) depends critically on the hippocampus and interactions of hippocampus with a set of brain regions that comprise a hippocampal-centered network (1–8). Episodic memory impairments due to a variety of neurological and psychiatric conditions have been associated with disrupted functional connectivity of hippocampal-centered networks, as measured via methods such as functional MRI (3, 4, 6, 9–14). Thus, to the extent that memory impairments result from disrupted network function, modulation of hippocampal network functional connectivity is a potential therapeutic strategy for episodic memory rescue.

Transcranial magnetic stimulation (TMS) induces electrical fields in the human brain with sufficient intensity to trigger focal action potential firing in neocortex (15). Repetitive TMS (rTMS) protocols influence neuroplasticity of distributed networks of the stimulated location, which can influence functional connectivity (16–21). This motivates the hypothesis that rTMS of hippocampal network locations could affect network functional connectivity and thereby influence episodic memory. Indeed, Wang et al. (8) applied multi-day rTMS to a parietal-cortex location identified via its high fMRI connectivity with the hippocampus. They reported that stimulation increased fMRI connectivity of the indirectly targeted portion of hippocampus with its network and improved episodic memory performance. We refer to this stimulation method as “Hippocampal Indirectly Targeted Stimulation” (HITS), as the overall goal is to use noninvasive rTMS to indirectly affect hippocampal interactions with its network. It is possible that HITS could be used to enhance memory function, both in healthy individuals and those with clinical memory impairments resulting from abnormal hippocampal network connectivity (22–24). However, the variety of neuromodulation protocols used by various research groups and variability in reported effectiveness of HITS (25–28) motivate systematic and quantitative evaluation of this potential.

Here, we report results from a systematic review and meta-analysis of studies that investigated the effects of HITS on episodic memory. We reviewed studies in which rTMS was applied to locations of the hippocampal network (Figure 1) and that assessed effects of stimulation on objective tests of episodic memory, conducted in healthy young and older adults and in clinical populations with memory impairments. Our main analysis goal was to determine whether HITS affects episodic memory reliably across studies, despite variation in study designs. We also tested whether the effects of HITS were greater on episodic memory than on other cognitive functions measured in the same studies, and whether study design factors modulated the effectiveness of HITS. We focused on hypothesis-driven design factors, including task format, targeting method, and study population, as well as eight factors identified post hoc as varying across studies (Table S1).

Figure 1.

Results

Sample characteristics

Our search identified 47 studies meeting inclusion criteria, 38 of which provided sufficient data to support meta-analysis (Table 1). These 38 studies included N = 1,009 subjects and reported 253 statistical comparisons (i.e., “effects”) of the influence of HITS on performance of episodic-memory (140 effects) and non-memory tasks (113 effects), which we transformed into normalized effects (Hedges’ g). The mean sample size was n = 23.2 for episodic-memory effects (range = 4 – 68) and n = 23.3 for non-memory effects (range = 8 – 58) (Figure S1). By definition, stimulation was applied to locations of the hippocampal network that were predominantly in left lateral parietal cortex (84% of studies) or in precuneus or other parieto-occipital neocortex locations generally considered as part of the hippocampal network (all remaining studies) (Figure 1).

Table 1.

HITS improved episodic memory performance

A meta-regression model with a random effect by study was used to evaluate the overall effect of HITS on episodic memory performance for the 140 effects derived from episodic-memory tasks. The overall effect was positive and highly significant, indicating that HITS improved episodic memory (Hedges’ g = 0.44; 95% CI [0.34, 0.54]; p < 0.001; Figure 2A). A similar result was obtained when outliers were removed in a sensitivity analysis (Hedges’ g = 0.38; 95% CI [0.29, 0.46]; p <0.001). Both the main analysis and sensitivity analysis showing lower bounds well above conventional thresholds for small effects (g ≥ 0.2) indicates robust evidence(29) for meaningful improvement in episodic memory performance due to HITS.

Figure 2.

Effects of HITS were greater on episodic-memory than non-memory tasks

Meta-regression was used to separately evaluate the overall effect of HITS on performance for 113 effects that concerned tasks assessing cognitive functions other than episodic memory. These included tasks designed to assess attention, working memory, executive functions, and language. The overall effect was almost zero and not significant, indicating that HITS did not reliably affect performance in these tasks (Hedges’ g = 0.04, 95% CI [-0.01, 0.09], p = 0.12; Figure 2B). A similar result was obtained when outliers were removed in a sensitivity analysis (g = 0.01, 95% CI [-0.05 0.06]; p = 0.79).

To more rigorously assess whether effects of HITS for tasks that measured episodic memory were significantly greater than for non-memory tasks, we ran a model pooling all 253 memory and non-memory effects and assessing overall moderation by task type. Effects for non-memory tasks were significantly less than for episodic-memory tasks (Hedges’ g modification = −0.39; 95% CI [−0.48, −0.29]; p < 0.001). A similar result was obtained when outliers were removed in a sensitivity analysis (Hedges’ g modification = −0.34; 95% CI [−0.43 - 0.25]; p < 0.001). Thus, the beneficial effects of HITS were significantly greater for tasks measuring episodic memory than other cognitive domains in both the main and sensitivity analyses.

Study factors modulated HITS effects on episodic memory

We categorized episodic-memory effects according to study design factors, based on a priori hypotheses as well as post hoc considerations of variability in study designs (Table S1). We evaluated which of these factors had the potential for independent modulation of HITS effects on episodic memory via cross-validated lasso regression selection with permutation. Factors that were retained in greater than 50% of the lasso iterations were included in the meta-regression. This led to the exclusion of the TMS Protocol factor, likely because there was strong redundancy of this factor with the Intensity and Sessions factors (Figure S2). As expected, the 10 retained factors were well represented among analyzed effects (Figure 3). In the sensitivity analysis with outlier removal, TMS Protocol was also discarded, as well as the Task Delay factor. Three of the retained factors significantly modulated the effects of HITS on episodic memory performance (Figure 3A), as described next.

Figure 3.

Consistent with our a priori hypothesis, effects for tasks that measured episodic memory using recollection test formats were significantly greater than for those that used recognition formats (Hedges’ g modification = 0.22 greater for recollection versus recognition; 95% CI [0.06, 0.37]; p = 0.006; Figure 3A). In contrast, effects for tests that used other episodic memory (non-recollection and non-recognition) formats were not significantly different from recognition (Hedges’ g modification = 0.07 less for other-format versus recognition; 95% CI [-0.33, 0.20]; p = 0.62; Figure 3A). The same pattern held in the sensitivity analysis with outliers removed (Hedges’ g modification = 0.19 greater for recollection than recognition; 95% CI [0.54, 0.33]; p = 0.006; Hedges’ g = 0.02 less for other-format than recognition; 95% CI [-.25, 0.21]; p = 0.84). Thus, among tasks assessing episodic memory, the effects of HITS were significantly greater for recollection.

Regarding the Timing factor, HITS applied before the period of memory encoding (from days to seconds before) was associated with significantly greater effects on episodic memory than HITS applied immediately after the period of memory encoding (Hedges’ g modification = 0.69 greater for pre-encoding versus post-encoding; 95% CI [0.34, 1.05]; p < 0.001; Figure 3A). The same pattern held in a sensitivity analysis with outliers removed (Hedge’s g modification = 0.61 greater for pre-encoding versus post-encoding; 95% CI [0.33, 0.89]; p < 0.001). The average effect for studies in which HITS was applied post-encoding was nearly zero (Figure 3A), with these being the majority (52.4%) of all observed negative effects (i.e., effects reflecting impairment by HITS). In contrast, effects when HITS was applied pre-encoding were almost all positive (91.5% of pre-encoding effects were positive). Notably, studies in the post-encoding category all followed the same general experiment design, whereby a brief encoding period was immediately followed by a single session of stimulation which was then immediately followed by the test. In contrast, pre-encoding studies included a variety of designs, including those in which brief stimulation trains were given immediately before individual memoranda as well as those in which many days or weeks of stimulation were given for days to weeks before memory encoding (Table S1). Thus, among tasks assessing episodic memory, the effects of HITS were significantly greater when it was applied sometime before the period of memory encoding.

Regarding the Targeting factor, effects differed for studies that used individualized (i.e., subject-specific) measures of functional or structural MRI connectivity to determine the stimulation location for HITS compared to studies that used a similar target for all subjects based on an atlas or a priori hypothesis (Hedges’ g modification = −0.32; 95% CI [-0.58, −0.07]; p = 0.013; Figure 3A). However, this effect was not significant in a sensitivity analysis with outliers removed (Hedges’ g modification = −0.19; 95% CI [-0.38, 0.01]; p = 0.059). Thus, evidence for modification of HITS effects by targeting method was weak, not surviving sensitivity analysis.

None of the other study factors significantly modulated the effects of HITS in either the main or sensitivity analyses (Figure 3B).

Of the 140 effects concerning episodic memory, 17.9% involved both the two significant positive moderating study factors identified in the main and sensitivity analyses (recollection test format and application of HITS before encoding). These effects thus can be considered as reflecting optimized experiment designs. To estimate the expected effects for future studies incorporating both of these factors, we used our meta-regression model to project the expected effect sizes for studies incorporating these factors while accounting for variance associated with other factors, held constant. Their mean effect estimate was g = 0.66 (95% CI [0.38, 0.94]), in contrast to that of all other effects incorporating non-optimized factors of all other effects, g = 0.37 (95% CI [0.07, 0.68]) (Figure 4).

Figure 4.

Similar outcomes in studies that were excluded from meta-analysis

Nine studies identified via systematic review were not included in the meta-analysis for reasons summarized in Supplementary Results. Excluded studies reported outcomes from 209 subjects and incorporated study design factors representative of those for included studies (Supplementary Results). Seven of the excluded studies reported results from statistical tests to assess the effects of HITS on episodic memory performance. Of those, five reported statistically significant improvement due to HITS in at least one test of episodic memory and two reported numeric improvements due to HITS that were not statistically significant (Supplementary Results).

Thus, excluded studies mostly reported episodic memory improvements due to HITS, with variability in outcomes that is consistent with the variability seen in studies that were included in the meta-analysis. It is therefore unlikely that meta-analysis results would vary significantly if it were possible to include these studies.

No serious adverse events were reported

None of the reviewed studies reported any unexpected or serious adverse events. The total sample reviewed included N=1,238 subjects, including healthy younger and older adults, individuals with mild to moderate Alzheimer’s dementia, and individuals with psychiatric symptoms. This suggests that the incidence of serious adverse events due to HITS is low, which is consistent with the general evidence for very low incidence of serious adverse events due to rTMS (30).

Discussion

Enhancement of episodic memory in humans is an important scientific objective with clear implications for the treatment of clinical memory disorders and for enhancing function in healthy individuals, including older adults experiencing normative memory decline. However, interventions for memory enhancement have been elusive. Our findings indicate HITS is a promising method to achieve memory enhancement via noninvasive stimulation. There were similarly positive effects among participant populations, from young healthy adults to those with memory disorders, and for a wide range of assessment delays, from seconds to weeks after subjects received HITS. In contrast, there was essentially no effect on non-memory functions, suggesting that memory enhancement did not come at the cost of any off-target cognitive detriment. There were no reported unexpected or serious adverse events. Findings suggest that HITS is safe and effective for targeted enhancement of episodic memory.

Many have speculated that brain stimulation targeting a given network can have specific effects on the cognitive abilities hypothesized to depend on that network (17, 25, 26, 31, 32). Although aspects of this hypothesis for HITS have been supported in isolated experiments (8, 33, 34), the current findings are notable in providing meta-analytic support. The hippocampus and its network are strongly implicated in episodic memory, and particularly in tests that that measure recollection compared to recognition (35–37). Indeed, we found that effects of HITS were specific to episodic-memory versus other cognitive tests, and were greater for recollection than recognition or other-format memory tests. Although previous individual experiments have shown that HITS affects activity of the hippocampal network in relation to recollection improvements more so than recognition (23, 34, 38), the current findings provide strong evidence that across many different experiment parameters, stimulation targeting the hippocampal network specifically affects the memory function thought to depend most heavily on this network.

Specificity of HITS effects on recollection is mechanistically informative. For instance, other interventions such as transcranial electrical stimulation yield nonspecific effects across a variety of cognitive domains (39). It is possible that such nonspecific effects could arise via modulation of general non-targeted factors, such as arousal via effects of stimulation on either peripheral or central nervous systems (40, 41). Such nonspecific influences could be problematic, as general and/or off-target effects could be ineffective or detrimental when applied to disorders affecting specific brain circuits. Along with neuroimaging evidence for selectivity (8, 23, 33, 34, 42, 43), the current findings suggest that HITS is suitable for targeted interventions of hippocampal network contributions to recollection. This could be useful in the treatment of memory disorders, which tend to disproportionately impact recollection relative to recognition (36). However, an important caveat is that the tests used as outcomes in the studies we analyzed were primarily designed to measure different cognitive abilities in isolation in the context of neuropsychological diagnosis. Our findings thus do not preclude off-target effects in specialized tasks that deliberately assess the interactions of brain networks. That is, enhancement of episodic memory could affect tests of other cognitive domains when those tests are at least partially sensitive to, or compete with, episodic memory (44).

It is notable that HITS had little to no effect in experiments that delivered a single session of HITS immediately after an encoding session and before the corresponding memory test (Figure 3A, HITS Timing factor of Post-Encoding). In contrast, effects were robustly positive when HITS was applied at some point pre-encoding, across a variety of potential stimulation-encoding delay intervals. This suggests that HITS improves memory formation rather than retention of recently learned information or memory retrieval. This is consistent with findings that hippocampus makes distinct contributions to learning versus retrieval (45). Regarding HITS as an intervention, it is notable that effects were similar across a variety of delays between when HITS was applied and when memory encoding occurred (Figure 3B, Task Delay factor). Unlike potential interventions that require delivery of stimulation concurrent with cognitive performance (e.g., transcranial electrical stimulation) or with highly specific information about the timing or successfulness of specific memory events (e.g., closed-loop deep-brain stimulation), this finding suggests that HITS does not require specific knowledge of when memory demands occur following intervention. This is of practical utility, as it is difficult to know a priori when memory encoding will be required during ongoing activities of daily living.

It is important to avoid over-interpretation of null findings concerning moderating effects of some study factors (Figure 3B), as many were correlated (Figure S2) and were selected in individual experiments based on a priori hypotheses. For instance, based on evidence that stimulation for more consecutive sessions of HITS leads to greater or more persistent effects (46), studies that included more HITS sessions also tended to measure outcomes after longer delays (correlation of Sessions and Task Delay factors; Figure S2). Thus, it is possible that studies with more stimulation sessions would have yielded greater effects had they used shorter testing delays, as in the studies with fewer stimulation sessions. Likewise, the Intensity factor was complicated in that studies using lower intensity were more likely to involve theta-burst stimulation than studies using higher intensity, which typically used high-frequency rTMS. These studies were based on hypotheses that hippocampal networks should respond to theta-burst stimulation more than high-frequency stimulation (25, 43). But, theta-burst stimulation typically used lower intensity and also fewer pulses of stimulation than high-frequency rTMS. Thus, no differences in outcome based on the Intensity factor thus could support the hypothesis that theta-burst stimulation is more effective than rTMS, as similar effect sizes resulted with fewer and less-intense stimulation pulses. Interpretation of null effect modulation by factors that are highly correlated with other factors is thus ambiguous. Although it was useful to identify those factors that robustly modulated effects of HITS via meta-analysis (Figure 4), future experimental studies will be needed to systematically assess whether effects of HITS vary by study design (Figure 3B), based on specific manipulation of individual factors while holding other factors constant.

A limitation of our approach is that we focused on studies applying HITS to lateral and medial parieto-occipital neocortex targets of the hippocampal network. Other prominent locations of this network include medial frontal and lateral temporal neocortex (47), although no studies targeting these locations were identified by our review, potentially because conventional TMS of these locations is not well tolerated. Notably, the hippocampal network includes portions of lateral cerebellar cortex (48), and one study identified by our review found episodic memory improvement following theta-burst TMS of this location (49). Theta-burst stimulation of lateral cerebellum has also been shown to affect hippocampal network fMRI connectivity (50). However, this study was excluded from meta-analysis to avoid heterogeneity in stimulation locations. Future work could systematically evaluate alternative stimulation targets within the hippocampal network, including those that may require stimulation methods other than TMS.

The current findings suggest that HITS has robust effects on episodic memory and that studies including optimized parameters yield especially robust effects (Figure 4). However, further research is needed to test clinical efficacy for specific conditions and to determine mechanisms of action. Although our review included studies of patients with episodic memory impairments due to mild to moderate dementia and psychiatric symptoms, the studies did not assess whether gains by HITS could be maintained for months to years. Further, it is not clear whether HITS will be useful for memory impairments in other disorders, which may involve distinct mechanisms. Implementations for specific disorders will likely require better understanding of how HITS affects neural function, such as could be obtained by directly recording hippocampal activity while HITS is applied (43, 51), in order to tailor HITS to target disorder-specific neural impairments. The current findings of HITS efficacy, specificity, and safety encourage such future explorations.

Materials and methods

Eligibility criteria

We included reports published between September 2014 (date of the seminal HITS publication (8)) and April 2025 that: (1) used rTMS to stimulate hippocampal network locations and (2) included at least one objectively scored test of episodic memory as an outcome. Studies were considered as stimulating hippocampal network locations if they delivered rTMS to: (1) an individualized/subject-specific location defined by fMRI connectivity analysis (e.g., a neocortical location identified by resting-state fMRI connectivity to hippocampus in an individual), (2) a location based on a group metric of hippocampal fMRI connectivity (e.g., a neocortical location identified in a group-based atlas of hippocampal resting-state fMRI connectivity), and (3) a similar location based on a hypothesis (e.g., a location that is in a group atlas of hippocampal connectivity but that was selected to test a hypothesis about its function). We limited spatial variability among studies by including only those that stimulated parietal and superior occipital neocortex. We excluded studies using other potential hippocampal network locations due to their rarity, including one study with cerebellar cortex stimulation (49).

Studies were considered as including an objective test of episodic memory if at least one of the reported outcomes used a memory test that allowed unambiguous computation of the effect of stimulation on performance accuracy (i.e., experimenter-controlled stimulus set presented to subjects for encoding, for which memory was subsequently tested after some retention interval).

Search strategy

Two publicly available online scientific databases were searched: PubMed (pubmed.ncbi.nlm.nih.gov) and Scopus (scopus.com). These databases were most recently searched on April 26, 2025. The primary search was conducted with PubMed, using the PubMed Advanced Search Builder. Two search terms of interest were entered: “memory” and “transcranial magnetic stimulation”. These terms were selected to be as broad as possible while aiming to exclude studies utilizing stimulation techniques (invasive and non-invasive) other than TMS. These terms were added to the Query Box using ‘AND’ Boolean operators; all possible fields (i.e., title, abstract) were searched. PubMed automatically included a search for similar terms for each of the terms of interest entered (i.e., ‘memory’ and ‘memories’), and included the Medical Subject Headings (MeSH) terms associated with them. This search returned 104 results. A secondary search was conducted using Scopus after completing the PubMed search, using the same primary search terms using ‘AND’ Boolean operators and constraining fields to Article title, Abstract, and Keywords and limiting results to peer-reviewed empirical reports by limiting the “Document Type” option to “Article”. This search returned 160 results. We also examined the references of publications identified as meeting the eligibility criteria for suggestions about other relevant publications. Publications were selected from the search results when two reviewers (PA and JLV) agreed that they met the two eligibility criteria. Additionally, only studies in human participants were selected.

Outcomes extraction

Following review of the full texts, we extracted data from statistical tests aiming to quantify the effects of rTMS on episodic memory and non-memory outcomes. Extracted data included sample sizes, means, and standard deviations of the outcome variables for experimental and control measurements, as well as results from reported statistical tests. These data were used to calculate effect sizes, using the test statistics or group means and variances following established methods (52). Cohen’s d was used to measure the effect size of pairwise comparisons. Partial eta squared was used to measure the effect size of comparisons using analysis of variance or other general linear models. We did not include effects from tests that provided “composite” measures spanning multiple cognitive domains (e.g., Montreal Cognitive Assessment, Mini-Mental State Examination, etc.) as these could not be readily classified as reflecting episodic-memory versus various non-memory functions. In tasks with multiple subscales (e.g., Wechsler Memory Scale, NIH Toolbox for Cognition, etc.), individual subscale scores were used and categorized based on the task format (e.g., recall versus recognition) and/or by the intended cognitive domain (e.g., episodic memory versus sustained attention). Effect sizes were computed such that improvement in performance (i.e., higher accuracy, faster response time in speeded non-memory tasks, or fewer errors) were positive and impairments (i.e., lower accuracy, slower responses, or more errors) were negative. Prior to input for meta-regression, all estimates of effect size were converted to Hedges’ g to minimize potential impact of inflated effect sizes associated with small sample sizes (52). Within and between estimation uncertainty around each Hedge’s g were approximated using its corresponding study effective sample size.

Meta-analysis of effect size

A mixed effects univariate meta-regression model estimated via restricted maximum likelihood was used to estimate the overall effect of rTMS on outcomes, pooling all effects reported in the final sample of studies collected at the outcome level and weighting them in terms of their within and between estimation uncertainty. We included a study random effect in the model to account both for inter and intra study variability in our final estimate, as a single study often reported various effect sizes for different outcomes. Heterogeneity in individual effect sizes was examined using a generalized Q-test, to assess whether variability in observed effects exceeds what would be expected from random sampling error alone (53, 54).

Meta-analysis of effect modifiers

Factors that varied among studies were identified based on a priori hypotheses and post hoc consideration of study design details that could influence outcomes (Table S1, Figure 3). We explored the association of effect size with these suspected modifiers by estimating a multivariate meta-regression model including all study factors. As with the main effects model, we pooled all effects reported in the final sample of studies collected at the outcome level and weighted them in terms of their within and between study estimation uncertainty. Since some study characteristics were a priori expected to be highly correlated with one another (e.g., theta-burst stimulation is typically performed at lower intensity than high-frequency rTMS), we performed a previous-variable-selection step before assessing effect modification. Indeed, some factors were highly correlated, as expected (Figure S2). We therefore used a bootstrapped (1000 iterations) lasso model with a cross-validated penalization term to identify study characteristics that were jointly consistently redundant in terms of statistically explaining HITS effects (i.e., discarded by lasso with a probability larger than 50% across all iterations), to be then excluded from the effect modification analysis. Study factors retained after the lasso variable selection were then simultaneously included in the multivariate mixed effects meta-regression, where their association with stimulation effects was estimated. Heterogeneity in individual effect sizes after accounting for effect modifiers was examined using a generalized Q-test, assessing whether significant variability in observed effects remained, even after accounting for the variation explained by included study characteristics (Q(df = 136) = 307.07, p < 0.001) . Analysis for this step was conducted using the glmnet package in R. All meta-analytic estimates were computed using the metafor package in R (54).

Outlier removal

To guard against the impact that unusually strong reported effects could have had in our findings, we carried out all meta-analyses, including the lasso variable selection step, after removing outliers identified using the studentized deleted residuals (55). We identified and excluded seven effect outliers in the analysis of episodic memory outcomes, leaving 133 of 140 effects for that analysis, and seven effect outliers in the analysis of non-memory outcomes, leaving 106 of 113 effects for that analysis. Categorical variables were included in final models if at least one level had a coefficient with magnitude ≥ |0.05| or their aggregate coefficients had magnitude ≥ |0.01|. Variables with all coefficients < |0.01| were not included. We report findings with and without removal of outliers, which had little impact.

Reporting bias assessment

Publication bias was assessed with a funnel plot of individual effect sizes against their standard errors (Figure S3 and S4). Special attention was paid to the asymmetry of the plot and the absence of effects in the lower-left corner of the funnel, which indicates the lack of published small-sample studies reporting negative effect sizes, further confirmed by an Egger’s regression test (p <0.001), which is commonly carried out to detect evidence of publication bias (56). We tested for heterogeneity (I²) to measure the inconsistency of effect sizes across studies within each cognitive domain. The I² statistic represents the percentage of variation in effect sizes beyond a random sampling error (57). Although the interpretation of heterogeneity can depend on multiple factors, an I² of 25 to 75% is typically considered moderate, whereas I² > 75% is considered a substantial heterogeneity (57). Further, we tested for the presence of outliers, and removed them in sensitivity analyses to assess the robustness of our main findings to outliers. There was no conclusive evidence of publication bias, with I² of 0.53 for episodic memory tests and of 0.0 for non-memory tests (Figure S3 and S4) and with replication of major findings in the sensitivity analyses.

Supplementary Materials

Figure S1.

Figure S2.

Figure S3.

Figure S4.

Table S1.

Supplementary Results

Summary of findings from studies that were not included in the meta-analysis

As indicated in the main text, nine studies were excluded from meta-analysis (Table 1). Five of these exclusions were because statistical information provided in the publications were not sufficient for calculation of standardized effect sizes, and authors could not be reached to provide additional information (studies 39, 42, 44, 45, 47 from Table 1). One exclusion (study 40) was made because the sample of subjects analyzed overlapped 94% with subjects that contributed data to a previous publication that was included, and so exclusion was necessary to avoid redundant statistical results in the meta-analysis. Two studies were excluded because they did not report the effects of HITS on performance of an episodic memory test, but instead quantified the correlation between fMRI connectivity and the effects of HITS on episodic memory performance (studies 43 and 46). One study was excluded (study 41) because re-analysis of the same data by the same research team (study 43) reported a different conclusion than the original report, suggesting that stimulation may not have been delivered to intended hippocampal network targets in the original report, and thereby calling into question findings from the original report and making it unclear how to categorize the Targeting or Target factors for meta-analysis.

Of the nine excluded studies, seven reported the effects of HITS on episodic memory performance (studies 39, 40, 41, 42, 44, 45, 47). Five of these seven reported that HITS significantly improved performance in at least one of the episodic memory tasks that were administered (studies 39, 40, 42, 44, 45). One of those studies (study 44) also reported numeric but non-significant decrease in memory performance in a subgroup analysis of subjects with mild Alzheimer’s dementia. Two of the seven studies (41 and 47) reported small numeric improvements in memory following HITS that were not statistically significant.

Regarding the study factors that modulated the effects of HITS in the main study (Figure 3A), the seven studies all all included at least one test using a recollection format (Task factor: Recollection) and all administered HITS before encoding (Timing factor: pre-encoding). Two used individualized targeting (studies 40 and 41), with the others using group/atlas-based targeting. Other factors that were not significantly related to memory outcomes following HITS in the main analysis (Figure 3B) varied among these seven excluded studies.

The other two excluded studies (43 and 46) investigated correlations between resting-state fMRI connectivity and the effects of HITS on episodic memory, without directly assessing the impact of HITS on episodic memory. One reported that fMRI connectivity of the hippocampus to the stimulated location predicted episodic memory improvement due to HITS (study 43). The other study reported that HITS changed the correlation between fMRI connectivity and memory performance scores (study 46).

Overall, reported results in the excluded studies were generally consistent with findings from the meta-analysis of included studies. That is, of the seven excluded studies that reported effects of HITS on episodic memory, the majority (5/7) reported statistically significant episodic memory enhancement by HITS, but with variability in effect sizes among studies and a minority (2/7) reporting small but statistically nonsignificant changes in memory due to HITS. Thus, if effects from these studies could have been included in the meta-analysis, it is unlikely that conclusions would have been substantially affected.

Data availability

Upon acceptance, the primary data will be deposited in a publicly available database (Dryad, datadryad.org) for use by other researchers without any restrictions on access or use. The data made available will include a table of all standardized effect sizes organized by study and by factor, such that reported analyses could be replicated.

Additional information

Author contributions

Conceptualization: EBG, PFA, JLV Methodology: EBG

Investigation: PFA, JMHR, ASA, JLV Writing – original draft: EBG, PFA, JLV

Writing – review & editing: EBG, PFA, JMHR, ASA, JLV

Funding statement

This study did not receive any funding.