Introduction

Affecting 46.8 million people globally, the dementia [i.e., Alzheimer disease (AD)] is the most common neurodegenerative disease characterized by progressive cognitive decline [1, 2]. In individuals aged between 65 and 95 years, its incidence doubles approximately every 5 years [3]. AD causes profound negative impacts on executive function and quality of life. Patients also experience memory impairment, behavioral disturbance, and insomnia [4]. Therefore, the amelioration of cognitive decline in the early stages of the disease is regarded as essential in AD management. The efficacy of AD medications, most of which are oral, do not provide satisfactory results [5, 6].

The unsatisfactory efficacy of AD leaves room for alternative treatments. Noninvasive brain stimulation (NIBS) interventions, which are based on the theory of alterations in synaptic function and neuroplasticity [7, 8] and target specific brain regions, have attracted increasing scholarly attention. NIBS techniques include repetitive transcranial magnetic stimulation (rTMS) [9] and transcranial electrical stimulation such as transcranial direct current stimulation (tDCS) [10]. According to the frequency applied, rTMS can induce different changes in brain activity. For example, high-frequency rTMS (HF-rTMS) induces higher brain activity, whereas low-frequency rTMS (LF-rTMS) suppresses cerebral cortex activity [11]. Studies have indicated that anodal [12, 13] and cathodal [10] tDCS exert stimulate and suppress activity in the targeted cortices, respectively. However, whether suppression or enhancement associated with the polarity of tDCS remains under debate [14, 15]. In AD management, rTMS or tDCS of specific cortical regions constitute promising NIBS techniques with regard to their enhancing [10] and neuroplasticity-related effects [16], as well as their effects on neurotransmitter modulation [17, 18].

Conventional pairwise meta-analyses have indicated the substantial efficacy of tDCS [19] and rTMS [20, 21] compared with that of sham treatment in improving cognitive function in patients with AD. In 2020, a recent network meta-analysis (NMA) comparing the efficacy of different NIBS techniques in improving cognitive function in patients with mild cognitive impairment (MCI) or AD suggested that rTMS was superior to tDCS and that patients with AD responded better to rTMS and tDCS than those with MCI did [22]. We identified 3 major limitations from this study. First, it did not address the regions that were stimulated in the interventions; for instance, rTMS over the left dorsolateral prefrontal cortex (DLPFC) and over multiple brain regions were considered to be the same technique. Second, heterogeneity risk was increased because patients with MCI were included along with patients with AD. Third, the potentially synergistic effects or priming effect [23] of stimulations over different cortex within the same session were not addressed. Similar limitations are observable in a past pairwise meta-analyses [24]. Specifically, methodological limitations meant that different NIBS interventions were not effectively compared. Conventional pairwise meta-analyses cannot provide information regarding the relative efficacy of interventions that have not been directly compared in head-to-head RCTs, which constitutes an essential indicator of interventions’ therapeutic value. The risk was heterogeneity among studies which was also increased in other meta-analyses that included RCTs of not only patients with AD but also those with MCI [19, 20, 25].

Given this background, an NMA including only patients with AD that addresses variations in the targeted brain regions in NIBS interventions enables the estimation of comparative efficacy or risk and the understanding of the relative merits of different interventions. To the best of our knowledge, no NMAs on this subject have been performed. Therefore, we conducted a systematic review and NMA to compare the efficacy and safety of various central NIBS interventions in cognitive function in patients with AD.

Methods

General aspects

The NMA was performed according to the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement 2020 guideline (eTable 1) [26] and AMSTAR2 (a measurement tool to assess systematic reviews) guideline [27]. The current study had been approval by the Institutional Review Board of the Tri-Service General Hospital, National Defense Medical Center (TSGHIRB No. B-109-29).

Search strategy and selection criteria

The systematic review involved searching the PubMed, Embase, ClinicalKey, Cochrane CENTRAL, ProQuest, ScienceDirect, Web of Science, and ClinicalTrials.gov databases from their inception to September 15, 2020, and final update at November 26, 2020 (the keyword used in each database had been listed in eTable 2). No language restrictions were applied. The reference lists of review articles and pairwise meta-analyses [19,20,21, 24, 25, 28,29,30,31,32,33,34,39]. (3) The initial MMSE scores were significant for determining the time to clinically meaningful decline during longitudinal follow-up [40]. Similarly, the MMSE total score can serve as an index of disease progression and cognitive decline in patients with AD [41]. (4) The MMSE is suitable for evaluating patients with varying severities of AD; by contrast, the ADAS-Cog is recommended only for patients with MMSE scores of ≥14 [42]. Whenever MMSE scores were not available, we have chosen ADAS-Cog scores. Notably, previous RCTs have found that patients with AD receiving sham controls exhibited improvement in cognitive function but returned to baseline in a longer follow-up (2–6 mo) [10, 43]. Therefore, to reduce the risk of these potential time effects, we extracted the data from the “final follow-up assessment.”

Secondary outcome, safety profile, and intervention acceptability

The secondary outcome was the changes in quality of life. The safety profile was calculated using the rate of any adverse event and the rate of any local discomfort, including headache, itching, swelling, or local erythematous changes. Intervention acceptability was calculated using the dropout rate, which was defined as the percentage of patients withdrawing their participation before the end of study period for any reason.

Node definition

Because the central NIBS methods varied widely among the studies, we categorized them into 2 major subgroups: (1) rTMS modalities, namely HF-rTMS (≥5 Hz) and LF-rTMS (<5 Hz), and (2) tDCS modalities, which were particularly categorized according to the anodal or cathodal placement position (e.g., a-tDCS over F3 or c-tDCS over F3). We further categorized the treatment arms on the basis of the stimulation position according to the electroencephalogram brain map. The nomenclature of the node definition was defined according to our previous four NMAs of NIBS studies in other different neuropsychiatric diseases [23, 44,45,46].

Cochrane risk-of-bias tool

Two independent authors (YW Chen and BS Zeng) evaluated the risk of bias (interrater reliability =.85) for each domain using the Cochrane risk-of-bias tool [47]. The studies were further categorized according to overall risk of bias.

Statistical analysis

The NMA was performed using STATA (version 16.0; StataCorp Statistics/Data Analysis, StataCorp LLC, College Station, TX, USA). We estimated the standardized mean difference (SMD) with a 95% confidence intervals (95%CIs) for the continuous variable (i.e., the primary outcome of changes in cognitive function). In the subgroup analysis based on specific cognitive rating scales (i.e., the MMSE and ADAS-Cog), we estimated the mean differences (MDs) with 95%CIs to provide additional clinical information. For the categorical variables (i.e., the safety profile and acceptability), we estimated efficacy using odds ratios and 95%CIs and applied a 0.5 zero-cell correction during the meta-analysis procedure. However, if both the intervention and control arms of a study contained zeroes, we did not apply this correction procedure because of the risk of increasing the bias but rather excluded such studies from the analysis [48, 49]. We used the frequentist model of NMAs to compare the effect sizes of studies with similar interventions. All comparisons were performed using 2-tailed t tests, and differences were considered significant at P < .05. Between-study heterogeneity was evaluated using the tau value, the estimated standard deviation of the effect across the studies.

The meta-analysis procedure was mixed comparisons with direct and indirect comparisons made using generalized linear mixed models [50]. Specifically, indirect comparisons were conducted using transitivity, in which the differences between treatments A and B could be calculated from their comparisons with a third treatment C. To compare multiple treatment arms, we combined the direct and indirect evidence from the included studies [51]. The mvmeta command in STATA software [52] was used in the NMA. The restricted maximum likelihood method was used to evaluate the between-study variance [53].

To provide additional information for clinical application, we calculated the relative ranking probabilities of the treatment effects of all treatments for the target outcomes. In brief, we used the surface under the cumulative ranking curve (SUCRA), which indicates the percentage of the mean rank of each treatment relative to an imaginary intervention that is without uncertainty the optimal one [54].

We evaluated the potential within-network inconsistencies between the direct and indirect evidence by using the loop-specific approach, identifying local inconsistencies through the node-splitting method. The design-by-treatment model was used to evaluate global inconsistencies in the NMA [55]. We evaluated the quality of evidence with the Grading of Recommendations Assessment, Development and Evaluation (GRADE) rating tools described previously [56, 57]. Finally, according to the rationale of a previous NMA study [58], we assessed the effectiveness of the sham interventions to justify our assumption of transitivity. Specifically, we determined cognitive changes following sham tDCS and rTMS using Comprehensive Meta-Analysis software (version 3; Biostat, Englewood, NJ, USA).

Results

After the initial screening procedure, 88 articles were considered for full-text review (Fig. 1). After 69 were excluded for various reasons (eTable 3), 19 remained for analysis (eTable 4) [2, 9, 10, 43, 59,60,61,62,63,64,65,66,67,68,69,70,71,72,73]. Figure 2 presents the overall geometric distribution of the treatment arms.

Fig. 1
figure 1

PRISMA2020 flowchart of the current network meta-analysis

Full size image
Fig. 2
figure 2

Network structure of the changes of cognitive function. Overall structure of the network meta-analysis. The lines between nodes represent direct comparisons in various trials, and the size of each circle is proportional to the number of participants in each specific treatment. The thickness of the lines is proportional to the number of trials connected to the network

Full size image

Characteristics of the included studies

The mean age of the 639 participants was 72.5 years (range = 65.7–80.5 y; interquartile range [IQR] = 69.0–73.6 y), and women constituted 54.7% (range =10.0–83.3%; IQR = 44.1–65.4%). The mean duration of central NIBS treatment was 5.7 weeks (range = 2–32 wk; IQR = 2–6 wk). The corresponding mean follow-up duration was 10.5 weeks (range = 2–32 wk; IQR = 3.8–13.3 wk). The mean baseline MMSE was 19.1 (range = 14.8–22.5; IQR = 15.9–21.5). The most included RCTs allowed participants to use concomitant medication during the interventions [9, 43, 59,60,61,62,63,64, 67, 68, 70,71,72]. The investigated NIBS approaches were tDCS [10, 43, 60, 63, 67, 69, 72] and rTMS [2, 9, 59, 61, 62, 64,65,66, 68, 70, 71, 73]. All the tDCS treatments applied were 2mA in intensity. Although we planned to consider other NIBS techniques (i.e., TBS), no relevant RCTs on patients with AD receiving such therapies were retrieved.

Primary outcome: changes in cognitive function (based on either MMSE or ADAS-Cog measurement)

The NMA revealed that only the cathodal tDCS of the left DLPFC (F3) and anodal tDCS over the right supraorbital region montage (Fp2; c-tDCS-F3 + a-tDCS-Fp2; SMD = 2.43 [95%CIs = 0.61-4.26]) were associated with significant beneficial effect on cognition compared with sham controls (Table 1, Figs. 2A and 3A). The associations between the NIBS methods and the beneficial effect on cognition were ranked according to the SUCRA. In brief, c-tDCS-F3 + a-tDCS-Fp2 was associated with the greatest benefit, followed by high-frequency rTMS over the bilateral DLPFC (HF-rTMS-F3F4; SMD = 1.12 [95%CIs = −0.24 to 2.49]) and anodal tDCS of the left DLPFC and cathodal tDCS over the right supraorbital region (a-tDCS-F3 + c-tDCS-Fp2; SMD = 1.14 [95%CIs = −0.17 to 2.45]; eTable 6A).

Table 1 League table of the changes of overall cognition function
Full size table
Fig. 3
figure 3

Forest plot of changes of cognitive function. When the effect size (expressed as standardized mean differences) exceeded zero, the specified treatment was associated with greater improvement in cognitive function in patients with Alzheimer disease than in patients receiving sham controls

Full size image

Notably, the sham therapy effect did not significantly differ between the sham rTMS and tDCS therapies (P = .061). However, a significant placebo effect was observed in the sham rTMS group (SMD = 0.171 [95%CIs = 0.009–0.333], P = .038) but not in the sham tDCS group (SMD= −0.124 [95%CIs = −0.386 to 0.138], P = .355; eFigure 1).

Subgroup of primary outcome: changes in cognitive function as measured by the MMSE

Only the HF-rTMS-F3F4 (MD = 6.11 [95%CIs = 2.38–9.83]), c-tDCS-F3 + a-tDCS-Fp2 (MD = 5.55 [95%CIs = 3.72–7.38]), a-tDCS-F3 + c-tDCS-Fp2 (MD = 4.61 [95%CIs = 2.75–6.47]), and anodal tDCS of the left frontotemporal lobe and cathodal tDCS over the right Fp2 region (a-tDCS-F7 + c-tDCS-Fp2; MD = 2.38 [95%CIs = 0.45–4.32]) were associated with significant beneficial effect on cognition (measured by MMSE) compared with sham controls (Table 2, Figs. 2B and 3B). The associations between an NIBS method and the changes of cognitive function (measured by MMSE) were ranked according to the SUCRA. In brief, HF-rTMS-F3F4 was associated with the greatest benefit, followed by c-tDCS-F3 + a-tDCS-Fp2 and a-tDCS-F3 + c-tDCS-Fp2 (eTable 6B).

Table 2 League table of the changes of overall cognition function: measured with MMSE
Full size table

Subgroup of primary outcome: changes in cognitive function as measured by the ADAS-Cog

The NMA revealed that only the high-frequency rTMS over the left DLPFC (HF-rTMS-F3; MD = −3.40 [95%CIs = −6.57 to −0.23]) and high-frequency rTMS multifocal stimulation (HF-rTMS-Mx; MD = −3.06 [95%CIs = −4.55 to −1.57]) were associated with significant beneficial effect on cognition (measured by ADAS-Cog) compared with sham controls (Table 3, Figs. 2C and 3C). The associations between NIBS interventions and benefit of cognitive decline (measured by ADAS-Cog) were ranked according to the SUCRA. In brief, HF-rTMS-F3F4 was associated with the greatest benefit (MD = −4.00 [95%CIs = −8.73 to 0.73]) compared with sham controls, followed by HF-rTMS-F3 and HF-rTMS-Mx (eTable 6C).

Table 3 League table of the changes of overall cognition function: measured with ADAS-Cog
Full size table

Secondary outcome: changes in quality of life

None of the NIBS methods were associated with differences in changes of quality of life compared with sham controls (eTable 5A, eTable 6D, eFigure 2A, and eFigure 3A).

Safety profile: any adverse event rate

The NMA revealed that none of the NIBS methods were associated with different rates of any adverse events compared with sham controls (eTable 5B, eTable 6E, eFigure 2B, and eFigure 3B).

Safety profile: any local discomfort rate

No differences between the NIBS interventions and sham controls with regard to local discomfort rate were observed (eTable 5C, eTable 6F, eFigure 2C, and eFigure 3C).

Intervention acceptability: dropout rate

No differences in dropout rate between the NIBS interventions and sham controls were noted (eTable 5D, eTable 6G, eFigure 2D, and eFigure 3D).

Risk of bias, publication bias, inconsistency, and GRADE ratings

Overall, 74.4% (99/133 items), 24.1% (32/133 items), and 1.5% (2/133 items) of the included studies had a low, unclear, and high risk of bias, respectively, to which unclear reporting of the allocation procedure was the major contributor of unclear risk of bias (eFigure 4A-B). Funnel plots of publication bias revealed general symmetry and Egger’s test results indicated no significant publication bias among the included articles (eFigure 5A-N). Overall, no inconsistencies in the NMA were demonstrated in either local inconsistencies, as assessed using the loop-specific approach and node-splitting method, or global inconsistencies, as determined using the design-by-treatment method, with the exception of significant inconsistency in the primary outcome (design-by-treatment method: P = .0125) (eTable 7-8). However, this inconsistency disappeared in the subgroup of the primary outcome: changes of cognitive function measured by MMSE or ADAS-Cog (design-by-treatment method: P = .7001 and .9044, respectively). The results of GRADE evaluation are listed in eTable 9. In brief, the overall quality of evidence of the overall NMA, direct evidence, and indirect evidence were low to medium.

Discussion

To the best of our knowledge, this is the first comprehensive NMA investigating the efficacy and safety of NIBS interventions in patients with AD. As mentioned, c-tDCS-F3 + a-tDCS-Fp2 was associated with significant beneficial effect on cognition compared with sham controls. When focusing on specific cognition-rating scales, HF-rTMS-F3F4, c-tDCS-F3 + a-tDCS-Fp2, and a-tDCS-F3 + c-tDCS-Fp2 were determined to be associated with significant beneficial effect on cognition (measured by MMSE) compared with sham controls. Only HF-rTMS-F3 and HF-rTMS-Mx were associated with significant beneficial effect on cognition (measured by ADAS-Cog) compared with sham controls. Notably, a study reported significant placebo effects in sham rTMS therapy [74]. In addition, all the NIBS methods were well tolerated with regard to safety profile, as reflected in the rates of adverse events or local discomfort, as well as acceptability, as indicated by dropout rate.

The first main finding of this study was that the c-tDCS-F3 + a-tDCS-Fp2 was associated with a significant beneficial effect on cognition compared with sham controls and was also associated with the greatest beneficial effect on cognition among all the NIBS methods. The role of tDCS in changes of cognitive function in patients with AD is believed to be associated with the cognitive-enhancing effect of tDCS targeting the DLPFC, an area widely connected to cortical/subcortical regions and associated with executive control and memory [75]. However, the effect of the direction of tDCS currents on the specific cortical region is under debate. Although cathodal tDCS is generally believed to be associated with inhibitory effects on the targeted cortex, tDCS administered at higher intensity (i.e., 2mA) appears to result in reversal to excitatory effects [15]. Studies have indicated that these cortical effects outlast the original stimulation due to synaptic long-term potentiation [10, 76]. Another hypothetical mechanism regarding the cognitive benefits of tDCS involves the concerns about the activation of the cognitive reserve pool through acetylcholine and dopamine modulation [17, 18]. In addition, the beneficial effect of tDCS on cognitive function might be derived not only from cortical targeting but also from the effect of currents spreading to nearby cortical regions [77]. Taken together, the potential beneficial effect of c-tDCS-F3 + a-tDCS-Fp2 on the beneficial effect on cognition, if any, may involve different effects on multiple cortical regions. Notably, the finding of a beneficial cognitive effect of c-tDCS-F3 + a-tDCS-Fp2 was mainly drawn from a single study using small sample sizes (n = 12 and 11) [10]. In addition, its confidence interval was relatively wide (i.e., 95%CIs = 0.61–4.26). Therefore, it should be interpreted with caution, and future large-scale RCTs are warranted to provide more evidence.

Significant inconsistency in the primary outcome was noted overall (design-by-treatment method: P = .0125). However, this inconsistency disappeared in the subgroup of the primary outcome: changes of cognitive function measured by MMSE or ADAS-Cog (design-by-treatment method: P = .7001 and .9044, respectively). When we reexamined the evidence of the individual treatment arm, we found that the potential beneficial effect of a-tDCS-F3 + c-tDCS-Fp2 was also demonstrated in the subgroup analysis of changes of cognitive decline as measured by MMSE. In the overall NMA results, a-tDCS-F3 + c-tDCS-Fp2 did not achieve significant outcomes, in contrast to the fact that anodal tDCS targeted to the left DLPFC and left temporal cortex, among other cortices, has been reported to enhance executive function and the memory process [12, 13]. The nonsignificant result may be ascribable to the relatively shorter follow-up duration in one RCT, in which the a-tDCS-F3 + c-tDCS-Fp2 and sham control groups did not differ significantly in cognitive changes at a 3-week follow-up [72]. RCTs have shown that patients with AD receiving sham tDCS exhibited improvement in cognitive function but returned to baseline in a longer follow-up (i.e., 2-6 mo) [10, 43]. In the present study, we tried to reduce the potential impacts of time by extracting data at the final follow-up. Nevertheless, the relatively short follow-up duration (mean follow-up duration was 10.5 wks) may have confounded with time. Therefore, future RCTs are encouraged to use longer follow-up duration (i.e., at least 2–6 mo).

In the present NMA, the tDCS studies had relatively consistent findings. By contrast, those of the rTMS studies were inconsistent. This may be partially attributed to bias in the rTMS studies, such as the significant placebo effect observed in sham rTMS (SMD = 0.171, 95%CIs = 0.009-0.333, P = .038, eFigure 1). The placebo effect of NIBS interventions on cognitive outcomes may be associated with several mechanisms. First, because mimicking active rTMS is challenging, several modifications have been developed for rTMS, such as the application of peripheral auditory clicking sounds for intersensory facilitation [78]. The potential placebo effect may also result from the introduction of the intervention itself, the expectation of outcomes, optimism, and emotional goal-seeking [79]. Therefore, as suggested in one study, the alleviation of negative emotion (i.e., depressive mood) in older adults and the presence or company of study staff members may contribute to cognitive improvement [70]. To reduce the potential bias from the placebo effect, the development of adequate rTMS sham control interventions should be urgent. Finally, the electromagnetic field of standard coils decays rapidly, and they can only penetrate to a depth of 2–3 cm [80], which may be insufficient to reach the outer brain layers of older adults and patients with brain atrophy, both characteristics of patients with AD. Therefore, the potential of alternative rTMS techniques that can reach deeper structures (e.g., deep TMS) should be explored.

Limitations

This study has several limitations. First, this NMA may have been underpowered because of the heterogeneity of the participants (e.g., with regard to comorbidities, effects of concomitant medications on cognition, baseline AD severity, differences in AD onset age, different methods to define the stimulation target, different NIBS stimulation protocols, and follow-up duration). Second, although all the RCTs included a sham control in their study design, they may not have been adequately blinded because of differences in the interventions used. The placebo effect by the sham control, especially that of sham rTMS (SMD = 0.171, 95%CIs = 0.009–0.333, P = .038, eFigure 1), may also have imposed bias with regard to patients with AD [70]. Third, given the relatively small number of RCTs (and by extension patients) included, the main present findings should perhaps be conservatively applied in clinical practice. Specifically, the evidence of potentially beneficial effect of c-tDCS-F3 + a-tDCS-Fp2 on cognition in patients with AD was mainly derived from one RCT [10], which evaluated patients with mild-to-moderate AD (MMSE score between 11 and 23) with a 10-week follow-up. Therefore, the corresponding finding in the present study should be limited to patients with the same severity of AD that were followed up for at least 10 weeks. Fourth, whether tDCS can target specific cortices remains under debate [77]. Therefore, whether the potential beneficial effect of tDCS resulted from enhancement on the specific cortex remains unclear. Fifth, although most of the included RCTs provided comparisons between NIBS and sham control interventions, few provided comparisons between different NIBS interventions [2, 10, 66]. Therefore, the geometric structure of the current NMA was relatively weak (Fig. 2A–C). Sixth, although we intended to also include RCTs of deep TMS and TBS by adding these modalities in our search keywords, there was a lack of such RCTs available at present time. Finally, significant inconsistency was detected in the primary outcome overall. Although this inconsistency disappeared after subgroup analysis, clinicians should remain cautious in practical applications of those findings.

Conclusion

In summary, c-tDCS-F3 + a-tDCS-Fp2 was associated with significant beneficial effect on cognition compared with sham control. Moreover, it was associated with the greatest benefit of cognitive decline among all the NIBS interventions. In addition, all of the methods were suggested to be well tolerated with regard to safety profile, as reflected in the rates of adverse events or local discomfort, as well as acceptability, as indicated by dropout rate. The significant placebo effect in sham rTMS indicates that adequate sham control interventions in rTMS therapy should be urgently developed. However, because the treatment duration of the included studies was relatively short (mean=5.7 weeks, IQR = 2–6 weeks), future RCTs of tDCS are encouraged to use longer treatment duration to obtain more evidence of the beneficial effect by long-term NIBS interventions.