The effects of aerobic exercise and transcranial direct current stimulation on cognitive function in older adults with and without cognitive impairment: A systematic review and meta-analysis

‘attention ’ , ‘memory ’ , ‘memory performance ’ (outcome terms). We included only randomized controlled trials (RTC) in humans if available in English full text over the past 20 years, with participants ’ age over 60. We assessed the methodological quality of the included studies (RTC) by the Physiotherapy Evidence Database (PEDro) scale. Results: Overall, 68 studies were included in the meta-analyses. AE (ES = 0.56 [95% CI: 0.28 – 0.83], p = 0.01) and tDCS (ES = 0.69 [95% CI: 0.12 – 1.26], p = 0.02) improved GC in all three groups of older adults combined (healthy, MCI, demented). In healthy population, AE improved GC (ES = 0.46 [95% CI: 0.22 – 0.69], p = 0.01) and EF (ES = 0.27 [95% CI: 0.05 – 0.49], p = 0.02). AE improved GC in older adults with MCI (ES = 0.76 [95% CI: 0.21 – 1.32], p = 0.01). tDCS improved GC (ES = 0.69 [90% CI: 0.12 – 1.26], p = 0.02), all three cognitive function (GC, WM and EF) combined in older adults with dementia (ES = 1.12 [95% CI: 0.04 – 2.19], p = 0.04) and improved cognitive function in older adults overall (ES = 0.69 [95% CI: 0.20 – 1,18], p = 0.01). Conclusion: Our systematic review with meta-analysis provided evidence that beyond the cardiovascular and fitness benefits of AE, pairing AE with tDCS may have the potential to slow symptom progression of cognitive decline in MCI and dementia. Future studies will examine the hypothesis of this present review that a potentiating effect would incrementally improve cognition with increasing severity of cognitive impairment.


Introduction
Normal cognitive function is a quintessential element of healthy ageing (Morley et al., 2015). Cognitive impairment starting in late mid-life is associated with functional dependence, morbidity, and mortality (Calderón-Larrañaga et al., 2019). Brain injury or disease can cause dementia, characterized by a set of related symptoms involving a progressive deterioration of global cognition (GC), working memory (WM), and executive function (EF) (Duong et al., 2017). The World Health Organization (WHO) predicts that the number of dementia patients will reach 82 million in 2030 and 152 million in 2050 (WHO, 2020). With no cure in sight and drug trials ending with disappointing (Cummings et al., 2016;Rice, 2014) or controversial results (Walsh et al., 2021), non-pharmaceutical interventions are needed. Low-cost, side effects-free, and logistically simple treatments should be applied that have the potential to delay the onset age of dementia and slow disease progression to its full form. Of the many non-pharmaceutical interventions such as cognitive and behavioural training, diet, social facilitation, and music therapy singularly or in combination (Duplantier and Gardner, 2021;Gavelin et al., 2021;Lissek and Suchan, 2021;Mansky et al., 2020;Whitty et al., 2020), exercise as a lifestyle modifier represents an increasingly advocated alternative to pharmaceutical treatments of dementia (Baranowski et al., 2020;Bhatti et al., 2020;Falck et al., 2019;Gupta et al., 2021;Herold et al., 2019;Intzandt et al., 2021;Kraal et al., 2021;Macaulay et al., 2020;McGurran et al., 2019;Ruiz-González et al., 2021;Siddappaji et al., 2021;Zhang et al., 2020). Perhaps due to the complexity of the disease (Ferrari and Sorbi, 2021), exercise interventions like other single modality treatments, can be of low efficacy (Sanders et al., 2020), improve symptoms inconsistently (Sanders et al., 2019), or can be even controversial (Diamond and Ling, 2019). Ineffectiveness of aerobic exercise (AE) for improving cognition with increasing disease severity is especially striking (Sanders et al., 2019), as long-term interventions with presumably the most effective form of exercise failed to improve cognitive outcomes in mild cognitive impairment (MCI) and dementia (Hall et al., 2021;Sanders et al., 2019).
In animal models and humans, AE activates brain areas known to be involved in GC, WM, and EF such as the medial prefrontal, perirhinal cortex, striatum, hippocampus, and raphe nuclei in animals, and the dorsal anterior cingulate cortex, supplementary motor area, superior and middle frontal gyrus, right inferior frontal gyrus, middle temporal gyrus, anterior white matter tracts, and hippocampus in humans (Colcombe et al., 2006;Jonasson et al., 2017;Pietrelli et al., 2018;Terjung, 2011;Voss et al., 2013). AE can induce neuroplasticity and neuroprotection activated in cognitive processes (Constans et al., 2016;McDonnell et al., 2013;Mellow et al., 2020), an effect that is less likely to arise from neurogenesis (Hvid et al., 2021). One potential mediator of AE-induced neuroplasticity is brain-derived neurotrophic factor (BDNF) (Szuhany et al., 2015), which can increase even after just a single session of AE, causing adaptive plasticity (Huang et al., 2017).
Because AE and tDCS share common neural substrates, it is conceivable that pairing of the two methods could produce synergistic effects on cognitive function and lead to a higher efficacy rate than what is achieved with each intervention singularly. Indeed, combining AE with NIBS improved cognitive function in healthy young individuals, but whether such boosting effect would occur in older individuals with and without cognitive impairment has not yet been systematically reviewed (Hendrikse et al., 2017;Moreau et al., 2015;Thomas et al., 2020Thomas et al., , 2021Steinberg et al., 2019;Clark et al., 2021;Manor et al., 2018Manor et al., , 2016Nissim et al., 2019;Manenti et al., 2014;Wrightson et al., 2015;Ma et al., 2020;Schneider et al., 2021;Zhou et al., 2014;Tahtis et al., 2014). K. Talar et al. Ageing Research Reviews 81 (2022) 101738 While the most effective order for sequencing AE and tDCS remains to be determined, it seems that AE could potentiate subsequent NIBSinduced plasticity under certain conditions (Mellow et al., 2020). It remains unclear if such combined treatment would, in fact, have behavioural effects. Therefore, we aimed to systematically and meta-analytically review existing data to determine the effects of AE and tDCS on selected cognitive functions (GC, WM, EF) and see if pairing the two treatments would potentiate the singular effects in older individuals with and without MCI or dementia. We hypothesized that the individual effects of AE and tDCS would decrease with increase in cognitive impairment (Sanders et al., 2019). Further, we expected that the combination of the two methods would potentiate the individual effects and cognitive function would improve in individuals with MCI and dementia.

Registration of the systematic review protocol
The protocol of the investigation was registered in the International Prospective Register of Systematic Reviews PROSPERO (ID: CRD42021240644). This systematic review and meta-analysis were conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Page et al., 2021) (PRISMA checklist, Fig. 1).

Inclusion and exclusion criteria
The inclusion criteria were determined according to the PICOS (P = population, I=intervention, C=comparator, O=outcome, S=study design) approach: (1) participants that were older individuals (≥ 60 years of age). Individuals aged ≥ 60 years are considered older according to WHO (WHO, 2021); participants were healthy, diagnosed with MCI or dementia, (2) the intervention consisted of AE of intensity > 40% of heart rate reserve (HRR) (as moderate intensity occurs at 40-60% of HRR) (Karvonen et al., 1957) or tDCS (anodal, cathodal), (3) passive controls; (4) with the outcomes for cognitive function (GC, WM and EF), (5) randomized controlled trial (RTC) as study design, (6) manuscripts that were published in English. The exclusion criteria were as follow: (1) other intervention than aerobic exercise or tDCS, (2) active control group (i.e., comprising AE/strength training; studies with stretching, toning, tDCS sham or cognitive training as a control were included), (3) participants aged below 60 years, with (4) no outcomes for cognitive function, and (5) no randomized controlled trail. (6) Master or PhD theses as well as conference proceedings were excluded.

Data extraction
The following data from the included studies were extracted: (1) characteristics of the study (publication year, geographical area), (2) the sample size and patient characteristics (age, gender, size, cognitive health status), (3) intervention parameters (exercise program, session duration, frequency, intensity), (4) outcome measures and (5) overall effect of the outcome of interest. For quantitative analyses (meta-analyses) the group size and mean differences of the outcomes of interest with a 95% confidence interval (CI) or standard deviation (SD) for intervention and control group were collected. The data were tabulated in an Excel spreadsheet (Microsoft Corporation, Redmond, WA, USA). In case of missing data, the original authors were reached.

Study quality assessment
To evaluate the quality of included studies the 11-item Physiotherapy Evidence Database (PEDro) scale was used (de Morton, 2009). The PEDro scale assesses the methodological quality of randomized controlled trials in evidence-based physical therapy (Herbert et al., 1998). The scores ≤ 3 indicate poor study quality, 4 − 5 fair quality and ≥ 6 good to excellent quality (Maher et al., 2003). Items were scored as either present (1) or absent (0) and the sum of 10 scores was obtained. The first item (eligibility criteria) was not included in the total score due to external validity.

Statistical Analysis
The effect sizes (ES) were computed as the standardized mean difference between the AE group and the control group or tDCS and the sham group. In addition to the meta-analyses exploring the overall effects of AE and tDCS on cognitive outcomes, the subgroup analyses were performed exploring the effects of both interventions on (a) cognitive outcomes in healthy, MCI and demented older adults separately, (b) individual outcome categories (WM, EF, GC) regardless of the participant health status. The subgroup analyses were not performed for individual outcomes in healthy, MCI and demented older adults separately were not performed due to the low number or non-existent studies for these subgroups. A random-effects meta-regression method called robust variance estimation (RVE) for multilevel data structures was used in all analyses, because it allows for the inclusion of the multiple dependent outcomes from the same study. RVE assesses the variance of meta-regression coefficient estimates with the use of the observed residuals and does not require the weights or distributional assumptions (Hedges et al., 2010;Tipton, 2015). To account for the correlated effects within the studies the study was used as the clustering variable. The observations were weighted with the use of the inversion of the sampling variance. They ensure that the choice of correlational values does not impact the results of the meta-analysis, the sensitivity analysis was performed using alternative correlational values to calculate the standard error (SE). Between-study heterogeneity was evaluated using I 2 statistics. The values of I 2 > 25% indicate the low, > 50% moderate and > 75% high heterogeneity (Melsen et al., 2014). All analyses were performed using the robumeta (version 2.0) and metaphor (version 3.0-2) packages in R version 4.41.42 (The R Foundation for Statistical Computing, Vienna, Austria).

Study selection
To assess the effect of AE and tDCS on cognition and behavioural outcomes, expert librarians constructed a syntax for older adults without and with MCI, dementia, respectively (Appendix, S1). In healthy older adults, the search yielded 2642 articles and 30 were included in the analyses. In older adults with MCI, the search yielded 606 articles and 26 were included in the analyses. In older adults with dementia, the search yielded 1390 articles and 19 were included. As one study reported its results in two separate articles, 18 studies were included in this review (Fig. 1). Table S2 shows the methodological quality of the studies, revealing a score ranging from 5 to 10, with a median 8 (Appendix, S2).

Aerobic exercise studies
In 53 RCTs with AE intervention, there were 3427 participants (n = 2120 women) in the three groups (healthy older adults: n = 23 studies, MCI: n = 19 studies, dementia: n = 11 studies). Participants' age ranged from 65 to 86 years, with mean age increasing with disease severity (healthy: 79, MCI: 73, dementia: 82 years). Mean AE programs duration was 19 weeks (range 1-60 weeks, median 16 weeks, and mode 24 weeks, n = 20 studies), and the mean frequency was 3 times per week (range 1 -7 times per week), session duration varied between healthy older adults (range 1 -60 weeks, mean duration 40 min, n = 23 studies), MCI (range 8 -48 weeks, mean duration 46 min, n = 19 studies) and dementia (range 6 -24 weeks, mean duration 34 min, n = 13 studies). AE intensity in healthy older adults was 59% of HRR (range 40 -75% of HRR), in MCI 66% of HRR (range 40 -80% of HRR) and in demented 58% of HRR (range 40 -75% of HRR). Table 1 presents cognitive and behavioural outcomes and measures of the 53 studies (Table 1). Table 1 shows the characteristics of the studies examining the effects of aerobic exercise training on 3 measures of cognition and behavioral outcomes in older adults (Supplement 1, Table 1).

tDCS studies
In 20 RTCs with tDCS intervention in the three groups combined (healthy, MCI, dementia), there were 772 participants (443 women). The mean age ranged from 61 to 82 years. Three groups were varied in age (healthy: 68 years, MCI: 75, dementia: 72). The mean duration of tDCS program was 3 weeks (range 1-10 weeks, median 2 weeks, and mode 1 and 2 weeks, n = 12 studies), and the mean frequency was 3 times per week. tDCS stimulation intensity was similar in the three groups (healthy: 2.0 mA, MCI: 1.8 mA, dementia: 2.0 mA) and mean session duration varied (healthy: 32 min, MCI: 28 min, dementia: 26 min). Table 2 presents the effects of tDCS on 3 measures of cognition and behavioural outcomes of the 20 studies included (healthy older adults: n = 6, MCI: n = 7, dementia: n = 7, Table 2). Table 2 presents the study characteristics concerning the effects of tDCS on 3 measures of cognition (Supplement 1, Table 2). Table 4. presents the weighted descriptive statistics (mean and total    program duration, session duration, frequency, and intensity) for doseparameters (Table 4). In sum, AE improved GC in healthy, and MCI older adults and in the three groups combined but the results were inconsistent for WM and EF.  Figs. 1-3).

Pairing AE with tDCS in the three groups
There was one study (n = 13 participants, 9 women, age: 73 years) concurrently delivering tDCS during AE in healthy older adults in 18 sessions over 6 weeks (Clark et al., 2021). Table 3 presents the characteristics of this study (Supplement 1, Table 3). There were no studies in older adults with MCI and dementia.
Moreover, Fig. 6 summarizes the effects of AE and tDCS on three measures of cognition (GC, WM, EF) in the three patient groups (healthy, MCI and demented older adults).

Summary of results
For the first time, we aimed to systematically and meta-analytically review existing data to determine the effects of AE and tDCS on selected cognitive function (GC, WM, EF) and see if pairing the two treatments would potentiate the singular effect of AE with tDCS in older adults with and without MCI or dementia. While suggested repeatedly (Hendrikse et al., 2017;Moreau et al., 2015;Steinberg et al., 2019), there are currently insufficient data to examine the hypothesis and to conclude that pairing AE and tDCS would produce a potentiated effect on cognition in ageing. Against the hypothesis, we found significant effects of individually delivered AE and tDCS on GC independent of cognitive status (healthy, MCI, demented). In agreement with the hypothesis, the severity of cognitive impairment affected the efficacy of AE to improve cognition so that AE did not improve cognition in dementia. tDCS was effective in a pooled analysis of cognition (GC, WM, EF) in older adults with dementia. Moreover, tDCS intervention improved global cognition in the three groups combined.
We discuss these data with a perspective on the potential of pairing AE with tDCS in an effort to reciprocally boost the effects of these individual treatments on cognition and provide individualized and disease-specific treatment options for reducing symptom evolution of    cognitive impairment in ageing.

Effects of AE combined with tDCS on three measures on cognition in older adults with and without MCI or dementia
Notwithstanding repeated calls and conceptual frameworks, our systematic search identified little to no data concerning the synergistic effects of pairing chronic AE with chronic tDCS on CG, WM, and EF in the three populations. Synergistic effects between concurrent AE and tDCS could arise because cognition and walking and running, activities often used in AE and rehabilitation, share common neural substrates (Dougherty et al., 2021;Kikkert et al., 2016;Moreau et al., 2015;Morris et al., 2016;Steinberg et al., 2019;Verlinden et al., 2014). In particular, the frontal lobe, where circuits controlling EF reside, becomes increasingly activated as the speed and complexity of walking increase (Clark et al., 2014;Wagshul et al., 2019). Conversely, impaired EF is associated with gait slowing and a reduced ability to perform complex gait tasks (Nutt, 2013;Steinberg et al., 2019). Intervention variables to produce dose effects on cognition are unclear from single-session cross-sectional studies, as cognitive, motor skill, sports skill, and AE training at varying intensities, duration, and complexity all seemed to improve cognition in combination with tDCS Manenti et al., 2014;Manor et al., 2018Manor et al., , 2016Moreau et al., 2015;Nissim et al., 2019;Schneider et al., 2021;Tahtis et al., 2014;Wrightson et al., 2015;Zhou et al., 2014).
A combination of tDCS at 1 mA (below the 2-mA maximal stimulator output) with a moderate level of AE in a single session appeared effective to immediately improve healthy young adults' EF . In the context of the current review, the only study that combined and concurrently delivered real and sham tDCS with complex walking ('AE') chronically in 18 sessions over 6 weeks in healthy older adults age ≤ 65 y, cautiously concluded that there is a potential for improving EF by adding frontal tDCS to walking rehabilitation (Clark et al., 2021). While this study administered complex walking tasks and tDCS concurrently (Clark et al., 2021), there is also evidence that AE could potentiate subsequent NIBS-induced plasticity under certain conditions, requiring additional studies to determine the most effective order for sequencing AE and tDCS (Mellow et al., 2020). While tDCS and AE act on overlapping brain areas, each method acts via different molecular mechanisms. tDCS presumably improves cognition by modifying the levels of acetylcholine, dopamine, and GABA and cortical activation, whereas AE modifies levels of growth factors (IGF-1, BDNF, VEGF), dopamine, glutamate, serotonin, and norepinephrine, promoting vascularization and neurogenesis (Hendrikse et al., 2017;Moreau et al., 2015;Steinberg et al., 2019). Our search identified no studies that examined the synergistic effects of pairing AE and tDCS on CG and WM in MCI and dementia. A potentiating effect might still occur between these two interventions to improve CG, WM, and EF because pairing cognitive training combined with tDCS can improve selected measures of cognition in older adults with and without MCI or dementia and in selected psychiatric conditions (Ciullo et al., 2021;Gonzalez et al., 2021;Lu et al., 2019;Siegert et al., 2021).
In sum, the strong conceptual framework of dual application of chronic AE and tDCS for augmenting the individual treatment effects on cognition is juxtaposed with scant experimental evidence. Future studies will need to examine the hypothesis of this present review that such a potentiating effect would incrementally improve cognition with increasing severity of cognitive impairment. In addition, sequencing effects of AE and tDCS should be elucidated.

Effects of AE on three measures of cognition in older adults with and without MCI or dementia
AE is considered as a highly effective strategy to improve cognition in older adults (Gheysen et al., 2018). Neuroplasticity, the mechanism responsible for creating and modifying synaptic connections improves cognitive function with AE in older people (Quigley et al., 2020). Chapman et al. study of shorter duration of AE (12 weeks) observed improvement in healthy older people's hippocampal size and blood flow (Chapman et al., 2013). Similarly, Voss et al. presented increase in temporal lobe connectivity after AE training for 48 weeks in older adults (Voss, 2010). Considering that each of the above-mentioned studies suggests the upregulation of growth factors and neuroplasticity as a key biological mechanism that appears to underline exercise-induced cognitive improvement in older individuals Vecchio et al., 2018), future studies should focus on identifying the role of BDNF, vascular endothelial growth factor (VEGF), and insulin-like growth factor-1 (IGF-1) following exercise-induced adaptations.
In mildly cognitively impaired older adults, AE carried beneficial effects for CG (ES = 0.76, p = 0.01) and cognition overall (ES = 0.59, p = 0.01), but there were no statistically significant effects of AE on WM and EF. These results are partially in line with previous meta-syntheses which suggest medium effect of AE on GC and a small effect on EF and WM in older adults with MCI (Biazus-Sehn et al., 2020;     -Hofstetter et al., 2018;Chen et al., 2020;Northey et al., 2018;Sanders et al., 2019;Zheng et al., 2016;Zhou et al., 2020). The reason for such differences could be the number of studies being substantially higher in the present review (Supplement 2, Fig. 2, n = 15 studies) than in previous reviews (n = up to 9 studies). Beneficial effects of AE on cognition in older adults with MCI may be caused by exercise-induced increases in cortical excitability, motor evoked potential (MEP) responses (Dai et al., 2013;McGregor et al., 2018), up-regulation of BDNF (Allard et al., 2017), increased left hippocampal volume (ten Brinke et al., 2015) and prevention of brain volume loss (Frodl et al., 2020). Another factor could be that the mean program duration of AE (Table 4, ~21 weeks) was relatively shorter than in the previous reviews (~23 weeks) and the mean session duration lasted for ~46 min, while in the other studies for ~51 min. Northey et al. recommended a session duration of 45 min that is beneficial for cognition (Northey et al., 2018). We suspect that GC, EF and WM responded differently to the exercise stimulus due to the significantly varying level of cognitive deterioration among the studies (decline measured objectively over time or subjective assessment of decline by the participant) (Winblad et al., 2004).
In demented older people, previous metasyntheses showed that AE has medium effects on GC and small effects on EF and WM (Heyn et al., 2004;Law et al., 2020). Our results do not support these data, as we found statistically non-significant small effects of AE on CG (ES = 0.19, p = 0.34) and small, non-significant effects of AE on cognition overall (ES = 0.08, p = 0.32). Moreover, there were no statistically significant results of AE on WM and EF. The source of discrepancy could be the number of studies being substantially higher in the present review ( Figure 9, n = 8 studies) than in previous reviews (n = up to 7 studies). Another factor could be that the mean program duration of AE (Table 7, ~13 weeks) was shorter than in the previous studies (~17 weeks) and the mean session duration lasted for ~34 min, while in the other studies for 45 min. Also, the mean age was higher (Table 7, 82 years) than in the previous studies (~79 years).
In conclusion, we found robust evidence for a decrease in the effectiveness of AE on three measures of cognition (CG, WM, EF) with increasing severity of cognitive impairment. In the spirit of discussion under 4.2, future studies will need to seek alternatives to singular treatments by blending interventions that are conceptually expected to produce a synergistic effect. We also found a positive effect of tDCS on GC individually and all cognitive function (GC, WM, EF) in the three groups combined. Moreover, AE was effective to improve GC individually in older adults with MCI and when the three cognitive measures were combined (GC, WM and EF).

Effects of tDCS on three measures of cognition in older adults with and without MCI or dementia
The increasing popularity of tDCS in sport science is observed due to the evidence of regulation of exertion markers, eg. RPE, HR . The effect of NIBS is measured using the change in motor evoked potentials (MEPs) when applied over the primary motor cortex (M1) (Huang et al., 2017). MEPs are the electrical signals recorded from the muscles via electromyography (EMG) that respond to the direct stimulation as an index of the motor cortex excitability (Legatt, 2014). It is recommended to measure MEPs with TMS before using tDCS due to the high variability of corticospinal excitability (CSE) in tDCS protocols (Bashir et al., 2019;Horvath et al., 2016Horvath et al., , 2015. The reasons of occurrence of variability are still unknown (Laakso et al., 2019). MEPs can be evoked by the paired pulse techniques: the short interval intracortical inhibition (SICI), obtained while giving 2 pulses close to each other (2 ms) to condition and test stimuli or by the long interval intracortical inhibition (LICI), when 2 pulses are distal to themselves (800 ms) or intracortical facilitation (ICF), when during the inter-stimulus intervals (ISI) (10 ms) the conditioned MEP is greater than the test MEP (Chen et al., 1998;Huang et al., 2017;Udupa et al., 2009;Ziemann, 1999  cognition and delays cognitive decline in healthy older people (Indahlastari et al., 2021).
Contrary to the previous results that tDCS has medium effects on GC, large effects on EF and medium effects on WM in healthy older adults Huo et al., 2021;Summers et al., 2016), present study shows large but non-significant effects on all cognitive function combined (ES = 0.88, p = 0.33). The source of discrepancy could be the small number of studies included in our meta-analysis (Supplement 5, Figure 16, n = 2 studies) than in the previous reviews Huo et al., 2021;Summers et al., 2016), up to 11 studies. Such difference in the number of the included studies occurred due to the different outcome measures (e.g., motor outcome) used in the meta-analysis (Summers et al., 2016), different non-invasive brain stimulation techniques  or proposed by Gavelin et al., 2021 classification of cognitive outcomes (Gavelin et al., 2021). This last is critical factor as there is not a generic classification of cognitive processes (Harvey, 2019;Hay et al., 2017) and differences in cognitive categories could lead to different assessment.
Moreover, previous studies showed that tDCS has large effects on GC, small effects on EF and medium effects on WM in older adults with MCI (Chu et al., 2021;Cruz Gonzalez et al., 2018;Xu et al., 2019b). Again, our results do not support these data, as we found no significant effects

Table 3
The effects of aerobic exercise training combined with transcranial direct current stimulation intervention on cognition and behavioural outcomes in healthy older adults. on GC, EF, and WM in older adults with MCI. We included small number of studies in our meta-analysis (Supplement 5, Fig. 2, n = 6 studies), similarly to other authors in the previous reviews (Chu et al., 2021;Cruz Gonzalez et al., 2018;Xu et al., 2019b), n = up to 4 studies). The difference in the obtained results occurs due to the different non-invasive brain stimulation techniques (Xu et al., 2019b) or division of short-term and long-lasting effects of tDCS on cognitive function (Chu et al., 2021;Cruz Gonzalez et al., 2018). Cruz Gonzalez et al. suggested that tDCS had small effects on WM in older adults with dementia (Cruz Gonzalez et al., 2018). Our results partially support these data, as we found large and significant effects on cognitive function (GC, WM, and EF) in older adults with dementia after tDCS intervention (ES = 1.12, p = 0.04). The source of discrepancy could be the number of studies being substantially higher in the present review (Supplement 5, Fig. 3, n = 6 studies) than in the previous (up to 4 studies). Moreover, to our knowledge, there were no reviews examining the effects of tDCS on WM and EF in demented older adults. Therefore, it is recommended to experimentally and systematically investigate the effect of tDCS on cognition in older people with dementia.
In summary, tDCS was effective on all three cognitive function combined (GC, WM and EF) in older adults with dementia. We found no statistically significant effects of tDCS on cognitive function in older adults with MCI.

Interaction of AE and tDCS with disease severity
The emerging picture supports the idea that the effectiveness of AE and tDCS differs with disease severity. While AE improves cognitive function in healthy (ES = 0.40, p = 0.001), and MCI (ES = 0.59, p = 0.003) older adults, tDCS is effective in demented older people (ES = 1.12, p = 0.04). Moreover, the present study found that older people overall (healthy, MCI, demented) significantly improved in GC after tDCS (ES = 0.69, p = 0.02) and AE (ES = 0.56, p = 0.01) interventions. Hence, a combination of the two methods (AE + tDCS) appears to be a promising non-pharmacological intervention to delay in midlife the onset age of clinical dysfunction and slow progression of cognitive impairment with ageing, which can possibly reduce the dependency of older adults and, improve quality of life (Fusco et al., 2012) and decrease care costs (Guralnik et al., 2002;Hazra et al., 2018).

Strengths and limitations
In this study, the RVE meta-regression was used to be able to include multiple dependent outcomes from the same study. This is an important strength of the present meta-analyses, as many of the included studies reported multiple cognitive outcomes, with the substantially varying effect of the intervention, sometimes even in the opposite direction Arcoverde et al., 2014;Eggenberger et al., 2016;Legault et al., 2011;Miu et al., 2008;Prehn et al., 2019;Scherder et al., 2005;Schoene et al., 2015;Yu et al., 2021). Using a standard meta-analytical approach would require choosing just one outcome and discarding others, leading to a potential selection bias. Thus, similarly to other meta-analyses of physical interventions (Pallarés et al., 2021;Talar et al., 2021), the use of RVE is better suited for the purpose and can also explain the discrepancies between our results and the results of some previous reviews (Chen et al., 2020;Law et al., 2020;Xiong et al., 2021;Y. Xu et al., 2019;Young et al., 2015;Zhu et al., 2016).
This systematic review has limitations. First, except for AE on GC in healthy older adults and tDCS on GC, WM in older people with MCI, most meta-analyses presented moderate to high levels of heterogeneity. This fact could be explained by the different methodologies (e.g., volume, intensity, type of AE intervention, stimulation tool, protocol, stimulation position, program duration, session duration and frequency), different variables included in the quantitative analysis (i.e., clinical diversity), as well as by the inconsistent statistical analyses performed in two studies Yu et al., 2021) that affected our results and the interpretation.
Second, only one study met the inclusion criteria for the combination of tDCS and AE intervention in older people (Clark et al., 2021). For this reason, future systematic reviews are encouraged to examine the effects of pairing AE with tDCS on GC, WM and EF in older people with or without MCI or dementia.
Third, the cognitive tools used by included studies, could assess the level of cognitive impairment, not measure GC. The studies where MMSE, MoCA and ADAS-Cog were measured only at baseline were excluded from our meta-analysis (Albinet et al., 2016(Albinet et al., , 2010Barnes et al., 2013;Brydges et al., 2020;Cancela et al., 2016;Davis et al., 2013;Esmail et al., 2020;Ferreira et al., 2015;Karssemeijer et al., 6. The effects of aerobic exercise and transcranial direct current stimulation training on global cognition, working memory, executive function in healthy, MCI and demented older adults. Pooled effect size greater than zero favour intervention (AE, tDCS training) vs. passive control and were computed using Random Variance Estimate meta-analytical modelling. Note that the figure does not contain any data on the combined effects of AE and tDCS on cognition, as only one study has examined such combined effects. Horizontal brackets denote confidence intervals (p < 0.05). AE, aerobic exercise; GC, global cognition; WM, working memory; MCI, mild cognitive impairment; EF, executive function. Maki et al., 2012;Nagamatsu et al., 2013;Raichlen et al., 2020;Scherder et al., 2005;Shimada et al., 2018;Suemoto et al., 2014;ten Brinke et al., 2015;van Uffelen et al., 2008). Similarly to Arevalo-Rodriguez et al., 2015 we recommend using additional and extensive cognitive tests to observe the disease severity from MCI stages to dementia. That could be achieved by MMSE, MoCA and ADAS-Cog changes over time instead of a single measurement (Arevalo-Rodriguez et al., 2015).
Fourth, dementia as a clinical syndrome is not a single disease, covers many medical conditions: Alzheimer's disease (60-80% cases), vascular dementia (5-10% cases) and dementia with Lewy bodies (5-10% cases) (Anon, 2021;Sheehan, 2012). For this reason, the studies in the dementia category included the above-mentioned diseases. Although it could be considered as a strength of this article, it is possible that such division influenced our results in older adults with dementia.

Recommendations for future research
The low number of studies with dual AE and tDCS (only 1 study found for healthy older adults) suggests the need for more studies examining the synergistic effects of AE and tDCS on cognitive function in older people. Future studies and reviews should also address the limitations of the present review (duration, timing, frequency, intensity of AE). It should be noted, that both techniques have similarities except separate mechanisms and modulating ability of brain functions. Moreover, both techniques may be used in a direct combination (i.e., tDCS can stimulate the brain while exercising). Moreover, other non-invasive brain stimulation techniques are recommended, e.g., TMS and tACS during exercise (Ross et al., 2018). These methods are relatively easy to apply, safe (no known severe side effects) and cost-effective. Lastly, the long-term effect of pairing tDCS and AE should be examined in future studies.

Conclusions
Beyond the cardiovascular and fitness benefits of AE, pairing AE with tDCS may have the potential to slow symptom progression of cognitive decline in MCI and dementia. Future studies will need to examine the hypothesis of this present review that a potentiating effect would incrementally improve cognition with increasing severity of cognitive impairment.

Conflicts of Interest
The authors declare no conflict of interest.

Data Availability
Data will be made available on request.