Physical Activity Improves Cognition and Activities of Daily Living in Adults with Alzheimer’s Disease: A Systematic Review and Meta-Analysis of Randomized Controlled Trials

Objective: The purpose of this meta-analysis was to examine the effects of physical activity (PA) on cognition and activities of daily living in adults with Alzheimer’s Disease (AD). Methods: Six electronic databases (MEDLINE, CINAHL, PsycArticles, SPORTDiscus, EMBASE and CNKI) were used to search for potential studies from inception until October 2021. Randomized controlled trials (RCTs) investigating the effect of physical activity (PA) on cognition and activities of daily living in AD patients compared to a control condition were included. The effect sizes were synthesized using a random effects model with a 95% confidence interval (CI). Results: Sixteen articles including 945 participants (aged 70 to 88 years, 34.6% male) met the inclusion criteria. The pooled effect sizes demonstrated that PA intervention was associated with significant improvements in global cognition (Standard Mean Difference (SMD) = 0.41, 95% CI [0.24, 0.58], p < 0.01) and activities of daily living (SMD = 0.56, 95% CI [0.32, 0.79], p < 0.01) in AD patients. Subgroup analyses suggested that PA for 3–4 times per week for 30–45 min for more than 12 weeks had a relatively strong effect on improving global cognition in AD patients. The sensitivity analysis showed robust results. Conclusions: The findings from the current meta-analysis suggested that AD patients can improve their global cognition and Activities of Daily Living (ADL) through engaging in aerobic and mixed exercise (aerobic and anaerobic exercise).


Introduction
Alzheimer's disease (AD) is the most common type of dementia, accounting for 60-80% of the confirmed cases. It is a chronically non-communicable, irreversible, neurodegenerative disorder, primarily appearing in the population aged 65 years and older [1,2]. The causes of AD are multifactorial, including gene mutation, age, having less physical activity, lower education levels, and cardiovascular disease and complications [1]. Individuals diagnosed with AD have different symptoms, including memory impairment, apathy, disorientation, behavior changes, and even difficulty walking and speaking. Moreover, AD can result in a poor quality of life and a loss of independence [1]. According to the World Health Organization (WHO), approximately 50 million people have been diagnosed with dementia worldwide, and the recently estimated number will be increased to 82 million in 2030 and 152 million in 2050 [3].
OR "Alzheimer*" OR "AD"; (3) "randomized controlled trials". Additional articles were manually found by studying the reference lists of the included studies. PRISMA (Preferred Repointing Items for Systematic Reviews and Meta-Analyses) was used in the current study [28].

Inclusion and Exclusion Criteria
The included articles met the following criteria: (i) participants had a clinical diagnosis of AD based on the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association (NINCDS-ADRDA), as well as the Diagnostic and Statistical Manual of Mental Health Disorders (DSM-IV); (ii) the interventions involved one or more similar PA therapies (e.g., walking, or walking and stretching) in the experimental group and either a placebo-control condition or a control condition, without using any other exercise interventions in the control group; (iii) the study design consisted of randomized controlled trials (RCTs); (iv) at least one outcome measurement (i.e., ADL and global cognition, e.g., (Mini-Mental State Examination (MMSE), the Cognitive section of the Alzheimer's Disease Assessment Scale (ADAS-Cog), and the Rapid Evaluation of Cognitive Function (ERFC))) was reported; (v) studies were published in English or Chinese. Studies were excluded when (i) they were animal or in vitro studies, abstracts, case reports, reviews, or trial protocols; (ii) there were participants with other types of dementia (e.g., vascular dementia) or mild cognitive impairment; (iii) outcome measurements could not be used to calculate the effect size (ES).

Data Extraction and Quality Assessment
Eligible studies were evaluated by two independent authors, screening the titles, abstracts, and full texts. When disagreement appeared, a third author would be consulted. The following information was extracted from each included article: author, year of publication, characteristics of subjects, sample size, gender, age, intervention protocol (frequency, time, type, duration and intensity), outcome measurements (ADL and global cognition (e.g., MMSE and ADAS-Cog)), and raw data for effect size calculation.
The eligibility studies' risk of bias was assessed by two independent reviewers, using the Physiotherapy Evidence Database (PEDro) scale [29], which is a systematic tool to assess the study quality. This scale encompasses 11 items, of which item 1 is not involved in calculating the total scores. The total valid scores range from 0 to 10. The methodological quality classifications are excellent (9-10 points), good (6-8 points), fair (4-5 points) and poor (<4 points) [29].

Statistical Analysis
Meta-analysis was performed using the Comprehensive Meta-Analysis software (v2 software, Biostat, Englewood, NJ, USA). The continuous data on cognition and ADL were extracted from the included studies. Standardized mean difference (SMD) was expressed as the effect size (ES) measure by calculating the mean change from pre-to post-intervention in the experimental and control groups, respectively. The standard deviation was calculated using the equation: SD = SE* √ n . The ESs were calculated using a random effects model with a 95% confidence interval (CI) to avoid a high risk of false-positive results [30]. The pooled ESs were presented in a forest plot, and a positive ES value reflected a more effective PA. ES was commonly classified into three levels: small (0.2-0.39), moderate (0.4-0.60), and large (≥0.6) [31]. Heterogeneity was examined using the I 2 statistic for each analysis: I 2 < 25% (low heterogeneity), I 2 < 50% (moderate heterogeneity), and I 2 < 70% (high heterogeneity) [32]. Prediction intervals (PIs) were measured to reflect the variation in intervention effects when more than 10 studies were included in the meta-analysis according to the formula: 95% PI = MD ± 2 Tau 2 [33]. Sensitivity analysis was performed to evaluate the stability of the results by excluding one study each time. Lastly, the funnel plot, Begg's test, and Egger's regression test were employed to examine the publication bias. Duvall and Tweedie's trim and fill analysis was used to recalculate a new ES by adjusting the outlier values. The significance level was set at p < 0.05. In addition, to investigate the confounding factors (study design, the stage of AD, location, and exercise principles), we performed subgroup analyses and meta-regression tests accordingly. In terms of the exercise principle, a prior study indicated that exercise modalities were likely to influence the findings [34]. Based on the included studies, we slightly modified the cut-off of the exercise duration (short-term: ≤12 weeks; medium-term: 13-23 weeks; long-term: ≥24 weeks) and session time (medium: 30-45 min; longer: >46 min) based on the American College of Sports Medicine [35]. With regard to gender, according to the proportion of males and females in original studies, we coded a high male percentage as ≥40% and a high female percentage as >60%. Figure 1 depicts the flowchart of the study selection for this meta-analysis. The initial search record of studies was 13,220. After removing duplicates (n = 2180), 11,040 studies were remaining. Furthermore, 10,994 studies were eliminated after reviewing the titles and abstracts. As a result of the screening process, 30 articles were excluded by assessing the full texts (n = 46), and 16 RCTs [36][37][38][39][40][41][42][43][44][45][46][47][48][49][50][51] met the inclusion criteria in this meta-analysis.

Characteristics of Included Studies
Detailed information of the included studies is shown in Table 1. All studies were published after 2010. The included studies were from Asia, South America, and Europe. The sample size ranged from 20 to 200. A total of 945 participants were diagnosed with mild to severe AD (approximately 34.6% of males). The mean age of the participants
According to the Physiotherapy Evidence Database (PEDro) scale, the methodological quality of each study is presented in Table 2. The study quality score fluctuated from six to eight, implying that all studies were considered to be of good methodological quality. Concealed allocation was used in eight studies [37][38][39][40]42,44,49,50]. Only two studies performed a blinding method to measure the outcomes [39,40].

Effects of PA Intervention on Global Cognition
The pooled result from 18 studies showed a significant improvement in favor of PA (SMD = 0.41, 95% CI 0.24 to 0.58, I 2 = 30%, p < 0.01; 95% PI = −0.04 to 0.85) ( Figure 2, Supplement Figure S1). However, the PI revealed that PA may not elevate global cognition in more than 5% of trial settings. By removing one study at a time, the pooled result of global cognition was stable. A symmetrical funnel plot was shown in Figure 3, indicating no publication bias for global cognition. Begg's test (Kendall's tau = 0.07, p = 0.74) and Egger's test (intercept = 1.48, p = 0.08) indicated no publication bias. Furthermore, the ES was recalculated using Duvall and Tweedie's trim and fill procedure, adding two studies on the right sides of the funnel plot, and an adjusted ES was SMD = 0.40 (95% CI 0.28 to 0.60, p < 0.05).
Additionally, we conducted subgroup analyses to examine potential factors affecting PA on cognition, shown in Table 3. There were significant differences in how exercise type (Q (1) Table 3 and Figure 4).  Note: PEDro, Physiotherapy Evidence Database scale; Studies were classified as having excellent (9-10), good (6-8), fair (4-5), or poor (<4). Scale of item score: , present. The PEDro scale involves (1) eligibility criteria; (2) random allocation; (3) concealed allocation; (4) similarity at baseline on key measures; (5) participant blinding; (6) instructor blinding; (7) assessor blinding; (8) more than 85% retention rate of at least one outcome; (9) intention-to-treat analysis; (10) between-group statistical comparison for at least one outcome; (11) point estimates and measures of variability provided for at least one outcome. more than 5% of trial settings. By removing one study at a time, the pooled result of global cognition was stable. A symmetrical funnel plot was shown in Figure 3, indicating no publication bias for global cognition. Begg's test (Kendall's tau = 0.07, p = 0.74) and Egger's test (intercept = 1.48, p = 0.08) indicated no publication bias. Furthermore, the ES was recalculated using Duvall and Tweedie's trim and fill procedure, adding two studies on the right sides of the funnel plot, and an adjusted ES was SMD = 0.40 (95% CI 0.28 to 0.60, p < 0.05).   more than 5% of trial settings. By removing one study at a time, the pooled result of global cognition was stable. A symmetrical funnel plot was shown in Figure 3, indicating no publication bias for global cognition. Begg's test (Kendall's tau = 0.07, p = 0.74) and Egger's test (intercept = 1.48, p = 0.08) indicated no publication bias. Furthermore, the ES was recalculated using Duvall and Tweedie's trim and fill procedure, adding two studies on the right sides of the funnel plot, and an adjusted ES was SMD = 0.40 (95% CI 0.28 to 0.60, p < 0.05).    Figure 5 shows a forest plot of the meta-analysis. The pooled result from nine studies revealed that the PA intervention had a significant benefit on ADL (SMD = 0.56, 95% CI 0.32 to 0.79, I 2 = 29.86%, p < 0.001). By removing one study at a time, the pooled result of ADL was stable. A symmetrical funnel plot was shown in Figure 6, indicating no publication bias for ADL. Begg's test (p = 0.60) and Egger's test (p = 0.29) indicated no publication bias. However, Duvall and Tweedie's trim and fill procedure revealed that no study was missing after adjusting for outlier values. A new ES was SMD = 0.54 (95% CI 0.32 to 0.79, p < 0.05).   Figure 5 shows a forest plot of the meta-analysis. The pooled result from nine studies revealed that the PA intervention had a significant benefit on ADL (SMD = 0.56, 95% CI 0.32 to 0.79, I 2 = 29.86%, p < 0.001). By removing one study at a time, the pooled result of ADL was stable. A symmetrical funnel plot was shown in Figure 6, indicating no publica- 0.32 to 0.79, I 2 = 29.86%, p < 0.001). By removing one study at a time, the pooled result of ADL was stable. A symmetrical funnel plot was shown in Figure 6, indicating no publication bias for ADL. Begg's test (p = 0.60) and Egger's test (p = 0.29) indicated no publication bias. However, Duvall and Tweedie's trim and fill procedure revealed that no study was missing after adjusting for outlier values. A new ES was SMD = 0.54 (95% CI 0.32 to 0.79, p < 0.05).

Discussion
This systematic review with meta-analysis examined the effects of PA on adults with AD, as well as the estimated ESs of PA on cognition and ADL. Compared to previous meta-analyses, there were six studies [36,39,43,44,49,50] newly included in this review. Thus, a total of 945 participants (34.6% male) were involved in 16 RCTs. Overall, the methodological quality of the included studies was high. The findings showed that PA significantly benefited global cognition and ADL in adults with AD.
Although many meta-analyses have indicated the effects of exercise on adults with AD [17,27,53], there are differences in selection criteria between our study and previous studies. Due to the different mechanisms of dementia, abnormal protein deposits may coexist with neuro-vasculature at different stages of dementia and influence brain function [54]. For example, Lewy body dementia is associated with the α-synuclein protein; AD is associated with the β-amyloid and tau-proteins; vascular dementia is attributed to a lack of blood flow [55]. Therefore, we included participants with only Alzheimer's type of dementia in this current study. Some meta-analyses intended to target people with AD, but the actual analysis included people with different types of dementia [17,27,53]. Zhu et al. conducted the meta-analysis by including both RCTs and non-RCTs, which directly increased the risk of bias [53]. Moreover, only exercise intervention was considered to meet the inclusion criteria. Some meta-analyses have combined the exercise with other therapies (e.g., psychotherapy) in the experimental group, which influenced the effectiveness of sole exercise [53,56]. Our study, compared to other systematic reviews, avoided the flaws of the above aspects and excluded some original studies. In this context, the findings from previous studies may not necessarily generalize the specific individuals with AD. On the contrary, our meta-analysis presented a low level of heterogeneity (approximately

Discussion
This systematic review with meta-analysis examined the effects of PA on adults with AD, as well as the estimated ESs of PA on cognition and ADL. Compared to previous meta-analyses, there were six studies [36,39,43,44,49,50] newly included in this review. Thus, a total of 945 participants (34.6% male) were involved in 16 RCTs. Overall, the methodological quality of the included studies was high. The findings showed that PA significantly benefited global cognition and ADL in adults with AD.
Although many meta-analyses have indicated the effects of exercise on adults with AD [17,27,53], there are differences in selection criteria between our study and previous studies. Due to the different mechanisms of dementia, abnormal protein deposits may coexist with neuro-vasculature at different stages of dementia and influence brain function [54]. For example, Lewy body dementia is associated with the α-synuclein protein; AD is associated with the β-amyloid and tau-proteins; vascular dementia is attributed to a lack of blood flow [55]. Therefore, we included participants with only Alzheimer's type of dementia in this current study. Some meta-analyses intended to target people with AD, but the actual analysis included people with different types of dementia [17,27,53]. Zhu et al. conducted the meta-analysis by including both RCTs and non-RCTs, which directly increased the risk of bias [53]. Moreover, only exercise intervention was considered to meet the inclusion criteria. Some meta-analyses have combined the exercise with other therapies (e.g., psychotherapy) in the experimental group, which influenced the effectiveness of sole exercise [53,56]. Our study, compared to other systematic reviews, avoided the flaws of the above aspects and excluded some original studies. In this context, the findings from previous studies may not necessarily generalize the specific individuals with AD. On the contrary, our meta-analysis presented a low level of heterogeneity (approximately I 2 = 28%, p > 0.1) by implementing strict inclusion criteria, showing reliable evidence of the effectiveness of PA on global cognition and ADL in adults with AD.
With regard to global cognition, all included studies, except for two studies using ADAS-Cog [39] and ERFC [42], examined the effect of PA on global cognition using the MMSE instrument. Both assessment tools have similar components (i.e., memory, attention, orientation, etc.) to evaluate global cognition. Our findings indicated that PA led to a significantly moderate improvement in global cognition (SMD = 0.42) for AD patients compared with the control group. This result was in line with previous meta-analyses, which demonstrated that exercise intervention contributed to an increase in global cognition amongst AD and dementia patients [54,57]. Although our result (SMD = 0.42, I 2 = 30%) is slightly smaller than those of the previous study (SMD = 0.48, I 2 = 85%) [54], the robustness of our result with lower heterogeneity is assured by including new RCT studies published in recent years. However, the PI for global cognition ranged from -0.04 to 0.85, implying that future research projects are likely to report a larger effect in the PA group, although all trials may not show a positive effect. Therefore, the results of this study are useful for researchers or therapists to better target intervention goals and methods, and for patients to know what to expect.
Additionally, subgroup analysis showed that exercise session time showed a significantly positive correlation with global cognition (p = 0.03), and exercise for 30-45 min had a more significant effect than exercise for more than 45 min. However, due to the moderate heterogeneity and a limited number of studies in interventions lasting 30-45 min, the evidence from the subgroup analysis of exercise session time does not very strongly determine whether an exercise session time of 30-45 min is better than one of more than 45 min. Jia et al. found that, in dementia patients, the positive effect on cognition was greater with an exercise session time of up to 30 min compared with one of more than 30 min [58]. Indeed, the included studies show that performing 30-45 min of exercise, compared with more than 45 min, produces better effects on global cognition. According to the current study, we moderately suggested that practicing moderate-intensity PA 3-4 times per week for 30-45 min for more than 12 weeks positively affected global cognition in adults with AD. However, future studies are warranted to verify our recommendation.
In terms of ADL, the ADCS-ADL is usually used to evaluate primary and operational activities of daily living for individuals with AD. In this current meta-analysis study, six studies directly reported ADL scores using the ADCS-ADL scale [36,38,39,41,49,50], and one study examined the ADL with the Barthel Index [44]. In older adults, impaired cognitive function is associated with difficulties in goal-directed activities (e.g., dressing and household duties) [59]. Reversely, improved cognitive performance leads to an improvement in physical function [59]. After participating in PA, participants showed a significant improvement in their ADL. The results from this current meta-analysis revealed that PA might have a considerable benefit in improving ADL in adults with AD. The calculated ES for ADL was 0.56, which is considered moderate. Because of the limited number of studies, the PI analysis for PA and ADL was not performed. Our finding is in line with previous review studies [53,57], which suggested that PA improved executive function and enhanced the ability of ADL in older adults. Meanwhile, exercise can adequately protect against a decline in ADL by improving skeletal muscle quality [60]. Given the limited number of included studies, subgroup analyses were not conducted to examine the impact of exercise dose on ADL. Further research investigating PA's effects on improving ADL is needed. It is possible to make firm recommendations on an optimal PA strategy for AD patients.
This meta-analysis was an updated and comprehensive investigation of the beneficial effects of PA on global cognition and ADL. RCTs and sole AD patients were included, which avoids the influence of the bias of non-RCTs or an inconsistency in population. In particular, sensitivity analysis indicated that the findings maintained robust. These findings are vital and timely to help clinicians or caregivers provide exercise guidance to AD patients during physical rehabilitation since PA is a cost-effective intervention for all age groups. It is possible to develop a public health policy to help AD patients prevent cognitive decline. In addition, PA can lead to other health benefits, such as reduced depression and improved quality of life; depression and reduced quality of life are common in AD patients [61,62]. In contrast, the most effective PA program for AD patients is yet to be developed. Long-term participation in PA is more difficult for this population because older adults initiate PA programs and commonly discontinue exercise within half a year [63]. Thus, establishing methods and strategies to encourage AD patients to maintain long-term exercise is needed in future studies.
Several limitations in this meta-analysis should be considered. First, only articles published in Chinese and English were included, likely resulting in language bias. Second, we only examined global cognition and ADL without analyzing other indicators, such as the subdomains of cognition (attention, processing speed, etc.). Further, the accuracy of these outcomes needs to be examined using other sophisticated instruments (i.e., magnetic resonance imaging) due to the possible bias of self-reported results.

Conclusions
To examine the pooled effects of PA on global cognition and ADL in adults with AD, our meta-analysis study included articles with a relatively high methodological quality.
The findings from this current study support that both aerobic exercise and mixed exercises benefited global cognition and ADL in patients with AD. Further research with large sample sizes and a high methodological quality is needed to acknowledge these findings.

Conflicts of Interest:
The authors declare that there is no conflict of interest.