A meta‐analysis of randomised controlled trials of physical activity in people with Alzheimer's disease and mild cognitive impairment with a comparison to donepezil

Physical exercise may benefit people with Alzheimer's disease (AD) and mild cognitive impairment (MCI). However, randomised controlled trials (RCTs) of exercise have shown conflicting findings and it is unclear if positive outcomes are comparable to a commonly used cholinesterase inhibitor, donepezil.


| INTRODUCTION
A growing number of studies have investigated the effects of exercise in people with Alzheimer's disease (AD) and Mild cognitive impairment (MCI). AD is characterised by severe progressive memory impairments which are associated with deposition of amyloid plaques (Aβ) in extracellular spaces leading to cortical dysfunctions and neuronal loss. 1,2 These deficits can be detected throughout late lifespan and often in individuals with MCI 3 which may be a prodrome to AD. There is an increasing interest in non-pharmaceutical interventions due to the current lack of disease-modifying drugs and long-term effective treatments. [4][5][6][7] A common class of medications used to treat dementia are acetylcholinesterase inhibitors. One of the most commonly prescribed cholinesterase inhibitors is donepezil which has been found to ameliorate cognitive symptoms in AD 8,9 but it is not recommended for the treatment of MCI. 10 After 6 months from the administration of these drugs in dementia, cognition tends to decline further 11 and it is acknowledged that these medications primarily exert a palliative effect without counteracting neurodegeneration. 6,12 Moderate and high intensity exercise has been demonstrated to diminish the progression of neuropsychological deficits in both AD and MCI. [13][14][15] Recent findings show that exercising increases levels of brain-derived neurotrophic factor (BDNF), an essential component for neuronal growth and neuronal plasticity. 16 Research in transgenic mice mimicking AD pathology has demonstrated that pharmacologicallyinduced neurogenesis in the mice's adult hippocampus ameliorates AD symptoms only if the mice had been physically active. 17 These promising findings suggest that exercising could alleviate or delay cognitive impairments by increasing neurogenesis. Results from the large Dementia and Physical Activity (DAPA) randomised controlled trial reported no benefits of physical activity on any cognitive domain 18 ; whilst other studies have suggested small to moderate positive effects on cognitive functions. 19 Previous meta-analyses examining the effect of physical activity on people with AD and MCI, [20][21][22][23][24][25][26] have shown that exercise compared to a control arm has beneficial effects, but it is not clear how this compares to pharmacological treatments. A recent meta-analysis examining the impact of both exercise and medications in AD and MCI revealed that medications have a small impact on cognition primarily in AD, whilst physical activity was shown to improve cognition in both clinical groups. 27 While this meta-analysis had a number of important strengths its main focus was on drug studies and it included a relatively small number of The aim of this meta-analysis is to strengthen the existing knowledge on the effects of physical activity on cognition in AD and MCI and to compare this to the effect of a common acetylcholinesterase inhibitor medication (i.e., donepezil). There has been an expanding interest in non-pharmacological interventions for these conditions since the study by Ströhle and colleagues 27 and this metaanalysis intends to provide at present a comparison between physical activity and donepezil. We also aim to address some of the weaknesses in the design of previous meta-analyses. Only randomized controlled trials (RCTs) are included to reduce the effects of bias and confounding. In line with prior research, 14,15,[20][21][22] strict diagnostic criteria for both groups and a specific definition of physical activity 28 are applied to reduce heterogeneity. The quality of the included studies is assessed according to established scale of Gates et al. 15 The results from specific cognitive scales are analysed separately rather than combined to reduce heterogeneity and to allow a comparison between exercise and donepezil for each reported measure where available. We have applied a method of calculating effect sizes recommended by Morris 29 who has systematically examined the accuracy and stability of effect sizes from repeated measures designs.
As cognitive decline may be associated with age, 1,20 and poor cognitive abilities at baseline 30 we have examined these potential moderators in a meta-regression. In line with prior research, we hypothesized that physical activity would be beneficial for cognition in people with AD and that exercise would be particularly effective in people with MCI as this disorder is earlier in the disease process. In comparison with medication we predicted the beneficial effect of exercise would be of a similar magnitude to the effects of donepezil.

| Identification of exercise and donepezil randomised controlled trials
The PRISMA guidelines were followed for this meta-analysis. 31 Two systematic searches of randomised controlled trials (RCTs) for exercise and for donepezil were conducted on Medline, Embase, PsycINFO, PsycARTICLES Full Text through the OVID database and also SCOPUS in October 2019. Free-text words and subject headings were employed.
To identify RCTs that examined the effects of physical activity broad search terms were used: 'Alzheimer's disease' or 'Alzheimer's*' or 'Mild cognitive impairment*' and 'physical activity' or 'physical exercise' or 'aerobic fitness' and 'randomized controlled trial.' To identify RCTs that examined the effects of donepezil the following broad search terms were used: 'Alzheimer's disease' or 'Alzheimer*' or 'Mild cognitive impairment*' and 'donepezil' or 'Aricept' and 'randomized controlled trial.' Full search strategies can be found in Supplementary Material 1 (S1). Two PRISMA flowcharts were created for each search strategy ( Figure 1, Figure 2).

| Inclusion and exclusion criteria
Only RCTs were included. Longitudinal, cross-sectional, case-control studies, systematic reviews, meta-analyses, cross-over randomized controlled trials and non-randomized controlled trials were excluded.
Studies were included if they provided evidence of AD diagnosis F I G U R E 1 PRISMA flowchart of the study selection process for physical activity randomised controlled trials (RCTs) [Colour figure can be viewed at wileyonlinelibrary.com] PISANI ET AL.

| Outcome measures
The mini-mental state examination (MMSE) and Alzheimer's disease assessment scale-cognitive subscale (ADAS-Cog) were chosen as the outcome measures because they were most commonly reported by studies. To ensure the different meta-analyses were sufficiently powered, outcomes were included if there were three or more independent studies that reported a mean and SD in both the control, donepezil and exercise groups. Where follow-up means and SDs were not available authors were contacted for this information.

| Data extraction
Sample characteristics (e.g., gender, age and diagnosis), type and Where confidence intervals or standard errors were provided, these were converted into SD.

| Quality assessment of studies
The Physiotherapy evidence database scale (PEDro) 15,34 is an 11-item quality rating scale of RCTs where each item is marked 0 or 1, with 1 indicating higher quality. The scale assesses eligibility criteria (item 1); internal validity (items 2-9) which include concealed allocation (item 3), blinding of participants (item 5), and of the assessor (item 7); and the appropriate outcome reported (items 10-11). Item 6 is blinding of the therapist, which is relevant for exercise studies, but is not applicable for pharmacological studies.
Therefore, for RCTs of donepezil this item was removed, hence not coded, to indicate no additional risk of bias.

| Calculation of effect size
Investigators have applied different methods when calculating effect sizes from repeated measure designs. Morris 29 has extensively investigated these different methods in terms of precision, robustness and bias, and has proposed the optimal methodology uses the pre and post intervention means and pre-intervention SD from both the treatment and control groups. Thus, we calculated the Hedges effect size (which is the Cohen effect size with a correction for bias from small sample sizes) and its variance from equations provided by Morris 29

| Meta-analysis
Effect sizes were pooled using a random-effects inverse-weighted variance model. 35 When conducting a meta-analysis there is balance between maximising the number of studies to increase power and being more selective to reduce heterogeneity. In this metaanalysis we performed separate meta-analyses for AD and MCI where there was a sufficient number of studies. Between-study heterogeneity was analysed using Cochran Q test and I 2 which indicates the percentage of total variation across studies due to heterogeneity. 36,37 Publication bias was assessed using Egger's regression 38 when at least five studies were included to ensure that the test was sufficiently powered. The calculation of effect sizes was performed using meta-analytical equations entered into Excel. These equations are identical to the METAN 39 command in STATA, 40 which is commonly used in meta-analyses publications. In terms of validation, the method has been used in parallel with STATA in a number of meta-analyses 41,42 and produced the same results.

| Meta-regression of mini-mental state examination
A random effects meta-regression was conducted on MMSE change in the AD and MCI groups using METAREG 43 in STATA 40 to investigate the contribution of age, baseline MMSE, on physical activity and donepezil. MMSE change was chosen because it was the most commonly reported outcome variable.

| Sensitivity analysis
To examine the strength of the results in relation to the variability of meta-analysis methods, a sensitivity analysis was conducted adjusting Rho from the standard value of 0.7 to a lower value of 0.5 and a higher value of 0.9.

| Study characteristics
Physical activity: 2110 articles were initially identified, and after applying inclusion and exclusion criteria, 19 RCTs were included ( Figure 1; AD, n = 7; MCI, n = 13). One study included both people with AD and MCI and conducted their analyses separate for each clinical group. 44 In total, 1793 individuals took part in these 19 RCTs, 970 allocated to physical activity and 823 to control arms (Table 1)

| Quality assessment or exercise randomised controlled trials
Physical activity: Across the 19 RCTs, the average PEDro score was 7.58 out of a maximum of 11 with a range between 6 and 9 points. A common reason for reduced scores was the lack of blinding participants and professionals administering the physical activity to the sample; as blinding of participants was deemed not feasible due to the nature of the intervention. Blinding of the researchers conducting the cognitive assessments was present in 14 (73%) out of the 19 studies (Table 4).
Donepezil: Across 18 RCTs, the average PEDro score was 9.33 out of a maximum of 10, after removing the item that assessed the blinding of the therapist (see methods). The range was between 8 and 10 points. Methodological weaknesses were primarily related to concealed allocation. Studies were double-blinded and always randomised (Table 5).

| Meta-analysis-physical activity
In AD, physical activity was associated with an improvement in MMSE compared to the control arm (g = 0.458, p = 0.013; Figure 3), however this result was associated with small sample bias (p = 0.02). Only two studies employed the ADAS-Cog, precluding a meta-analysis. Physical activity also had a significant beneficial effect on cognitive functions assessed with the MMSE compared to the control condition for MCI (g = 0.631, p = 0.001) (S5, Figure 4). There was no significant effect of physical activity on the ADAS-Cog (p = 0.399) ( Figure 5; Table 3).

| Meta-regression
In the AD group, there were no significant moderating effect of age  Table 6).

| Sensitivity analysis
There was no change in results classified as significant or nonsignificant when rho = 0.9, or when rho = 0.5.
For people with MCI, there was a trend towards significance for improved cognition assessed with the ADAS-Cog (g = −0.130, p = 0.059; Figure 8).

| Meta-regression
There were no significant moderating effects of age or baseline MMSE scores for AD or MCI groups (all p > 0.05; Table 6).   For the AD group, exercise was associated with nominally larger effect sizes than donepezil for MMSE, however the exercise metaanalysis also showed evidence of small sample bias which may include publication bias, so additional studies will be needed to confirm the difference in effect size. In the MCI group, exercise was associated with a robust improvement in MMSE, but there was no evidence that donepezil had a beneficial effect on cognition in this clinical group.
Our findings are in line with previous research but are arguably more robust as the meta-analysis was limited to only RCTs, 14,15,20,21,24,25 pre-and post-treatment measures have been taken into account using Morris' recommendations 29 and the cognitive measures were meta-analysed using the same scales rather than pooling different instruments. There was a significant impact of baseline MMSE scores for people with MCI, whereby those whose MMSE scores were low experienced more benefit from practising physical activity. An alternative explanation for these results could be due to a compensation effect, 81 whereby individuals who are cognitively poor may benefit more from physical activity, that is the effect of this intervention is stronger on this population because of the extend of their cognitive impairment (compensation). 81

| Strengths, limitations and future directions
There are a number of limitations to this study. Significant heterogeneity was present across most of the physical exercise analyses.