Meta-Analysis of Randomized Controlled Trials on the Efficacy and Safety of Donepezil, Galantamine, Rivastigmine, and Memantine for the Treatment of Alzheimer’s Disease

To study the impact of donepezil, rivastigmine, galantamine, and memantine on cognitive, functional, behavioral, global changes and adverse effects in patients with mild, moderate and severe Alzheimer’s disease (AD), we screened the literature published before September 2017 in the Pubmed, Embase, Cochrane library and Web of Science Electronic databases according to the inclusion criteria. Thirty-six studies were finally determined from 1560 preliminary screened articles. The AD Assessment Scale-cognitive Subscale (ADAS-cog), AD Cooperative Study-Activities of Daily Living (ADCS-ADL), Neuropsychiatric Inventory (NPI), and Clinician’s Interview-Based Impression of Change Plus Caregiver Input scale (CIBIC+) were used as valid endpoints. Of the 36 trials included, meta-analyses of these placebo-control trials showed that there were significant differences between the donepezil, rivastigmine and placebo groups using ADAS-cog, ADCS-ADL, and CIBIC+. Meta-analyses of these placebo-controlled trials showed that there were significant differences between the galantamine and placebo groups using ADAS-cog, ADCS-ADL, NPI, and CIBIC+. These observations suggest that memantine is beneficial for stabilizing or slowing the decline in ADAS-cog and ADCS-ADL19 changes in AD patients. However, there was no significant effect according to the ADCS-ADL23, NPI, and CIBIC+ tests, which indicated that memantine treatment has no significant effect on these cognitive aspects of AD patients. Different effects of donepezil, rivastigmine, galantamine, or memantine on AD were found in this study. According to the results, we conclude that galantamine is effective in treating all aspects of AD and is the first choice for the treatment of AD. However, due to limited data, we should consider additional data to obtain more stable results.


INTRODUCTION
The relationship between cognitive dysfunction or impairment and Alzheimer's disease AD has been reported in the literature (Stern et al., 1990;Chen et al., 1998;Perry and Hodges, 2000;Caro et al., 2002;Pereira et al., 2008). Meanwhile, cholesterol esterase inhibitors (ChEs) and memantine (Supplementary Table 1), which is a non-competitive N-methyl-D-aspartate (NMDA) receptor antagonist, can normalize dysfunctional glutamatergic neurotransmission (Parsons et al., 2013), which has shown effective efficacy in the treatment of AD. Since the intrinsic mechanism of acetylcholinesterase inhibitors (AChE-Is) requires a sufficient amount of residual endogenous acetylcholine, which is available, the therapeutic efficacy is expected to decrease with the severity of dementia. Cholinesterase inhibitors, which can reduce acetylcholine breakdown in the brain, are widely considered as a treatment options for AD (Doody et al., 2001). Therefore, greater atrophy in the brain regions that are responsible for the cholinergic pathway was found in those patients who had no response to donepezil (Bottini et al., 2012). It has been reported that donepezil is a type of AChE-I that improves cerebral blood flow (CBF), as well as its primary effect on memory function (Kogure et al., 2017). Galantamine is a newly available cholinergic drug that counteracts AD by specifically and reversibly inhibiting acetylcholinesterase (AChE) and altering the nicotinic cholinergic receptors, thereby subsequently reducing central cholinergic neurotransmission (Tariot, 2001). The deterioration of cognitive function in patients with AD appears to be mediated by the use of cholinergic drugs such as rivastigmine (Birks, 2006). Rivastigmine is a novel brain-selective inhibitor of "pseudo-irreversible" AChE, whose metabolism is almost completely independent of the cytochrome P450 system (Sramek et al., 1996). Memantine protects neurons against the overstimulation of NMDA receptors, which occurs in AD and thus causes glutamate-and calcium-mediated neurotoxicity (Jiang and Jiang, 2015). Evidence of the efficacy of memantine has been shown primarily in patients with moderate or severe AD (Areosa et al., 2005). The etiology and pathogenesis of AD are not well understood, but central cholinergic neurons are found to be impaired in AD patients with low choline intake and reduced choline synthesis (Inestrosa et al., 2005). The central cholinergic system plays a key role in regulating learning, memory and attention. AChE-Is is a major drug in the clinical management of AD (Tan et al., 2014), which can improve cognitive function by prolonging the duration of action of acetylcholine (Ach) in the CNS to improve cholinergic function and slow down memory loss. AchE-Is such as donepezil and galantamine show a significant effect on mild-to-moderate AD (Winblad et al., 2001;Raskind et al., 2004). Donepezil, rivastigmine and galantamine, which belong to the group of ChE inhibitors (ChE-Is), are capable of cognitive, functional and behavioral improvement; however, none of them has been shown to be effective in the progression of AD (Zemek et al., 2014). Currently, AchE-Is, including donepezil, rivastigmine, and galantamine, are standard treatments for slowing disease progression (Li et al., 2015). In addition, high dropout rates and adverse-effectinduced dropouts were observed in randomized clinical trials of these drugs. Therefore, the safety of ChE-Is and memantine has been proposed.
We conducted a systematic review and meta-analysis of donepezil, galantamine, rivastigmine and memantine in AD to elucidate the efficacy and safety of these drugs. We sought to elaborate on previous reviews and include a broad range of outcome measures to determine the extent to which these drugs have varying degrees of effects on cognitive, behavioral and functional impairment in AD patients at different stages of severity.

Selection Criteria
The inclusion criteria for the meta-analyses were as follows: (1) full-text publications written in English; (2) double-blind, parallel-group, placebo-controlled, with random assignment to donepezil, rivastigmine, galantamine, or memantine; (2) inclusion of patients with or probably with AD diagnosis, according to the fourth edition of the Mental Disorders Diagnostics and Statistics Manual (DSM-IV) and the National Association of Nervous and Communicative Disorders and Stroke/Alzheimer's Disease Institute of Standards (NINCDS-ADRDA) (McKhann et al., 1984;American Psychiatric Association, 2000); (3) includes treatment duration for at least 52 weeks and at least one measure that reflects changes in cognitive, functional, behavioral or global assessment of change, as well as the number of adverse events (AEs) that led to dropout, and AE changes; (4) drug dosage and dosage form specifications. Studies with fatal defects in research design or data analysis were excluded, and trials with no readily available data were also excluded.
Studies were excluded for the following reasons: (1) not randomized controlled trials, such as case reports, reviews, and meta-analysis; (2) family-based studies; (3) lack of original data, such as meeting abstracts, and case reports/series; (4) non-human studies; and (5) publication in a language other than English.

Search Strategy and Selection Criteria
Data abstraction was performed as a collaboration between two researchers using standard data extraction by discussing with other team members or by asking for the exact data from the original investigators. For missing data, we sought missing information and essential clarification from the authors. For measure variables, there is an approximate or direct algebraic relationship with standard deviation (SD), which we obtained from the standard error, confidence interval, t-value, or p-value related to the differences between the two sets of means.
We obtained the following baseline variables from each study: sample size, age, sex, race, design, dosing, blinding, duration of the trial, baseline cognitive score (Mini-Mental State Examination, MMSE), random numbers, secondary outcomes, and AEs that led to dropouts during the double-blind trials.

Types of Outcome Measures
The measurement scales used in the tests varied. Therefore, we recorded measurement scales based on the general areas being assessed, namely, cognition, function, behavior and global assessment of change. Next, we attempted to determine a single measurement scale, which is the most commonly used key outcome measure in each area. We engaged the ADAScog, Alzheimer's Disease Cooperative Research-Activities of Daily Living (ADCS-ADL), NPI and Clinician's Interview-Based Impression of Change Plus Caregiver Input scale (CIBIC+) as primary measurements to assess the effects of different drugs on the cognition, function, behavior, and global assessment of change (Supplementary Table 2). The Alzheimer's Disease Cooperative Study Activities of Daily Living 19-or 23-item scale (ADCS-ADL 19/23), which is based on interviews with caregivers or others close to the patients, was used as a scale to assess how the patients cope with daily activities. The 19-item subset was used for patients with moderate to severe AD, and the 23-item subset was used for patients with mild to moderate AD. In addition, to assess the efficacy and safety of these drugs, we recorded discontinuations of the trials due to AEs and dropouts due to adverse effects and other reasons.

Statistical Analysis
Meta-analysis was performed using the Review Manager 5.3 software. For continuous data collected using the same measurement scale (e.g., cognition and behavior), we calculated the mean difference (MD) or standardized mean differences (SMD) with 95% confidence intervals (CIs) for changes from the baseline or final values. For dichotomous clinical outcomes, dropouts, and AEs, we performed odds ratios (ORs), absolute risk differences, 95% CI and p values to assess the efficacy and safety of the studied drugs. Heterogeneity was assessed using the Cochran Q-statistic and I 2 tests (Doody et al., 2001). I 2 approximates the ratio of the total variance in effect estimates due to heterogeneity rather than sampling error. Heterogeneity of the index is considered when p < 0.10 and I 2 > 50%. The latent retrieval bias was evaluated by the MD or SMD funnel plot in the main results of each test.
In a pooled study of Activities of Daily Living Inventory 23 (ADCS-ADL 23 ), the pooled analysis of both studies showed that no significant benefit was found with memantine treatment [SMD = 0.00, 95% CI (-0.11, 0.12); p = 0.93] (Supplementary Figure S1B). The heterogeneity among most pooled studies was low (p = 0.99, I 2 = 0%). The funnel plots (data not shown) did not show a symmetric distribution, which indicated no hint of publication bias.

Galantamine
We pooled NPI data from three articles (Tariot et al., 2000;Rockwood et al., 2001;Brodaty et al., 2005) (Figure 4B), which contained detailed information on the baseline and final observation times for galantamine and placebo. Galantamine treatment showed a significant effect on the behavioral outcome when assessed by NPI [SMD = -0.15, 95% CI (-0.24, -0.06); p = 0.001]. The heterogeneity among most pooled studies was low (p = 0.91, I 2 = 0%). The funnel plots (data not shown) did not show a symmetric distribution, which indicated no hint of publication bias.

Rivastigmine
We pooled NPI data from one study (Winblad et al., 2007) (Figure 4C), which contained detailed information on the baseline and final observation times for rivastigmine and placebo. Rivastigmine treatment showed no significant effect on behavioral outcome when assessed by NPI [MD = -0.50, 95% CI (-2.68, 1.68); p = 0.65].

Rivastigmine
Two articles (Rosler et al., 1999;Feldman and Lane, 2007) based on interviews and CIBIC+ used the clinician's impression of change to evaluate the clinician's global impression (Figure 5C). We pooled the results of studies and found a significant difference in the rivastigmine group [OR = 2.08, 95% CI (1.53, 2.83); p < 0.00001]. The heterogeneity among most pooled studies was low (p = 0.49, I 2 = 0%). The funnel plots (data not shown) did not show a symmetric distribution, indicating no hint of publication bias.

Memantine
Two articles (Tariot et al., 2004;Bakchine and Loft, 2008) based on interviews and CIBIC+ used the clinician's impression of change to evaluate the clinician's global impression (Supplementary Figure S1C). We pooled the results from the studies, and no significant difference was found in the memantine group [OR = 1.23, 95% CI (0.85, 1.78); p = 0.28]. The heterogeneity among most pooled studies was low (p = 0.70, I 2 = 0%). The funnel plots (data not shown) did not show a symmetric distribution, which indicated no hint of publication bias.

Donepezil
When any AE was considered, there was a statistically significant association between the donepezil group and the placebo group [OR = 1.24, 95% CI (1.04, 1.49); p = 0.02] (Figure 6A). The heterogeneity among most pooled studies was low (p = 0.87, I 2 = 0%). The funnel plots (data not shown) did not show a symmetric distribution, indicating no hint of publication bias. Overall, when considering the number of dropouts caused by any reason, no difference was found between patients treated with donepezil and placebo [OR = 1.12, 95% CI (0.91, 1.37); p = 0.28] (Figure 6B). The heterogeneity among most pooled studies was low (p = 0.20, I 2 = 25%). The funnel plots (data not shown) did not show symmetric distribution, which indicated a hint of publication bias. When we deleted the greatest positive outlier (Gault et al., 2015), the merged effect size was not statistically significant [OR = 1.10, 95% CI (0.90, 1.35); p = 0.35]. The heterogeneity among most pooled studies was low (p = 0.18, I 2 = 28%). When we eliminated the most unfavorable point (Krishnan et al., 2003), the merged effect size was not significant [OR = 1.15, 95% CI (0.94, 1.41); p = 0.19]. The heterogeneity among most pooled studies was low (p = 0.24, I 2 = 21%). When the number of dropouts was caused by adverse effects, a significant difference was found between the donepezil group and the placebo group [OR = 1.58, 95% CI (1.22, 2.05); p = 0.0006] (Supplementary Figure S2). The heterogeneity among most pooled studies was low (p = 0.80, I 2 = 0%). The funnel plots (data not shown) did not show a fairly symmetric distribution, which indicated no hint of publication bias.
Among these side effects, some gastrointestinal and nervous system side effects such as nausea, vomiting, diarrhea, anorexia, dizziness, depression and headache were observed. Since some studies did not report these events in detail, we did not separately compare the incidence of AEs.

DISCUSSION
Different methods of measurement were used in assessing treatment outcomes. We divided the impact into four parts, cognitive function, functional outcome, behavioral outcome, and global assessment. In this article, we presented a metaanalysis of the effects of donepezil, galantamine, rivastigmine, and memantine (Supplementary Table 1) on mild-to-moderate, moderate-to-severe, and severe AD using the ADAS-cog (cognitive function), ADCS-ADL (functional outcome), NPI (behavioral outcome) and CIBIC+ scores (global assessment). We obtained data from 6611 AD patients across 36 trials. Meta-analyses of these placebo-controlled trials showed that there were significant differences between the donepezil and placebo groups using ADAS-cog, ADCS-ADL, and CIBIC+, between the galantamine and placebo groups using ADAS-cog, ADCS-ADL, NPI, and CIBIC+, between the rivastigmine and placebo groups using ADAS-cog, ADCS-ADL, and CIBIC+, and between the memantine and placebo groups using ADAS-cog, and ADCS-ADL 19 (Supplementary Table 6). This observation suggests that donepezil is beneficial for stabilizing or slowing the decline in cognitive function, functional outcome, and global assessment change in AD patients. Analysis of the entire database showed consistent results, which indicated positive results with the donepezil treatment and improvement in the condition of mild-to-moderate, moderate-to-severe, and severe AD patients. To our knowledge, this is the first meta-analysis to reveal the favorable performance of donepezil with mild-tomoderate, moderate-to-severe, and severe AD. This observation suggests that galantamine is beneficial for stabilizing or slowing the decline in cognitive function, functional outcome, behavior outcome and global assessment change in AD patients. To our knowledge, this is the first meta-analysis to reveal the favorable performance of galantamine with mild-to-moderate, moderate-to-severe, and severe AD. This observation suggests that rivastigmine is beneficial for stabilizing or slowing the decline in cognitive function, functional outcome, and global assessment change in AD patients. Analysis of the entire database showed consistent results, which indicated positive results with rivastigmine treatment and improvement in the condition of mild-to-moderate, moderate-to-severe, and severe AD patients. There was no significant effect on the outcome of NPI-assessed behavioral therapy, except galantamine. This observation suggests that memantine is beneficial for stabilizing or slowing the decline in cognitive function, and functional outcome change in AD patients. However, there was no significant effect according to the ADCS-ADL 23 , NPI, and CIBIC+ tests, which indicated that memantine treatment has no significant effect on these cognitive aspects of AD patients.
Safety and tolerability are as important as the effectiveness of the interventions in clinical research. In the current metaanalysis, when any AE, dropout caused by any reason, or adverse effect was considered in all patients or subgroups, significant differences were found between the galantamine or rivastigmine groups and the placebo group, which indicated that donepezil and galantamine were not sufficiently safe and tolerable for AD treatment. However, no significant difference was found in all patients or subgroups when the donepezil group and the placebo group were compared, which indicates that donepezil is safe and tolerable enough for AD treatment. When any AE, dropout caused by any reason, or adverse effect was considered, no significant difference was found between the memantine group and the placebo group in all patients or subgroups, which indicates that memantine is safe and tolerable enough for AD treatment. Doody et al. (2007) reported a lower rate of discontinuation due to AEs in the memantine group and a lower discontinuation rate than the placebo group. On the other hand, Yang et al. (2013) and Jiang and Jiang (2015) did not find the differences.
Compared with the published meta-analysis (Loveman et al., 2006;Raina et al., 2008;Bond et al., 2012;Di Santo et al., 2013), galantamine has potent therapeutic effects on all aspects of the treatment of AD, but the other three drugs do not have effective therapeutic effects on some aspects (Supplementary Table 6). Therefore, based on the current statistical conclusion, we have determined that we prefer galantamine for the treatment of AD. However, due to limited data, we should also consider additional data to obtain more stable results.
This analysis is subject to the number of limitations. First, although all studies are short-lived, the duration of the study is still a variable and may be a factor to consider. Second, the results of neuropsychiatric symptoms are based on a relatively small number of trials. This can affect the wider range of estimates. Third, meta-analysis data come only from published scientific literature, and some negative results and non-statistical data are difficult to publish; therefore, there is publication bias. Fourth, the genetic backgrounds of these patients involved in our analysis were different, which could potentially influence the rivastigmine treatment. Finally, some of our trials used flexible drug doses. Overall, the results indicate benefits in cognition but the efficacy on functional, behavioral, and global change symptoms is questionable in patients with mild to moderate to severe AD. Our results might suggest a possible perspective for anti-dementia drug trials, such as increasing placebo effects over time and heterogeneity of neuropsychiatric symptoms in AD. The results of this study still need to be confirmed by further studies.

CONCLUSION
Our analysis is the first attempt to incorporate available direct or indirect evidence to evaluate the efficacy and safety of 4 drugs in the treatment of AD. The results suggest that donepezil exhibited a significant positive efficacy with respect to cognition, function, behavior, and global change. However, the efficacy of rivastigmine or galantamine in behavioral outcome is questionable in patients with mild to moderate to severe AD. The efficacy of memantine on global assessment is questionable in patients with mild-to-moderate and moderate-to-severe AD. This review shows that donepezil, galantamine, rivastigmine, and memantine can delay cognitive impairment in patients with mild-to-moderate-to-severe AD for at least 52 weeks. Based on the current statistical conclusions, galantamine is effective in treating all aspects of AD and may be the first choice in the treatment of AD. However, due to limited data, we should also consider additional data to obtain more stable results.

AUTHOR CONTRIBUTIONS
WZ and PZ designed the study. D-DL performed the experiments and data analysis with the help of Y-HZ, WZ, and PZ. D-DL, WZ, and PZ wrote the manuscript.

FUNDING
This work was supported in part or in whole by the National Natural Science Foundation of China (31301108) and (31401215)