FormalPara Key Points

This network meta-analysis of EGb761, three cholinesterase inhibitors (ChEIs), including donepezil, galantamine and rivastigmine, and memantine focused on randomized double-blinded trials in patients with mild-to-moderate Alzheimer’s disease (AD) or probable AD without any other types of dementia or neurological disorders.

EGb761 and memantine did not provide benefits for cognition, function, behavioral symptoms, clinical global impression, and tolerability compared to placebo and ChEIs.

Galantamine provided the greatest beneficial effects on cognition and behavioral symptoms, and donepezil provided the best clinical global impression and tolerability outcomes, with a similar benefit for cognition as galantamine; therefore, donepezil should be selected as the first-line treatment if not contraindicated.

1 Introduction

Alzheimer’s disease (AD), a progressive neurodegenerative disorder characterized by cognitive deterioration, functional impairment, and behavioral problems, is the most common form of dementia in the elderly. Approximately 30 million people with dementia live in low- and middle-income countries, and 60–70% of them have been diagnosed with AD [1, 2]. The primary goal of pharmacological interventions for AD is to slow functional impairment. The three cholinesterase inhibitors (ChEIs), donepezil, galantamine and rivastigmine, and the N-methyl-d-aspartate (NMDA) receptor antagonist memantine offer modest benefits to cognition and aspects of functioning and have been proposed as treatment options by guidelines in developed countries [3]. However, these medical treatments have a relatively high cost in developing countries, and their potential adverse effects can cause patient non-adherence. Therefore, complementary choices of natural products that are expected to be well-tolerated, such as extracts of Ginkgo biloba and antioxidants, have been widely used in patients with AD, particularly in developing countries [4].

EGb761, a standardized extract of Ginkgo biloba, has been registered in several countries as a phytomedicine or dietary supplement [5, 6] and is reimbursable by health insurance in some countries, including Germany [7, 8]. EGb761 has been hypothesized to restore age-related impairment of memory through a mechanism of action as an antioxidant that inhibits mitochondrial dysfunction that leads to neuronal cell death [9, 10]. During 2014–2017, several systematic reviews reported that EGb761, at a dose of 240 mg/day for at least 22 weeks, had beneficial effects in mixed patients with mild-to-moderate AD and vascular dementia [11,12,13,14]. A few head-to-head trials have compared EGb761 with ChEIs, and the results were inconclusive [15,16,17].

To our knowledge, treatment comparisons among multiple agents, including EGb761, ChEIs and memantine, within a single study have not been available to date. Two direct, pairwise meta-analyses (MAs) compared individual ChEIs or memantine with placebo [18, 19]. Two network meta-analyses (NMAs) compared active agents: one assessed three ChEIs for mild-to-moderate AD [20], and the other involved memantine in addition to the ChEIs for mild-to-severe AD [21]. The former included double-blind and open-label trials, and the latter did not restrict the analysis to randomized controlled trials (RCTs). The purpose of the present study was to compare the therapeutic benefits and tolerability of all active agents—EGb761, three ChEIs (donepezil, galantamine, and rivastigmine), and memantine—for mild-to-moderate AD on the basis of an NMA of RCTs with treatment concealment.

2 Methods

2.1 Search Method and Study Selection

An electronic search for relevant studies was conducted using PubMed, the Cochrane Controlled Trials Register, the Cochrane Database of Systematic Reviews, the WHO International Clinical Trials Registry, and ClinicalTrials.gov up to 30 June 2017. The search terms included “Alzheimer’s disease”, “dementia”, “Ginkgo biloba”, “EGb761”, “cholinesterase inhibitor(s)”, “donepezil”, “galantamine”, “rivastigmine”, and “memantine”. References in relevant systematic reviews were manually scanned for additional studies. Only articles in English were retrieved.

Eligible studies were double-blinded RCTs that compared EGb761 with placebo, ChEIs with placebo, or memantine with placebo, or compared the active agents. For EGb761, only trials that used the oral form of treatment were selected. Participants in eligible studies were diagnosed as having AD or probable AD based on the third and fourth editions of the American Psychiatric Association’s Diagnostic and Statistical Manual of Mental Disorders (DSM-III, IV) or the National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) criteria. Study eligibility criteria included trials involving AD and probable AD patients without other types of dementia or neurological disorders and trials reporting study endpoints specific to the mild-to-moderate AD state and having at least one cognition endpoint. Mild-to-moderate AD was operationalized as the baseline Mini-Mental State Examination (MMSE) scores within a range of 10–26 or the equivalent when using other measures, such as the Syndrome Kurz test (SKT) with a score range of 9–23. Trials were excluded if the reported outcomes of the study participants with AD could not be separated from those with dementia from other causes. Eligible studies had to prove therapeutic doses for an adequate duration. Dosages were required to be either fixed or titrated up to specified targets, as follows: 240 mg/day for EGb761, 5 and 10 mg/day for donepezil, 16 and 24 mg/day for galantamine, 12 mg/day for rivastigmine in oral form, 9.5 mg/24 h (10 cm2) for the rivastigmine patch, and 20 mg/day for memantine [22]. There was evidence supporting the effectiveness of EGb761 when it was used for more than 22 weeks [14]. For the ChEIs and memantine trials, the efficacy for AD was commonly evaluated after 12 weeks of treatment. The minimum treatment duration for the included studies was 22 weeks for EGb761 and 12 weeks for ChEIs and memantine. Other exclusion criteria included trials that allowed concomitant use of psychotropics, anticonvulsants, and antiparkinson agents with the study drugs and trials that had a sample size of fewer than 30 patients per arm.

2.2 Data Extraction and Quality Assessment

A training session on data extraction was conducted to assure that two of the researchers (OT and CL) had more than 98% agreement on the data abstracted based on a standardized data collection form. Both researchers (OT and CL) independently searched the literature. Based on the titles and abstracts, potential articles were screened in the full-text version for eligibility. For each selected article, the two researchers (OT and CL) independently extracted the predefined contents, including trial characteristics, patient characteristics, sample sizes, treatment regimens and outcomes, on the basis of an intention-to-treat analysis. If studies had both a double-blinded phase and an open-label phase, outcomes measured at the end of the double-blinded phase were used as the endpoints. Quality assessment was performed using the Cochrane risk of bias tool (RoB 2.0) [23, 24]. Any disagreements between the data abstractors were resolved by discussion, and other researchers (OW, TR, PL, and PL) were asked to resolve any remaining differences.

2.3 Study Outcomes and Measurements

2.3.1 Cognition

For trials testing ChEIs and memantine, cognitive function of patients with AD was usually assessed by a clinician using the Alzheimer’s Disease Assessment Scale–cognitive subscale (ADAS-cog). This scale consisted of 11 items, with a total score ranging from 0 to 70 [25, 26]. Higher scores indicated worse cognition, and a cutoff ≥ 18 indicated cognitive deficit. An improvement of at least 4 points was considered a clinical improvement in cognition [27]. For trials testing EGb761, cognition was most often assessed using the SKT [15, 28,29,30]. The SKT contained nine items, with a total score ranging from 0 to 27, and a higher score indicated more impairment. Transformation of SKT into ADAS-cog scores has been reported as ADAS-cog = 5.3 + 1.3 × SKT [31].

2.3.2 Functional Outcomes

The most common measure of functional impairment in patients with AD was the Alzheimer’s Disease Cooperative Study–Activities of Daily Living Inventory (ADCS-ADL). The 23-item version (ADCS-ADL23), with a total score of 0–78, was rated by a clinician using information supplied by caregivers [32]. The second most commonly used measure was the Disability Assessment for Dementia (DAD), an informant-rated questionnaire consisting of 40 items, with a total score of 0–100 [33]. Other functional measures included the Progressive Deterioration Scale (PDS), Interview for Deterioration in Daily Living Activities in Dementia (IDDD), Gottfries-Brane-Steen Activities of Daily Living Scale (GBS-ADL), Alzheimer’s Disease Activities of Daily Living International Scale (ADL-IS), and Nurnberger Alters-Beobachtungsskala (NAB). A higher score on the ADCS-ADL, DAD, and PDS indicated better function. In contrast, a higher score on IDDD, GBS-ADL, ADL-IS, and NAB indicated more impairment.

2.3.3 Behavioral Symptoms

Behavioral symptoms in patients with AD were most often assessed by the Neuropsychiatric Inventory (NPI), which measured both the frequency and severity of neuropsychiatric symptoms in the previous month [34, 35]. The original version of the NPI covered ten subdomains of behavioral functioning—delusions, hallucinations, agitation/aggression, dysphoria, anxiety, euphoria, apathy, disinhibition, irritability/lability, and aberrant motor activity—with each subdomain having one item scored 0–12, which summed to 0–120 in total [34]. In the 12-item version, two additional subdomains were night-time behavioral disturbances and appetite/eating abnormalities, which resulted in a total score of 0–144 [35]. Scores of less than 20, 20–50, and > 50 indicated mild, moderate, and severe behavioral disturbances, respectively.

2.3.4 Clinical Global Impression

The global impression of patients with AD was commonly assessed using one of four similar measures: the Clinician Interview-Based Impression of Change Plus Caregiver’s Input (CIBIC-plus), Clinical Global Impression of Change (CGIC), Alzheimer’s Disease Cooperative Study–Clinical Global Impression of Change (ADCS-CGIC), and Clinical Global Impression (CGI) scales. CIBIC-plus, the most common measure, was a semistructured interview covering four domains, including general, mental/cognitive, activities of daily living, and behavior, which guided the clinician’s judgment of global change. The CGIC was an interview-based global rating scale that quantified the clinician’s judgment of the magnitude of changes in overall impairment [36]. The ADCS-CGIC had a semistructured format allowing a clinician to gather necessary information from both patients and caregivers. Each of these measures had a score range of 1–7, which was assessed by comparing the clinical status at a particular visit with the baseline. Interpretation of the scores for these measures was consistent [37, 38]. Scores of 1, 4, and 7 indicated a marked improvement, no change, and worsening, respectively [38, 39]. Most trials reported the outcomes in terms of the number of patients who had clinical improvement or remained unchanged at the time of follow-up.

2.3.5 Tolerability

The tolerability of the treatments was assessed in terms of the percentage of patients withdrawing from the trials due to any adverse events. In addition, selected adverse events, including nausea, vomiting, diarrhea, headache, dizziness, weight loss, and anorexia, were assessed.

2.4 Statistical Analysis

For the continuous outcomes, including cognition, function, and behavior symptoms, which were assessed using several measures, differences between treatments and controls in changes from baseline were measured in terms of standardized mean difference (SMD). The binary outcomes, such as an improvement in clinical global impression, discontinuation of treatment, and specific adverse events compared with controls, were measured in terms of relative risk (RR).

We took a frequentist statistical approach for pooling treatment effects across individual trials using Stata’s mvmeta, a multivariate, random-effects MA and meta-regression [40, 41]. According to the multivariate approach, the dependent variables were contrast-based outcomes between treatments with a common control as the reference. Values of the treatment effects in the trials with missing references were replaced by a minimal value for the mean (0.001 by default) with a very large variance (10,000 by default). This is a technique to make all contrasts with the common control possible by augmentation with such a non-missing but ignorable value [42]. The random-effects model to capture an observed outcome as contrast of a treatment of interest in each trial was composed of four components: (1) effect of the treatment as contrasted to the reference; (2) heterogeneity of the treatment effect across trials (or between-study variance) within the same study designs; (3) inconsistency in the treatment effects across designs; and (4) within-trial error terms. Heterogeneity in the treatment effects was assumed to have the same between-study variance for all treatment contrasts in the network. Inconsistency in the treatment effects was assessed globally by comparing the consistency and inconsistency models using a design-by-treatment interaction approach [43]. For local inconsistency, a difference between the direct and indirect estimates of each treatment effect was examined in all closed loops in the network using a node-splitting method that separated a particular treatment effect from the others, based on the trials that contained the direct, head-to-head comparison, and other trials through a common comparison (the indirect comparison) [44].

Within a network for each outcome, pooled effects presented as point estimates with 95% confidence intervals (CIs) were summarized across all treatments sequentially using league tables. For every outcome, the probability of being the best for a treatment was estimated using surface under the cumulative ranking (SUCRA), which is a numerical summary representing an overall probability of the rank distribution of each treatment [45]. The SUCRA of a treatment represents the averaged proportion of all other treatments performing worse than the treatment of interest; in other words, the probability that the treatment of interest was more efficacious than the rest. If the treatment always ranks first or is 100% to be the best, the SUCRA is equal to 1.0, and if the treatment always ranks as the worst, the SUCRA is 0. The hierarchical performance of every treatment in both beneficial outcomes and tolerability was illustrated by cluster rank diagrams.

To detect possibilities of publication bias due to systematic differences in treatment effects between small and large studies, a funnel plot between effect sizes and the study sample sizes was examined visually. An asymmetric plot indicated evidence of a small-study effect, whereas symmetry in the funnel plot supported the absence of small-study effects.

To assess the robustness of the results, a sensitivity analysis was performed for each study outcome by removing selected studies across a preplanned characteristic concerning quality of the trials in terms of the risk of bias assessment. The impact of excluding certain studies from the network was deemed substantial if the statistical significance of the treatment effects disappeared, or rather appeared, by the operation.

3 Results

3.1 Study Characteristics and Quality Assessment

Through a literature search, 4320 study articles were identified, of which 4244 were excluded after screening their titles and abstracts. Seventy-six articles underwent full-text review, and 49 of these were excluded (Fig. 1). The remaining 27 double-blinded RCTs, published between 1998 and 2016, met eligibility criteria, representing 9774 patients with mild-to-moderate AD in Europe, North America, and Asia. The trial characteristics, including countries, interventions, and patient characteristics are summarized in Table 1. Three trials conducted head-to-head comparisons between EGb761 and donepezil, donepezil and galantamine, or rivastigmine in oral and patch forms. Eighteen trials had at least 100 participants per group. Three trials compared EGb761 with placebo and had a treatment duration of 22–26 weeks. Of the 23 trials involving ChEIs or memantine, nine trials were 12–20 weeks in duration, and 14 trials were 22–26 weeks in duration. The mean age of the participants ranged from 64 to 78 years. In most trials, the proportion of females was higher than that of males. Most trials enrolled participants with mean baseline MMSE scores ranging from 16.0 to 21.6, with standard deviations ranging from 2.3 to 5.1. The majority of the included trials assessed both cognition and clinical global impression outcomes.

Fig. 1
figure 1

Flowchart of study selection. AD Alzheimer’s disease, AEs adverse events, ChEIs cholinesterase inhibitors, EGb761 standardized Ginkgo biloba extract, MCI mild cognitive impairment, RCT randomized controlled trial, VCI vascular cognitive impairment, WHO ICTRP World Health Organization International Clinical Trials Registry Platform

Full size image
Table 1 Description of included studies and patient characteristics
Full size table

The risks of biases of individual studies are listed in Electronic Supplementary Material, Table S1 and summarized in Fig. 2. The methodological quality was sufficient since there were no studies with high risk of bias regarding the randomization and blinding process, only one study with modified intention-to-treat samples, and four studies with dropout rates higher than 30%. The randomization procedure and allocation concealment were not reported by two and four studies, respectively. Random sequence generation and allocation concealment were clearly described with adequate quality in 21 trials. Twenty-five trials provided the number of patient dropouts and reasons. The dropout rates ranged from 3.2 to 17.6% for the EGb761 treatment arms (from two of four trials) and from 1.7 to 35.5% for the ChEIs and memantine arms. Seven trials had dropout rates above 20%. Twenty-six trials reported outcomes based on intention-to-treat approaches. Application of the Cochrane risk of bias assessment criteria found 67%, 11%, and 22% of the selected trials to have low, high, and unclear risk of bias, respectively. All EGb761 trials were judged to have a low risk of bias. Of the ChEI trials and memantine trials, 61%, 13%, and 26% were judged to have a low, high, and unclear risk of bias, respectively. All trials were funded by pharmaceutical companies.

Fig. 2
figure 2

Summary of risk of bias assessment: percentages across all included studies

Full size image

Outcome measures at baseline are presented in Table 2. Three trials reported baseline SKT scores ranging from 15.7 to 17.4, which were equivalent to ADAS-cog scores of 20.4–22.6. The mean ADAS-cog scores at baseline in most trials ranged from 18.7 to 34.9, reflecting mild-to-moderate cognitive impairment. Eighteen trials reported mild impairment in baseline activities of daily living. Nine trials reported baseline behavioral problems with mean NPI scores ranging from 8.7 to 21.5, reflecting mild impairment. Treatment discontinuation rates ranged from 1.5 to 28.6% (median, 7.4%; interquartile range, 9.74%).

Table 2 Study endpoints measured at baseline and discontinuation rate of included studies
Full size table

Visual inspection of the comparison-adjusted funnel plots for cognition, behavior, and clinical global impression outcomes showed no evidence of small-study effects or publication bias (data available on request). For function and tolerability outcomes, the plots were slightly asymmetric. Networks of treatments compared for cognition are shown in Fig. 3, and other major outcomes are shown in Electronic Supplementary Material, Fig. S1. Comparisons of specific adverse event rates are shown in Electronic Supplementary Material, Fig. S2.

Fig. 3
figure 3

Network plot for cognition. DON donepezil, EGb761 standardized Ginkgo biloba extract, GAL galantamine, MEM memantine, PBO placebo, pts patients, RCT randomized controlled trial, RIV oral rivastigmine, R-Patch rivastigmine patch

Full size image

Use of the networks allowed comparison across all drugs and EGb761. Both global and local tests for inconsistency in treatment effects showed no significant differences between the direct and indirect evidence for all outcomes. The assessment of inconsistency is presented in detail in Electronic Supplementary Material, Tables S2–S7. In addition, findings from the direct, pairwise MA were aligned with those from the NMA for all study outcomes (data available on request).

3.2 Cognition

The NMA was based on 27 trials with cognitive outcomes in 9774 patients and showed that all ChEIs were significantly better than placebo in improving cognition (SMD from − 0.52 to − 0.26) (Table 3). The benefits were generally modest; the largest difference in mean changes from baseline was between galantamine and the placebo (SMD [95% CI]: − 0.52 [− 0.62 to − 0.41]). EGb761 and memantine showed no benefits in cognition compared to placebo. Galantamine was superior to rivastigmine in both oral and patch forms, EGb761, and memantine (SMD [95% CI]: − 0.22 [− 0.40 to − 0.05]; − 0.26 [− 0.45 to − 0.07]; − 0.34 [− 0.56 to − 0.12]; and − 0.42 [− 0.71 to − 0.13], respectively), but not better than donepezil (SMD [95% CI]: − 0.09 [− 0.22 to 0.05]). Donepezil was significantly more effective than EGb761 and memantine in bolstering cognition (SMD [95% CI]: − 0.26 [− 0.46 to − 0.05]; − 0.34 [− 0.62 to − 0.05], respectively). By cumulative rankings, galantamine had the highest probability (98%) of being ranked higher than the other agents, whereas the probability of EGb761 being more effective than the others was 35%. Memantine was the least effective drug for cognition (22% chance of being more effective than the others) (Table 4).

Table 3 League table comparing treatments for the major outcomes
Full size table
Table 4 Surface under the cumulative ranking for treatments of the major outcomes
Full size table

3.3 Functional Outcomes

Oral rivastigmine, the rivastigmine patch, and galantamine were significantly better than placebo in delaying functional impairment based on NMA of 19 trials with 7445 patients. Functional outcomes were measured mostly by the ADCS-ADL, PDS, or DAD, and NMA showed benefits with small effect sizes (SMD [95% CI] 0.24 [0.16–0.33] for oral rivastigmine; 0.23 [0.12–0.33] for the rivastigmine patch; 0.21 [0.12–0.31] for galantamine) (Table 3). In comparisons between active agents, oral rivastigmine was significantly more effective than memantine (SMD [95% CI] 0.31 [0.09–0.52]) and donepezil (SMD [95% CI] 0.14 [0.01–0.28]). The rivastigmine patch and galantamine were significantly more effective than memantine (SMD [95% CI] 0.29 [0.07–0.52]; 0.28 [0.06–0.50], respectively). Based on the probabilities of being more effective than the others, oral rivastigmine was the most effective drug (86%), followed by the rivastigmine patch (79%) and galantamine (75%). The probabilities of being more effective than the others for EGb761 and donepezil were 51% and 39%, respectively. The least effective drug on functional outcomes was memantine (7%) (Table 4).

3.4 Behavioral Symptoms

Galantamine was the only agent significantly better in improving behavioral symptoms than placebo (SMD [95% CI] − 0.15 [− 0.26 to − 0.04]), and it was also superior to memantine (SMD [95% CI] − 0.25 [− 0.48 to − 0.03]), based on NMA of ten trials with 4000 patients (Table 3). For cumulative rankings, galantamine had an 80% probability of being more effective than other agents (Table 4); however, the NMA effect sizes were small.

3.5 Clinical Global Impression

The NMA for clinical global impression included 21 trials of 8865 patients, mostly using the CIBIC-plus, ADCS-CGIC, and CGIC. Compared with placebo, donepezil (RR [95% CI] 1.69 [1.42–2.01]), oral rivastigmine (RR [95% CI] 1.48 [1.24–1.77]), the rivastigmine patch (RR [95% CI] 1.37 [1.10–1.70]), and galantamine (RR [95% CI] 1.20 [1.01–1.44]) had significantly greater proportions of patients with improved or unchanged global impressions from baseline (Table 3). Donepezil was significantly better in improving clinical global impression than galantamine (RR [95% CI] 1.40 [1.09–1.80]) and EGb761 (RR [95% CI] 1.40 [1.06–1.85]), whereas memantine and EGb761 had no significant benefit compared to placebo. The probability of having a higher cumulative ranking than the others was greatest for donepezil (96%), followed by oral rivastigmine (78%), the rivastigmine patch (62%), and galantamine and EGb761 (40% each). Memantine was the least effective drug for clinical global improvement, with a probability of being more effective than the others of 28% (Table 4).

3.6 Tolerability

Twenty-four studies with a combined total of 9634 patients provided data on treatment discontinuation due to adverse events. The NMA revealed that ChEIs had a significantly higher risk of discontinuation than placebo (RR [95% CI] 2.14 [1.49–3.06] for oral rivastigmine; 2.04 [1.30–3.20] for rivastigmine patch; 1.79 [1.28–2.49] for galantamine; and 1.49 [1.03–2.17] for donepezil) (Table 3). However, risk of discontinuation of EGb761 did not significantly differ from placebo, ChEIs, and memantine. Oral rivastigmine had the highest probability of being less tolerable than the others (77%), whereas donepezil tended to be tolerated better than the others, with the lowest probability of being more at risk of discontinuation (38%). The cumulative probability of being the worst option regarding discontinuation risk was 70% for the rivastigmine patch, 63% for memantine, 57% for galantamine, and 42% for EGb761 (Table 4).

3.7 Adverse Events

Sixteen trials provided data on at least one selected adverse event. The NMA results showed that oral rivastigmine had a significantly higher risk than placebo for nausea, vomiting, dizziness, weight loss, and anorexia. Donepezil and rivastigmine oral and patch had a significantly higher risk for diarrhea than placebo. Galantamine and oral rivastigmine had a significantly higher risk for headache than placebo (Electronic Supplementary Material, Table S8). In comparison to the other active agents, oral rivastigmine had the highest probabilities for the risks of nausea (100%), vomiting (100%), dizziness (92%), weight loss (86%), and anorexia (100%) (Electronic Supplementary Material, Table S9).

3.8 Cluster Rank

Taking both treatment efficacy and tolerability into a single perspective for all agents and placebo, Fig. 4 depicts, for every outcome on the horizontal axis, the probability of an agent being more effective than the others and, on the vertical axis, the probability of having a higher risk of treatment discontinuation due to adverse events. An agent located at the right-most and lowest corner is deemed the most effective and safest treatment for patients with mild-to-moderate AD. Overall, except for placebo, donepezil was the best tolerated drug, and rivastigmine, in both oral and patch forms, was the least tolerated drug, as indicated by the two extreme probabilities of having the highest proportions of patients who discontinued treatments. Donepezil was the most effective agent on the clinical global impression, as indicated by the highest probability of being more effective than the others. Donepezil followed galantamine and led rivastigmine in oral form and patch for the probability of being more effective on cognition. Donepezil followed EGb761 and galantamine for being more effective on the behavioral outcomes. For function outcomes, donepezil had a lower probability than all other ChEIs and EGb761. On the other hand, memantine was the active agent that performed the worst on all efficacy outcomes, and its probability of having a higher risk of discontinuation was close to that of rivastigmine.

Fig. 4
figure 4

Clustered ranking plots for efficacy and tolerability: a cognition versus tolerability; b functional outcome versus tolerability; c behavioral symptoms versus tolerability; and d clinical global impression versus tolerability. DON donepezil, EGb761 standardized Ginkgo biloba extract, GAL galantamine, MEM memantine, PBO placebo, RIV oral rivastigmine, R-Patch rivastigmine patch, SUCRA surface under the cumulative ranking

Full size image

3.9 Sensitivity Analysis

When trials with high or unclear risk of bias were excluded, most of the sensitivity analysis results were consistent with results from the base-case analysis. However, some of the effect estimates that were statistically significant initially became non-significant. Changes in the significance level occurred in three of 21 comparisons: between oral rivastigmine and EGb761 on cognition, oral rivastigmine and donepezil on function, and the rivastigmine patch and placebo on clinical global impression (data available on request). Overall, the sensitivity analysis did not have a major impact on the main findings from the base-case analysis with respect to the relative efficacy on a given outcome.

4 Discussion

The present NMA summarizes data from double-blinded RCTs on the efficacy and tolerability of all active agents for mild-to-moderate AD. All three ChEIs contributed a statistically significant improvement in cognition compared to placebo despite a modest magnitude, and galantamine performed the best. A better and modest effect on functional outcomes was found for rivastigmine in both oral and patch forms and galantamine, whereas a better effect on behavioral symptoms was found for galantamine only. For the clinical global impression, all ChEIs had significantly better effectiveness than placebo and, among them, donepezil performed the best and better than galantamine. EGb761 and memantine were not significantly better than placebo for all beneficial outcomes. For tolerability, rivastigmine, regardless of the formulation, posed the highest risk of treatment discontinuation due to any adverse events compared with other agents. Oral rivastigmine was the worst with respect to risks of specific adverse events, including nausea, vomiting, dizziness, weight loss, and anorexia, which was in accordance with findings from previous NMAs [20, 21]. Our findings did not support that EGb761 was better than ChEIs for treatment discontinuation.

A previous NMA by Kobayashi et al. [20] reported that the most effective ChEI for cognition was galantamine, and for clinical global impression, the most effective agent was donepezil; however, no ChEIs were better for the behavioral outcomes than placebo. A subsequent NMA of ChEIs and memantine by Tricco et al. [21] reported that the most effective drug for cognition was donepezil and galantamine was the most effective drug for behavior, whereas all ChEIs were no better than placebo for function. For clinical global impression, Tricco et al. [21] reported that galantamine was the most effective drug based only on the CIBIC-plus on an interval scale. The previous two NMAs did not examine EGb761 and used broader inclusion criteria that included open-label trials [20] and observational studies [21]. The findings in the present study confirmed that galantamine for cognition and donepezil for clinical global impression were the most effective ChEIs, whereas galantamine was the only ChEI that had a better benefit than placebo for behavior, although these effects were of modest magnitude. In addition, oral rivastigmine was consistently the least tolerable drug.

Our NMA is the first of its kind to compare EGb761 with ChEIs and memantine. In previous direct MAs, compared with placebo, EGb761 was reported to be more effective for cognition, function, behavior, and clinical global impression with a very high inconsistency (I2 93–96%) [11,12,13, 70,71,72,73]. The reported therapeutic benefits of EGb761 were modest and likely confounded by an inclusion of vascular dementia and AD with cerebrovascular diseases [11,12,13, 70,71,72,73]. The present study focused on patients with mild-to-moderate AD without any other types of dementia or neurological disorders and found no better therapeutic benefit and tolerability of EGb761 versus placebo, all ChEIs, and memantine.

The use of ChEIs in patients with AD has been based on a hypothesis of memory impairment, which states that cholinergic deficits are responsible for the cognitive and behavioral changes [74, 75]. In addition to being a selective reversible and competitive inhibitor of acetylcholinesterase (AChE), galantamine acts as an allosteric modulator of neuronal nicotinic acetylcholine receptors (nAChRs), which have been shown to be reduced in AD, and acts synergistically with acetylcholine to increase nAChR activation [76]. However, the mechanism of galantamine regarding the behavior outcomes remains unknown. On the other hand, memantine has been used mostly in patients with moderate-to-severe AD. Memantine is believed to prevent the neuronal dysfunction that is due to the toxicity of the elevated glutamate [77, 78].

Strengths of the study were the criteria used to select trials and factors that could lead to inconsistencies in the effect estimates. We excluded seven trials that allowed for concomitant drug use, which could contaminate the intended effects of the studied drugs. Based on the 27 included trials, 23 trials excluded patients who received ChEIs or EGb761 before enrollment, and the remaining four trials included patients who discontinued ChEIs for at least 4 weeks. These exclusion criteria help to reduce the influence of previous treatments. In addition, all included trials had an adequate duration and dosage of treatment. The restrictive study selection criteria supported the transitivity assumption of the analysis.

An important limitation was attrition bias, which was relatively high among the included trials. The majority of RCTs used the last-observation-carried-forward strategy for the intention-to-treat analysis when data were missing. Our results for efficacy and tolerability were based on the studies that reported short-term outcomes and adverse events. However, reports of long-term outcomes of ChEIs or EGb761 were limited and usually from the extended open-label phase of the RCTs. There was an interest in the benefit of combining ChEIs with EGb761; however, limited reports were available. Cognition and functional outcomes were not significantly different between the combination and monotherapy [17, 79]; therefore, the present study did not include the combination.

The aims of AD treatment are to promote independence, maintain function, and treat symptoms. However, the evidence revealed from this NMA showed that ChEIs provided minimal benefit. Most guidelines recommend donepezil, galantamine, and rivastigmine for mild-to-moderate AD but do not specify the preferred option [4,5,6]. In low- to middle-income countries such as Thailand, if the generic products of donepezil and oral rivastigmine are available, donepezil should be first-line treatment if not contraindicated because it showed better clinical global impression and tolerability than other ChEIs and EGb761, with a similar benefit for cognition as galantamine, a high-cost medicine. In AD patients who have behavioral neuropsychiatric symptoms, galantamine, which is the only ChEI with behavioral benefits, is preferred.

5 Conclusion

In patients with mild-to-moderate AD, EGb761 and memantine had no therapeutic benefits compared with placebo, whereas galantamine provided higher benefits for cognition and behavioral symptoms than other ChEIs. Donepezil, an expired patent drug, should be used as a first-line option because it provides a better clinical global impression and tolerability with a similar benefit for cognition as galantamine.