Background
Alzheimer’s disease (AD), a progressive neurodegenerative disorder, is the most common form of dementia affecting 46.8 million people with an enormous public health impact [
1,
2]. Primary manifestations of AD include progressive deterioration of cognition, impairment in functional ability, and alterations of neuropsychiatric symptoms. Currently, there are no therapeutic interventions that can delay the disease progression, but available medications have provided symptomatic benefits [
3]. Two main classes of drugs are recommended by the US Food and Drug Administration (FDA) for pharmacological management of AD: cholinesterase inhibitors (ChEIs) donepezil, galantamine, and rivastigmine, which are licensed for mild to moderate AD [
4,
5]; and glutamate antagonist memantine for moderate to severe stage [
6]. Donepezil is the only ChEI to be indicated for use all across the full spectrum of AD [
7]. Owing to fewer treatment options for more serious patients, donepezil 23 mg was approved for moderate to severe AD in recent years [
8,
9] and the combination therapy of donepezil plus memantine is proposed for treating patients in this stage [
10].
Previous meta-analyses focused only on comparing the efficacy and tolerability of drugs with placebo, which gave the clinical application directions [
6,
11,
12]. Those findings showed modest benefits for improving the symptoms related to cognition, function, behavior, and clinical global changes [
13,
14]. But considering the lack of direct comparative evidence among available drugs, the results of those studies were inconclusive on how to choose the optimal therapeutic regimens to achieve the maximum efficiency [
15]. The network meta-analysis combines evidence from a network of all included trials to rank all available treatments in terms of efficacy and tolerability, providing estimates for interventions even if they have not been directly compared [
16]. Here, we therefore conducted a network meta-analysis to comprehensively compare and rank different types and dosages of cognitive enhancers at different clinical stages for guiding treatment decisions.
Methods
Search strategy and selection criteria
In this network meta-analysis, potentially eligible randomized controlled trials (RCTs) were searched through PubMed, Embase, and the Cochrane Central Register of Controlled Trials, which were published between database inception and July 21, 2017. Additional trials were retrieved from the cited references of relevant published meta-analyses and systematic reviews. Included studies were completed RCTs with English language publication that met the following criteria: only double-blind RCTs with follow-up of 12–104 weeks; the trials compared four primary FDA-approved treatments (donepezil, galantamine, rivastigmine, or memantine) alone or in combination (only including the donepezil–memantine combination approved by the FDA) with placebo or other treatments; drug dosages were specific and within the therapeutic range; eligible participants had a clinical diagnosis based on the
Diagnostic and Statistical Manual of Mental Disorders (DSM) for dementia of the Alzheimer’s type or the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association (NINCDS-ADRDA) for probable AD [
17]; criteria for disease severity classification were reasonable; and at least one of the five outcomes of cognition, function, behavior, global assessment, or adverse events was covered. We excluded quasi-randomized trials, trials with too short-term or too long-term follow-up (< 12 weeks or > 104 weeks), trials that included patients with mixed dementia or neuropsychiatric symptoms, and trials that recruited fewer than 10 participants per group.
Data extraction and quality assessment
Basic study characteristics—for instance, sample size, age, gender, race, drug dosage, disease severity, diagnostic criteria, trial duration, cognitive scores of baseline level, efficacy outcomes of the change from baseline, and individuals who experienced all-cause adverse events—were extracted from each trial. For all of the studies included, we analyzed the intention-to-treat (ITT) population results if the trials adopted the ITT approach [
18]. Three investigators (K-XD, M-ST, and C-CT) independently abstracted information from original articles with the standardized data extraction table. If no consensus was reached, further discussion would be carried out with other members of the team or authors. We appraised the risk of bias using the Cochrane Risk of Bias Tool [
19].
Outcomes
Measurement scales that were used in the trials were different from each other. We paid close attention to cognitive function (the mean overall changes from baseline to endpoint) and tolerability (the responders who had any adverse events during the treatment period) for the primary outcomes. Efficacy in cognition was mainly evaluated by the Alzheimer’s Disease Assessment Scale—cognition subscale (ADAS-cog), the Severe Impairment Battery (SIB), and the Mini-Mental State Examination (MMSE). Secondary outcomes included daily functions assessed by the Alzheimer’s Disease Cooperative Study—Activities of Daily Living (ADCS-ADL) and the Bristol Activities of Daily Living Scale (BADLS), neuropsychiatric symptoms assessed by the Neuropsychiatric Inventory (NPI), and the global assessment of changes assessed by the Clinician’s Interview Based Impression of Change Plus Caregiver Input (CIBIC plus) and the Clinical Global Impression of Change (CGIC). Based on all included RCTs in this network meta-analysis, we summarized that most of the patients with mild to moderate AD have MMSE scores of 10–26 and those with moderate-to-severe AD have scores of 0–15.
Statistical analysis
At first, we conducted a pairwise meta-analysis using the random-effects model. We chose the standardized mean difference (SMD) with 95% confidence interval (CI) as the effect sizes for continuous results, while dichotomous results were calculated using pooled odds ratios (ORs). We quantitatively investigated the statistical heterogeneity in each direct comparison using the I2 statistic and P value. Publication bias was examined with the Egger’s regression test. The pairwise meta-analysis was conducted in STATA (version 14.1) software.
Secondly, we performed a random-effects model Bayesian network meta-analysis using WinBUGS software (version 1.4.3; MRC Biostatistics Unit, Cambridge, UK) and drew relevant diagrams with STATA [
20]. We summarized the results of the network meta-analysis by choosing the SMD or OR with corresponding credible intervals (CrIs) as effect sizes. We adopted noninformative priors and changed the precision of the prior distribution in sensitivity analyses. Details about the WinBUGS codes are presented in Additional file
1: Supplementary 1. The global heterogeneity of network meta-analyses was assessed by the
I2 statistic with the gemtc R package (version 3.2.2). Inconsistency was statistically examined by calculating the significant discrepancies between direct and indirect evidence in each closed loop with the loop-specific method and the node splitting method [
21‐
23]. The intervention hierarchy was estimated and expressed by rankograms, the surface under the cumulative ranking curve (SUCRA), and mean ranks [
24]. The comparison-adjusted funnel plot could be drawn to detect publication bias in the network meta-analysis [
24].
Study characteristics, the main source of heterogeneity, might affect the accuracy of the results. Correspondingly, we conducted the subgroup network meta-analyses for secondary outcomes to adjust for the difference in severity of disease. Additionally, sensitivity network meta-analyses were conducted to examine the robustness of primary outcomes by omitting short-term or longer follow-up studies (only including studies with 20–30 weeks of follow-up).
Discussion
This network meta-analysis comprehensively compares and ranks efficacy and tolerability among current available cognitive enhancers approved by the FDA for AD. The cholinesterase inhibitors result in better outcomes on cognition than memantine for mild to moderate patients, among which galantamine and donepezil are probably the interventions most strongly associated with cognitive improvements. For moderate to severe AD, the combination therapy of donepezil 10 mg with memantine 20 mg is the most effective regimen, followed by donepezil 23 mg alone. However, none of the cognitive enhancers was likely to improve behavior. The higher-dose rivastigmine transdermal patch (15 cm2) was probably the best option considering the benefits of both function and clinical global impression. Memantine shows the best profile of acceptability, while rivastigmine oral form is associated with a high incidence of adverse events. Both clinical efficacy and adverse events related to cognitive enhancers are shown to be dose dependent. Our findings may help physicians choose targeted pharmaceuticals for patients with different stages and clinical symptoms.
In this network meta-analysis, we extend the previous meta-analysis by including a large number of studies with a broader dosage range and enrolling more patients in all stages of disease. Two dosages of galantamine 24 mg and 32 mg daily were involved for mild to moderate AD. The corresponding SMD (efficacy) of 0.01 and OR (tolerability) of 0.64 indicated no additional risk–benefit profile from higher doses. Hence, this evidence suggested that galantamine 24 mg daily probably had a favorable balance between benefits and tolerability. For patients with moderate to severe AD, it was observed that the combination therapy of donepezil 10 mg with memantine 20 mg offered the highest level of cognitive benefits, and the results generally supported previous studies and current guidelines [
10,
25‐
27]. Moreover, we evaluated the effects and safety of donepezil 23 mg approved recently. This treatment showed beneficial effects on cognition, but had the lowest safety profile. The Asia-Pacific Expert Panel for donepezil 23 mg recommended that we should perform a stepwise escalation of donepezil to 23 mg daily and monitor the incidence of adverse events in a timely manner [
28].
All pharmacological interventions are well tolerated for most patients. Compared with placebo, more side effects occurred with ChEIs and combined treatment, but not with memantine. Although ChEIs share a similar mode of action, they differ in pharmacologic characteristics and routes of administration, which can determine the pharmacological and toxicity profile [
29]. Rivastigmine capsules 12 mg daily were strongly associated with high incidence of all-cause adverse events among ChEIs. The transdermal patch formulation approved for use across all stages of AD has been shown to have a better tolerability in comparison to the oral form. A higher dose of the rivastigmine transdermal patch (15 cm
2) is also approved for the treatment of moderate to severe AD [
30]. Better outcomes were observed as improvements in daily activities and global clinical impression in our analysis. The transdermal system delivery is innovative because it reaches the point of steady-state plasma concentration in a shorter time and alleviates the adverse reactions of the gastrointestinal tract better than the oral form based on the principle of pharmacokinetics [
31]. Overall, the transdermal patch may provide a new approach to AD therapy.
Neuropsychiatric symptoms are common features throughout all stages of AD, which lead to a heavy burden for patients and caregivers [
32]. Previous meta-analyses have already concluded that the ChEIs and memantine have benefits in the treatment with neuropsychiatric symptoms, and various drugs show mixed results [
12,
33]. The results of our pairwise meta-analyses showed that only galantamine’s beneficial effects were verified compared with placebo, but no significant difference for cognitive enhancers was found through the network meta-analysis. Furthermore, there was no statistical difference between five active treatments and placebo in our subgroup meta-analysis. Fewer patients with neuropsychiatric symptoms based on our inclusion criteria may provide a reasonable explanation for the discrepant result. Many systematic reviews and meta-analyses recommended that antipsychotics, ginkgo biloba, and nonpharmacological treatment provide evidence of effects for patients with behavioral symptoms [
12,
34‐
36]. Developing new drugs targeting neuropsychiatric symptoms and further research to strengthen evidence of therapeutic benefits are needed.
This network meta-analysis has several limitations. We found that the main inconsistency in our network analysis was in the loop of donepezil 5 mg–placebo–donepezil 10 mg, and we considered that this inconsistency was related to the difference in durations of the included trials. We conducted sensitivity network meta-analyses restricting the study period to 20–30 weeks, which did not result in any further inconsistency. The global heterogeneity of network meta-analyses was high in cognition among the population with moderate to severe AD, probably because different measurements (MMSE and SIB scale) were included in this network. In the pairwise meta-analysis, the heterogeneity was high for donepezil 10 mg vs placebo, and we found that two trials had a relatively shorter follow-up (12 weeks), which may have led to increased heterogeneity. The research implications of these results were limited by study characteristics (such as trial durations, simple sizes, and inclusion or exclusion criteria) and methodological models, and they also suffered from the bias due to quite fewer studies in a pair of comparisons or selective reporting, which might result in potential confounding factors to be cautious of. Although we used multiple databases to search published articles as much as we can and sent emails to authors for additional detailed information, we are still unable to exclude some possibilities that several unpublished studies are unavailable or that included eligible trials might overvalue the research outcomes. Furthermore, this network meta-analysis offers a clearly hierarchic relationship, but intervention rankings have a certain degree of imprecision, because most interventions have overlapping 95% CrIs [
37].
In this network meta-analysis, we did not involve a cost-effectiveness analysis. Medication cost accounts for a large proportion of the total dementia healthcare cost. It is indicated by most economic evaluations that pharmacological treatments for AD are reasonable in terms of clinical effects and costs. The probabilistic sensitivity analyses suggested that donepezil and memantine were cost-effective with slightly greater quality-adjusted life years (QALYs) [
38,
39]. Treatment with donepezil plus memantine was cost-effective within the willingness-to-pay threshold in moderate-to-severe AD [
40,
41]. Nevertheless, the results generally are associated with a degree of uncertainty, which is related to country-specific data. In England, the use of prescription drugs doubled after introducing the national dementia strategies [
42]. The proportion of people who receive the concomitant use of AChEI and memantine has increased in Europe [
43]. Indeed, drug costs are very high in some countries (China, Indonesia, South Africa) because some drugs remain on patent and these countries are reluctant to use generic medicines.
Conclusion
Our analysis provides some evidence that galantamine and donepezil may show the highest level of efficacy in cognition for mild to moderate AD, and the combination therapy (memantine 20 mg with donepezil 10 mg) and donepezil 23 mg daily are recommended for moderate to severe patients. The higher-dose rivastigmine transdermal patch (15 cm2) is the best option considering the benefits of both function and clinical global impression. None of the medicines is likely to improve behavior through current network meta-analysis. Memantine shows the best profile of acceptability, while rivastigmine oral form is associated with a high incidence of adverse events. Although the clinical effects are uncertain in the multifactor environment, these findings are helpful for guiding treatment decisions. Hopefully, further research should try to differentiate more clearly the effects of monotherapy versus combined therapies.