The effectiveness and cost-effectiveness of donepezil, galantamine, rivastigmine and memantine for the treatment of Alzheimer's disease (review of TA111): a systematic review and economic model

Definitions

Dementia is usually a disease of later life and has been defined as:

A syndrome consisting of progressive impairment in memory and at least one other cognitive deficit (aphasia, apraxia, agnosia, or disturbance in executive function) in the absence of another explanatory central nervous system disorder, depression, or delirium.

Diagnostic and Statistical Manual of Mental Disorders-Fourth Edition (DSM-IV)3

People with dementia may also show other symptoms, such as depression, psychosis, wandering and aggression.

Alzheimer’s disease is the most common form of dementia, and is additionally characterised by the presence of neurofibrillary tangles and amyloid plaques in the cerebral cortex, observed at post-mortem.

Diagnosis

In the distant past, diagnosis of AD before death had been on the basis of excluding other causes. However, there are now agreed criteria that accurately predict up to 90% of AD cases (Box 1). Alternatively, diagnosis can be made from ICD-10 (International Classification of Diseases, 10th Revision4) and DSM-IV(Diagnostic and Statistical Manual of Mental Disorders-Fourth Edition3).

The diagnosis of dementia may happen many months after onset, as the development of symptoms is usually insidious. It may take some time for the individual to realise that significant memory, mood or ability changes are taking place. Other possible diagnoses, such as depression, delirium, vitamin B₁₂ deficiency and hypothyroidism, have to be excluded first. Further testing is necessary to determine the particular cause of dementia. 5

Examination by the National Audit Office (NAO) of data from primary care trusts (PCTs), the Dementia UK report8 and the Office for National Statistics (ONS) has indicated that, in England, more than 50% of people with dementia never receive a correct diagnosis. 9 See Figure 1 for 2006 estimates of the diagnosis gap.

FIGURE 1.

The gap between prevalence and diagnosis of dementia in England. Source: Knapp et al. 8 Dementia UK report to the Alzheimer’s Society, King’s College London and London School of Economics and Political Science (estimated actual average prevalence) and General Practice Research Database report to the NAO (reported prevalence based on diagnoses). The graph shows reported prevalence of dementia, based on levels of diagnosis within PCTs, for ages 65 years and upwards in 2006. The estimated actual average prevalence has been calculated using data from the 2007 Dementia UK report8 in conjunction with population estimates from the ONS. The latter does not take into consideration those aged 85 years and above, owing to restrictions on the data available, and as such forms a very prudent estimate of dementia in the > 65-year population.

Significance for patients

It can take several years of slow deterioration for the full effects of AD to be felt. 21 In the early stages there can be severe memory loss for recent events with associated repetitive questioning and loss of the ability to learn. 21,22 There may be a general deterioration in the ability to socialise, which can be difficult for both sufferer and carer to cope with. 23,24 As mild AD takes hold, normal activities of daily living, such as shopping or managing finances, become increasingly difficult as cognitive function deteriorates. 25 Communication also becomes a problem as vocabulary shrinks and fluency falters. 25,26 At this stage the sufferer may still be aware of their failing abilities, and the experience and known outcome of AD can frequently lead to associated depression.

Disease progression to moderate AD leads to further loss of cognitive abilities, including the ability to remember and/or understand words. Activities of daily living become increasingly affected as the ability to perform purposeful movements decreases, for example getting dressed or cooking. Commonly there are also neuropsychiatric symptoms such as anxiety, wandering, irritability, disinhibition and apathy. Visual and auditory hallucinations occur in about 30–59% of sufferers. 7 Managing these symptoms can be a very difficult burden for carers, who may well be elderly themselves. Indeed, the main predictors of full-time institutional care are caregiver exhaustion,27 the degree of patient dependence28 and the rate of disease progression. 29

In developed countries, sufferers of AD usually end their days in institutional care, as the last stages of AD bring complete dependence. This final stage is characterised by limitations such as inability to walk, manage personal care, mutism, inability to recognise familiar people and objects, and incontinence. There may also be seizures and involuntary twitching.

Significance for carers

Being the main carer for a person with AD can have an enormous impact on physical, psychological and social well-being. 30 From the early frustrations, prior to diagnosis, of living with others’ impaired cognitive function, through the devastating diagnosis, to the knowledge that the relation/friend is going to get progressively worse and die, the outlook for carers is bleak. Many carers are elderly spouses, perhaps with health concerns of their own or grown-up children who now have their own families to care for as well. 31 Carers may cope reasonably well with the early stages of the disease, but as the behavioural and psychological symptoms of dementia (BPSD) become more severe, full-time institutional care becomes increasingly likely. 5 For some carers this brings feelings of guilt and depression,32,33 possibly leading to the cognitive decline of the carers themselves. 34 Behavioural and psychological symptoms are common in AD and may be difficult to manage, causing distress to carers and patients alike. They have been shown to be better predictors of institutionalisation35 and carer distress36 than cognitive symptoms.

As AD progresses, increasing grief and feelings of loss may be experienced by carers. 37 Findings from the EUROCARE European study of co-resident spouse carers of dementia sufferers showed that co-resident carers carried a heavy burden and that mental distress was high. They concluded that issues of behavioural disturbance, negative social reactions, financial worries and younger spouse carers predicted greater distress. 38 However, there is evidence that enhanced counselling and support can relieve symptoms of depression in caregivers and delay admission to institutional care of people with AD. 39

Significance for the NHS and social services

With an increasingly elderly population, the burden of AD upon the NHS and social services is considerable. Of the estimated 520,000 people in the UK with AD,10 It is estimated that approximately 63.5% live at home and 36.5% are in residential care. 8 Unsurprisingly, the risk of moving into residential care increases with age and disease severity. The proportion of people with severe dementia increases from 6.3% among those between 65 and 69 years old to 23.5% among those aged 95 years or older. 8 Consequently, the proportion of people in the UK with dementia who live in residential care rises from 26.6% of those aged 65–74 years to 27.8% of those aged 75–84 years, 40.9% of those aged 85–89 years and 60.8% of those aged 90 years or older. 8

As the disease progresses, the balance of burden of care shifts from predominantly falling on the informal carers to the NHS and social services as patients are sustained with medication and support at home, until finally financial costs fall mostly on social services as patients move into institutional care, although a proportion of this cost may be borne by the carer or their family. Another proportion of people will qualify for the NHS Continuing Care programme, which will meet their care needs either in the community or in an institution. A longitudinal cohort study by Banerjee and colleagues40 has found that when a person with dementia lives with their main carer they are 20 times less likely, over the course of a year, to move into residential care than those who do not [odds ratio 0.05 (95% CI 0.01 to 0.42)]. They also found that the carer’s psychological quality of life (QoL) and the severity of behavioural problems shown by the patient were predictors of institutionalisation [odds ratios 1.10 (95% CI 1.02 to 1.19) and 1.08 (95% CI 1.01 to 1.15), respectively]. 40 However, in a similar study, de Vugt and colleagues41 found that it was the carer’s response to the behavioural symptoms (rather than the symptoms themselves) that predicted institutionalisation.

Measurement of disease

Details of individual measures used in the included trials can be found in Appendix 1.

A review of outcome measures used in clinical trials of drugs for AD by Wolfson and colleagues42 revealed a number of shortcomings in these measures. In particular, they found that several of the scales had weak psychometric properties, for example lack of responsiveness to change. Some studies had small sample sizes and others used inappropriate statistical analyses. 43

The progress and symptoms of AD can be measured through cognitive tests, behavioural measures, measures of functional ability/QoL and global rating scales.

A thorough assessment of cognitive ability would include measures of attention, processing speed, visuospatial function, praxis, language, executive function and abstraction. The most commonly used scales for this domain are the Mini Mental State Examination44 (MMSE) and the Alzheimer’s Disease Assessment Scale – Cognitive Subscale45 (ADAS-cog). Whereas the MMSE’s validity and reliability as a screening tool for AD have been established,44 it has problems with identifying change over time and scores are affected by people’s level of education. 46,47 Similarly, the ADAS-cog has been criticised for its insensitivity to change in cognitive ability at either end of the severity continuum. 48 It is concerning that the most commonly used instruments to measure change in cognitive function in drug trials for AD should be insensitive to change.

The measurement of behaviour change is important as it is these symptoms that many caregivers find most difficult to cope with, precipitating the transition into institutional care. 35,36 The most frequently used measure of behavioural change in AD trials is the Neuropsychiatric Inventory (NPI). 49 This is a proxy-rated scale, usually completed by the main carer; its validity and reliability have been demonstrated by Cummings and colleagues. 49

There are two kinds of global rating scales for AD: those that measure the severity of illness at a point in time, for example the Clinical Dementia Rating (CDR) Scale50 and can, if used repeatedly, plot mental deterioration over time; the other sort of global instruments are change scales, such as the Clinician’s Interview-based Impression of Change – plus Caregiver Input (CIBIC-plus). 51 These measure broad changes in AD. However, their use may be biased towards cognitive abilities, as Claus and colleagues52 have found that clinicians may have a bias towards this aspect of AD, whereas carers place more emphasis on behavioural and psychological symptoms and functional ability. However, the use of CIBIC-plus may help to overcome this.

Measures of functional status in clinical trials are most commonly taken using the Activities of Daily Living (ADL) Scale53 or the Instrumental Activities of Daily Living (IADL). 54 Their reliability and validity has been described by McDowell and Newell. 55 However, this is not in the specific context of dementia.

Although the DEMQOL has been validated as a measure of health-related quality of life (HRQoL) in people with dementia,56 in clinical trials the most frequently used measure is the patient-rated QoL scale. This is a seven-item patient-rated scale that measures feelings of well-being in the domains of relationships, eating, sleeping and social and leisure activities, on a 0–50 analogue scale. 57

Informal care

Formal care

The formal care of people with AD falls mainly to the NHS and social services, although private and voluntary sector agencies are also involved. The NAO has produced a diagram to show the types of providers that are currently involved in dementia care (Figure 3).

FIGURE 3.

Key public providers involved in the formal care of people with dementia. Source: NAO. 9 a, Royal College of General Practitioners;68 b, Royal College of Psychiatrists. 69 c, Barnes and Lombardo. 70 d, Membership Office. 71 e, Patients receiving domiciliary care may also be in receipt of day care and vice versa. f, Knapp et al. 8 Dementia UK report to the Alzheimer’s Society, King’s College London and London School of Economics and Political Science (older people are aged 65 years and over). The 2005 census data and Dementia UK both report that there are over 8 million older people in England.

In addition to informal care provided by unpaid carers, at any one time people with dementia will have, or need to, access to a range of health and social care services. For example, a person with dementia may receive services from the district nurse or community matron, see their GP, attend day care and be in receipt of domiciliary care. As the disease progresses, the needs of most people with dementia for domiciliary and other care and support will increase. Ultimately, a care home may become appropriate or the person with dementia may stay in their own home supported by various aspects of health and social care.

The cost of care: overview

Around two-thirds of the care for people with AD comes informally from the community and it is the family who bares the greatest burden of cost. 9 The following diagram from the NAO report (Figure 4) shows how the cost of dementia was spread in England in 2005–6. [The categories of mild, moderate and severe dementia are based on the schedule named the Cambridge Mental Disorders of the Elderly Examination (CAMDEX). 72]

FIGURE 4.

Dementia costs in England 2005–6 by severity and place of residence. Source: NAO. 9 Dementia UK report to the Alzheimer’s Society, King’s College London, and London School of Economics and Political Science. Informal care is costed using (1) the hourly cost of a home care worker for specific tasks and (2) the minimum wage for time spent on general tasks and supervisory activities. The estimated service cost of late-onset dementia is over £14B per annum. Most of this is accounted for by supported accommodation (i.e. care home) costs and informal care (i.e. notional) costs. Some £6.87B is borne by people with dementia and their families.

It has been estimated in the recent Dementia 2010 report that these costs have risen to an annual cost to the UK economy of £23B per year (2007–8) (this compares with £12B per year for cancer and £8B per year for heart disease). The majority of this £12.4B cost fell on unpaid carers (55%); social services funded 40% (£9B) of the cost and the NHS funded 5% (£1.2B). 10

Presuming that 62% of people with dementia have AD (see Epidemiology), this translates to an annual cost to the UK economy of AD of over £14B per year. For each person with AD this gives an estimated annual cost of £27,647 – more than cancer (£6000), stroke (£5000) and heart disease patients (£3500) put together. 10 A full description of the costs of care for people with AD can be found in Chapter 7 (see Cost of health and social care received by Alzheimer’s disease patients).

Variation in services

The Dementia UK report8 indicated that there is a wide variation in the prescribing of anti-dementia medication in England and Wales. 8 Information specific to AD was not available, but it may be reasonable to suggest that the picture would be similar. The data were collected by IMS, a medical information consultancy, between October 2005 and September 2006 from 50% of the pharmacies in England and Wales, and represent 90% of all UK prescribing. This information is a reflection of national commissioning practice and shows, by PCT, the likelihood of being prescribed medication for dementia. The number of prescriptions per person with dementia in primary care varied from 12.0 prescriptions per year in Knowsley to 0.4 in West Berkshire. Most PCTs (75%) prescribed between 1.0 and 4.0 prescriptions per year. 8 The reason for this variation in provision is unclear.

Three licensed acetylcholinesterase inhibitors

This technology assessment report (TAR) will consider four pharmaceutical interventions. Three have marketing authorisations in the UK for the treatment of adults with mild-to-moderately severe AD (measured by the MMSE score 26–10). These are donepezil (Aricept^®, manufactured by Eisai Ltd), rivastigmine (Exelon^®, manufactured by Novartis) and galantamine (Reminyl^®, manufactured by Shire Pharma). They are acetylcholinesterase inhibitors (AChEIs), which work by restricting the cholinesterase enzyme from breaking down acetylcholine, thus increasing the concentration and duration of acetylcholine at sites of neurotransmission.

Study design

Internal validity

The instrument sought to assess the following considerations:

1. Was the assignment to the treatment groups really random?

2. Was the treatment allocation concealed?

3. Were the groups similar at baseline in terms of prognostic factors?

4. Were the eligibility criteria specified?

5. Were outcome assessors blinded to the treatment allocation?

6. Was the care provider blinded?

7. Was the patient blinded?

8. Were point estimates and a measure of variability presented for the primary outcome measure?

9. Did the analyses include an intention-to-treat (ITT) analysis?

10. Were withdrawals and dropouts completely described?

In addition, methodological notes were made for each included study, including the reviewer’s observations on sample size and power calculations, participant attrition, methods of data analysis, and conflicts of interest.

External validity

External validity was judged according to the ability of a reader to consider the applicability of findings to a patient group and service setting. Study findings can only be generalisable if they describe a cohort that is representative of the affected population at large. Studies that appeared representative of the UK AD population with regard to these considerations were judged to be externally valid.

Pair-wise meta-analysis

We used random-effects meta-analyses (DerSimonian and Laird model80) only, regardless of any statistical evidence of interstudy homogeneity. Heterogeneity was explored by visualisation of results and, in statistical terms, both Cochran’s Q-test (compared with a chi-squared distribution)81 and the I²-statistic. 82,83 Small-study effects (including publication bias) were visualised using funnel plots and quantified using Egger’s test84 (see Appendix 4). Analyses were conducted using bespoke software, written in Microsoft Visual basic for Applications (version 6) and applied in both Microsoft Access 2003 and Microsoft Excel 2003 (Microsoft Corporation, Redmond, WA, USA). Stata 10.1 (StataCorp LP, College Station, TX, USA) was used to generate forest plots (‘metan’ command) and to assess small-study effects (‘metabias’ command).

Where more than one arm of a contributing trial was relevant to any analysis, data were pooled to form a single meta-arm as the unit of analysis, as recommended in the Cochrane Handbook for Systematic Reviews of Interventions (section 16.5.4). 85 For the continuous outcome measures reported in this review, the mean for the combined arm is estimated as the weighted mean from the multiple separate arms (where the numbers in each arm provide the weights), and the SD for the combined arm is calculated according to the usual formula:

where i indexes a total of k arms being combined, n_i is the number of participants in each arm, and s_i is the SD for that arm.

All meta-analyses were stratified according to the measurement population. Where multiple-measurement populations were reported in an individual study, we used the highest ranking according to a prespecified hierarchy:

1. true ITT

2. last observation carried forward (LOCF)

3. observed cases (OCs) only.

The issue of how to deal with missing data point from dropouts in ITT analysis of dementia patients is a contentious one. Owing to the natural course of this degenerative disease, the assumption of LOCF that disease progression stops at the last data point clearly does not reflect reality. Similarly, to use OCs only (i.e. not estimating any data points after dropout) may give misleading results. A better solution may be to apply the rate of decline found in the control group to all dropouts. 86

We performed separate analyses for different periods of follow-up. The two lengths of follow-up for which data were generally available were approximately 3 months (12–16 weeks of treatment) and approximately 6 months (21–28 weeks) (figures showing these results are in the body of the text).

Where different dosages of drugs were found in various studies, we meta-analysed comparable groups separately (figures for commonly used doses in the UK are in the body of the text). We also performed a single analysis in which all dosages were combined (figures from these analyses are in Appendix 5). For continuous outcomes measured over a longitudinal period of follow-up, it is possible for investigators to report outcomes in two ways: the mean of each participant’s observed change from a measured baseline score (mean change from baseline) or absolute measurements at the relevant juncture (absolute value). If randomisation is adequate, the difference between these values should be the same (i.e. the mean of the differences will be the same as the difference in the means). However, the dispersion of each measure may vary. It is stated in the Cochrane Handbook for Systematic Reviews of Interventions (section 9.4.5.2) that ‘[t]here is no statistical reason why studies with change-from-baseline outcomes should not be combined in a meta-analysis with studies with final measurement outcomes’. 85 However, exploratory analyses showed that the inclusion of both types of data led to large differences in the results of meta-analyses, although this may be because the studies that only report final measurement data tend to be of a lower methodological standard (and, therefore, may also be more susceptible to biases that would distort reported treatment effect). As a result, we were not prepared to pool the two types of measurement, and all of our meta-analyses rely on studies reporting mean change from baseline only.

Pooling of multiple outcome measures

In addition to pair-wise meta-analyses of treatment effect pooled on each outcome’s natural scale [weighted mean difference (WMD)], we combined outcomes in a series of broad domains – cognitive, functional, behavioural and global – to investigate the overall characteristics of reported effectiveness evidence in each area (see figures in the body of the text and data sets used in the meta-analysis of pooled multiple outcome measures in Appendix 6).

In order to combine studies using different outcome measures within each domain, effect sizes were expressed as a standardised mean difference (SMD). The SMD expresses the size of the treatment effect in each trial relative to the variability observed in that trial. Accordingly, for a given trial i,

where m_1i and m_2i represent the reported means in active treatment and control cohorts, respectively, and s_i is the pooled SD across both groups, estimated as

where n_1i, n_2i and N_i represent the sample sizes of treated, control and combined cohorts, respectively, and the reported SDs of measurements in treated and control groups are SD_1i and SD_2i. In order to pool SMDs, it is necessary to derive the standard error (SE), which is estimated as follows:

The method assumes that the differences in SDs between studies reflect differences in measurement scales and not real differences in variability among study populations.

Where studies reported more than one outcome contributing to the same domain, a weighted average of all SMDs was calculated, using the precision of the estimates as the weighting factor (this could be seen as a submeta-analysis, adopting a fixed-effects model with inverse variance weighting). So that such studies were not given spurious weight, the sample size for each outcome measure was divided by the total number of outcomes.

This approach has the advantage of enabling a broader evidence base to be combined, but it has the disadvantage of requiring estimates to be pooled on a scale that has no direct clinical meaning. Accordingly, we used these analyses solely to explore the characteristics of the evidence base, and not to draw direct conclusions about the magnitude of relative effectiveness of the comparators. In particular, we used the analyses as a basis for metaregression (see below), and for assessing small-study effects.

Metaregression

Where there was sufficient evidence (at least five individual data points in a meta-analysis), study-level regression (‘metaregression’) was used to explore the statistical heterogeneity across studies. Three prespecified covariates were explored: population gender, population gender, and baseline disease severity (as measured by MMSE). Because of inconsistencies in the evidence base, it was not possible to undertake multivariate analyses, so regressions were conducted solely on a univariate basis. Metaregression was undertaken in Stata 10.1 (‘metareg’ command), using the restricted maximum likelihood estimator, as recommended. 87,88 These figures are in Appendix 7.

Mixed-treatment comparison: indirect comparison

In addition to pair-wise meta-analyses, where sufficient data were available, we synthesised information on all technologies and their comparators simultaneously, in a mixed-treatment comparison (MTC) using Bayesian Markov Chain Monte Carlo (MCMC) sampling. 89–92 The analyses were performed using WinBUGS 1.4.1 (MRC Biostatistics Unit, Cambridge UK) (model code is reproduced in Appendix 8).

Vague prior distributions were used in the analyses (Normal[0, 0.000001] for mean difference between treatments; Uniform[0,2] for SD of random-effects distribution). Point estimates and 95% credible intervals were calculated from 100,000 simulated draws from the posterior distribution after a burn-in of 10,000 iterations.

Outputs are presented in terms of treatment effect compared with a common baseline. In each case in the presented analyses, the available evidence networks included at least one placebo arm; therefore, the baseline treatment is always placebo. This is helpful, as it enables all MTC outputs to be interpreted on a common level. In addition to treatment effect relative to placebo, the posterior probability that each treatment is most effective is presented, simply calculated as the proportion of MCMC trials in which the given treatment had the highest (or lowest, for negative scales) estimated treatment of all comparators.

This approach assumes ‘exchangeability’ of treatment effect across all included trials, such that the observed treatment effect for any comparison could have been expected to arise if it had been measured in the populations reported in all other included trials. Exchangeability was judged through examination of the trial populations and comparability of outcomes in the common treatment group facilitating the comparison. Figures representing these analyses are in the body of the text.

As for pair-wise syntheses, we generated separate MTCs for different periods of follow-up (12–16 weeks and 21–28 weeks). We also generated separate analyses according to measurement population: LOCF only; ITT plus LOCF; OC only; and all measurement populations combined (see Appendix 9). Where multiple measurement populations were reported in an individual study and more than one was pertinent to one of these analyses, we used the highest ranking according to the hierarchy given above. Multiple relevant arms within a single study were pooled according to the methods detailed above (see Pair-wise meta-analysis), before being entered into the MTC.

Donepezil

No systematic reviews of donepezil were found that matched our inclusion criteria.

Galantamine

No systematic reviews of galantamine were found that matched our inclusion criteria.

Rivastigmine

Birks and colleagues94 conducted a Cochrane review of rivastigmine compared with placebo for people with mild-to-moderate AD. They found nine trials with a total of 4775 participants. The review found that the use of rivastigmine (6–12 mg daily) was associated with a two-point improvement on the ADAS-cog compared with placebo [ITT WMD –1.99 (95% CI –2.49 to –1.50)] and a 2.2-point improvement on the Progressive Deterioration Scale (PDS) for ADL [ITT WMD –2.15 (95% CI –3.16 to –1.13)] at 26 weeks. The authors concluded that rivastigmine gave benefit to people with mild-to-moderate AD when compared with placebo. The review also considered delivery of the drug by transdermal patch. It found that the lower dose patch (9.6 mg/day) was associated with fewer side effects than the capsules or the higher dose patch (17.4 mg/day), and produced similar efficacy. The main AEs were gastrointestinal (nausea and vomiting), usually occurring during the titration phase.

All cholinesterase inhibitors

The German Institute for Quality and Efficiency in Health Care (IQWiG) conducted a systematic review and meta-analysis of all of the cholinesterase inhibitors included in this report for people with mild-to-moderate AD. 95 They included RCTs up to June 2006 in their systematic review, and found 27 studies with a total of 9883 participants. Only four of these trials met our inclusion criteria. 96–99 The IQWiG concluded that all the AChEIs provided benefit in improving or maintaining cognitive function and ADLs, and that galantamine alleviated psychological symptoms. However, none of the studies provided evidence of improvement in QoL. A summary table of these results can be found in Appendix 13.

Memantine

A systematic review of memantine for dementia was carried out by Raina and colleagues. 100 Their inclusion criteria were broader than ours and included all major types of dementia, patients with mild-to-moderate disease severity and all drugs for treating dementia. Of the 59 studies they included, only two met our inclusion criteria for trials. 96,101 The data syntheses from this systematic review are not relevant to this TAR and will not be discussed.

All included drugs

Hansen and colleagues102 conducted a systematic review and meta-analysis of functional outcomes from the use of donepezil, galantamine, rivastigmine and memantine for people with mild-to-moderate AD. They included 13 RCTs, 12 of which were included in the previous TAR (No. 111). The new study, which is included in this review, is Brodaty and colleagues. 93 Overall, they found a small effect size (d = 0.1–0.4) favouring drug treatment. A metaregression showed that this effect was not affected by disease severity, age, gender and drug dose. AEs were most commonly gastrointestinal.

Identified evidence

The 2004 review2 identified 14 RCTs investigating the effectiveness of donepezil compared with placebo. 93,103–116

Notes:

• The Nunez and colleagues trial111,112 was reviewed in poster form in 2003; a full publication, authored by Johannsen and colleagues,112 is now available, from which we have extracted data; however, for consistency with the 2004 review, we continue to refer to this RCT as ‘Nunez and colleagues’.

• The Seltzer and colleagues trial115 was reviewed on a commercial-in-confidence (CiC) basis using information supplied by the manufacturer in 2004 – a full publication, authored by Seltzer and colleagues. 115

• Winblad and colleagues,116 with additional information contained in the trial of Wimo and colleagues. 117

Our searches identified an additional five RCTs. 118–122 A summary of their design characteristics can be found in Table 2 and the interventions, comparators and baseline characteristics of the participants in Table 3. Critical appraisal of these small studies showed that none was of good quality; neither study was reporting adequate randomisation nor allocation concealment. A summary of the markers of internal validity is presented in Table 4.

Evidence of clinical effectiveness

Cognition

In 2004, Loveman and colleagues2 summarised the evidence they found for donepezil versus placebo for cognitive outcomes as follows:

Six RCTs showed that donepezil appears to confer a statistically significant benefit to participants on the ADAS-cog scale when compared to placebo. The benefit varies according to the dose of donepezil with higher doses of donepezil tending to show increasing benefit. Because the mean change scores varied quite considerably between the included studies, this dose-related trend can particularly be seen within individual trials, although no direct statistical comparisons were made in any of these. The mean change scores were however varied between the included studies. Eight RCTs showed trends towards better MMSE score in the donepezil treated groups when compared to the placebo groups, although this was not always demonstrated to be statistically significant. These trends were mirrored in one unpublished trial of people with mild Alzheimer’s disease.

New data

In the studies we found published since 2004, four showed significant cognitive benefit for donepezil versus placebo. 118–121 However, only two of these trials118,122 estimated the missing values from dropouts using ITT analysis. The others made estimates using OCs only, thus potentially magnifying any benefit from donepezil and biasing their results in favour of the intervention. A summary of the results from cognitive measures can be seen in Table 5.

TABLE 5 - Measures of cognition in included studies: donepezil vs placebo

A, absolute value at specified juncture; MC, mean change from baseline at specified juncture; SKT, Syndrom Kurztest.

Analysis of variance (ANOVA).

Student’s t-test (calculated by reviewer).

Two-way ANOVA, with treatment group and treatment time as the main factors.

ANOVA, covarying age, gender, and severity of cognitive impairment at baseline.

Reported 95% CI is asymmetrical, suggesting calculation error.

The t-test.

Study	Subgroup	Outcome	Type	Donepezil		Placebo		p-value
				n	Mean (SD)	n	Mean (SD)
Moraes et al. (2008)120	OC population	ADAS-cog – 13 weeks	A	11	29.7 (15.7)	12	31.8 (18.5)	< 0.05a
Winstein et al. (2007)122	ITT population	ADAS-cog – 4 weeks	MC	5	–5 (2)	5	0 (4.85)	0.066b
		Serial Reaction Time Task – 4 weeks	MC	5	3.325 (8.39)	5	1.65 (10.1)	0.782b
dos Santos Moraes et al. (2006)119	OC population	ADAS-cog – 13 weeks	A	17	30.7 (13.9)	18	40.9 (19.4)	0.085b
		ADAS-cog – 26 weeks	A	17	28.3 (12.3)	18	42.8 (18.7)	< 0.01c
Mazza et al. (2006)118	ITT population	MMSE – 24 weeks	A	25	19.8 (3.16)	26	18.6 (3.66)	NSd
			MC	25	1.2 (12.2)	26	–0.25 (5)e	0.06d
		SKT – 24 weeks	A	25	11.8 (2.9)	26	16.9 (3.9)	0.01d
			MC	25	–3.3 (–2.55)	26	0.9 (1.3)	< 0.001d
		CGI-item 2 (cognitive) – 24 weeks	A	25	3.6 (0.94)	26	5.2 (0.95)	0.01d
			MC	25	–0.9 (1.02)	26	0.15 (0.338)	< 0.001d
Peng et al. (2005)121	OC population	MMSE – 12 weeks	A	46	22.1 (2)	43	18.7 (2.4)	< 0.01f

Synthesis with existing evidence base

The data from the new trials were synthesised with those from Loveman and colleagues report2 by random-effects meta-analysis. This was conducted considering ADAS-cog and then MMSE as the outcomes, measuring differences between donepezil (all doses) and placebo at 12 and 24 weeks post randomisation. The results can be seen in Figures 7 and 8. We also meta-analysed the data by 5 mg/day and all doses combined; these results can be found in Appendix 5. We then went on to explore the effect of pooling the entire cognitive outcome measures at 24–26 weeks; the results of this can be seen below (see Figure 11).

FIGURE 7.

Random-effects meta-analysis: ADAS-cog at 12 weeks (mean change from baseline): donepezil (10 mg/day) vsplacebo.

FIGURE 8.

Random-effects meta-analysis: ADAS-cog at 24 weeks (mean change from baseline): donepezil (10 mg/day) vsplacebo.

Alzheimer’s Disease Assessment Scale – Cognitive Subscale

We found no new studies reporting the ADAS-cog at 12 or 24 weeks. The meta-analyses presented below are of studies included in the previous assessment report. The overall pooled estimates shows a benefit from donepezil compared with placebo, which increases over time: 12 weeks’ WMD = –1.97 (95% CI –3.38 to –0.56), p = 0.006; 24 weeks’ WMD = –2.90 (95% CI –3.61 to –2.18), p < 0.001 (Figures 7 and 8).

Similarly, no new evidence was found for the outcome measure MMSE at 12 weeks post randomisation, but one new study was found with measures at 24 weeks’ follow-up. The meta-analyses below show an overall benefit from donepezil versus placebo when measured on the MMSE: 12 weeks’ (10 mg/day) WMD = –1.17 (95% CI 0.88 to 1.45), p < 0.001; 24 weeks’ (5 and 10 mg/day) WMD = 1.21 (95% CI 0.84 to 1.57), p < 0.001 (Figures 9 and 10).

FIGURE 9.

Random-effects meta-analysis: MMSE at 12 weeks (mean change from baseline): donepezil (10 mg/day) vsplacebo.

FIGURE 10.

Random-effects meta-analysis: MMSE at 24 weeks (mean change from baseline): donepezil (all dosages) vsplacebo. a, WMD and error bars provided in publication; SE estimated on assumption that error bars represent 95% CIs; b, Pooled 5 and 10 mg/day arms.

Pooled multiple outcome measures

Two new studies were found to add to this combined meta-analysis of cognitive outcome measures at 24–26 weeks. The overall pooled estimate showed a significant cognitive benefit from donepezil compared with placebo: SMD = 0.40 (95% CI 0.29 to 0.50), p < 0.001 (Figure 11). The data set used in this meta-analysis can be found in Appendix 6.

FIGURE 11.

Random-effects meta-analysis: cognitive outcomes (SMD) at 24–26 weeks: donepezil (all dosages) vs placebo.

Functional

The 2004 assessment report2 found that:

A variety of functional measures were used in eight RCTs. Donepezil had some effect in improving or limiting further deterioration on ADLs when compared with placebo, but this was not always statistically significant, particularly over longer durations of follow-up. One trial reported time to loss of ADL and/or time to institutional care and found that donepezil conferred no advantage to placebo.

New data

We found only one new RCT measuring functional outcomes for this comparison. This small, poorly reported trial showed a significant benefit from donepezil (5 mg/day) for ADLs in an OC-measured population at 12 weeks’ follow-up – mean difference: I = 40.5 (SD 7.6), C = 49.5 (SD 6.3), p < 0.01 (Table 6).

TABLE 6 - Measures of functional ability in included studies: donepezil vs placebo

A, absolute value.

The t-test.

Study	Subgroup	Outcome	Type	Donepezil		Placebo		p-value
				n	Mean (SD)	n	Mean (SD)
Peng et al. (2005)121	OC population	ADL – 12 weeks	A	46	40.5 (7.6)	43	49.5 (6.3)	< 0.01a

Synthesis with existing evidence base

When the 2004 and post-2004 evidence bases were collected together, there was an extremely heterogeneous collection of outcome measures for this domain. As a result, we have not been able to perform any quantitative synthesis of individual outcome measures on a natural scale.

Pooled multiple outcome measures

There were no new studies that measured functional outcomes at 24 weeks; therefore, we pooled the functional outcome data from the studies in the previous assessment. This showed a significant benefit for donepezil at all doses compared with placebo: SMD = 0.30 (95% CI 0.14 to 0.45), p < 0.001 (Figure 12). The data set used for this meta-analysis can be found in Appendix 6.

FIGURE 12.

Random-effects meta-analysis: functional outcomes (SMD) at 24 weeks: donepezil (all dosages) vs placebo.

Behavioural and mood

In 2004 the assessment group2 reported that:

The NPI was used as a measure of mood and behaviour in four RCTs. Data were varied but suggested that donepezil may have some effect in improving or limiting further deterioration on the NPI scale compared to placebo, at least over shorter durations of follow-up.

New data

None of the newly identified studies provided any additional data on the effect of donepezil as indicated by measures of behavioural function. Therefore, we conducted random-effects meta-analysis of the studies included in 2004 for the NPI at 12 and 24 weeks, which showed no significant benefit from donepezil measured by the NPI (Figures 13 and 14).

FIGURE 13.

Random-effects meta-analysis: NPI at 12 weeks (mean change from baseline): donepezil [all dosages (all are 10 mg/day)] vsplacebo. a, Score inverted from published figure to reflect usual direction of NPI; b, WMD and error bars provided in publication – SE estimated on assumption that error bars represent 95% CIs.

FIGURE 14.

Random-effects meta-analysis: NPI at 24 weeks (mean change from baseline): donepezil [all dosages (all are 10 mg/day)] vs placebo. a, Score inverted from published figure to reflect usual of direction NPI; b, WMD and error bars provided in publication – SE estimated on assumption that error bars represent 95% CIs.

Pooled multiple outcome measures

Because NPI is the only outcome measure used in this domain of the evidence base, it was not necessary to pool outcomes on a standardised level.

Global effect

Loveman and colleagues2 summarised their findings on global outcomes comparing donepezil and placebo as:

Seven RCTs assessed the effect of donepezil compared with placebo on the Clinical Global Impression of Change (CGIC) or CIBIC plus, showing overall that CGIC/CIBIC-plus scores were statistically significantly better with donepezil. The range of scores varied between the included studies. Higher proportions of participants receiving donepezil were considered as responders to treatment, although this was not compared statistically in many cases. On the CDR scale trends were also demonstrated towards improved global function in the donepezil-treated groups compared with the placebo groups in five trials but statistical significance was not demonstrated. In one unpublished trial with participants with mild Alzheimer’s disease, no benefit on the CDR was noted in the donepezil treated group.

New data

Only one of the new studies measured global outcomes. 121 They also found significant benefit on the CDR: I = 1.2 (SD 0.2), C = 2.0 (SD 0.2), p < 0.01 (Table 7).

TABLE 7 - Measures of global effect in included studies: donepezil vs placebo

A, absolute value.

The t-test.

Study	Subgroup	Outcome	Type	Donepezil		Placebo		p-value
				n	Mean (SD)	n	Mean (SD)
Peng et al. (2005)121	OC population	CDR – 12 weeks	A	46	1.2 (0.2)	43	2 (0.2)	< 0.01a

Synthesis with existing evidence base

Clinician interview-based impression of change-plus

Only the previously included studies had data for meta-analysis of the CIBIC-plus. We pooled studies at 12 and 24 weeks and found that at both time points there was a significant overall pooled estimate of benefit from donepezil at 10 mg/day compared with placebo [12 weeks’ WMD = –0.38 (95% CI –0.49 to –0.26), p < 0.001; 24 weeks’ WMD = –0.43 (95% CI –0.55 to –0.31), p < 0.001] (Figures 15 and 16). Meta-analyses of CIBIC-plus results for 5 mg/day and all doses combined can be found in Appendix 5.

FIGURE 15.

Random-effects meta-analysis: CIBIC-plus at 12 weeks (mean change from baseline): donepezil (10 mg/day) vs placebo.

FIGURE 16.

Random-effects meta-analysis: CIBIC-plus at 24 weeks (mean change from baseline): donepezil (10 mg/day) vs placebo.

Clinical Dementia Rating

The pooled results on the CDR scale showed a significant advantage from taking donepezil at 12 and 24 weeks’ follow-up: 12 weeks’ WMD = –0.26 (95% CI –0.44 to –0.09), p < 0.003 and 24 weeks’ WMD = –0.57 (95% CI –0.85 to –0.29), p < 0.001 (Figures 17 and 18).

FIGURE 17.

Random-effects meta-analysis: CDR at 12 weeks (mean change from baseline): donepezil (all dosages) vs placebo. a, Pooled 5- and 10 mg/day arms.

FIGURE 18.

Random-effects meta-analysis: CDR at 24 weeks (mean change from baseline): donepezil (all dosages) vs placebo. a, 5 and 10 mg/day arms pooled.

Pooled multiple outcome measures

We did not find any new studies that measured global outcomes at 24–26 weeks; therefore, we pooled the global outcome data from the studies in the previous assessment. This showed a significant benefit for donepezil at all doses compared with placebo: SMD = 0.38 (95% CI 0.27 to 0.48), p < 0.001 (Figure 19). The data set used in this meta-analysis can be found in Appendix 6.

FIGURE 19.

Random-effects meta-analysis: global outcomes (SMD) at 24–26 weeks: donepezil (all dosages) vs placebo.

Metaregression

We also conducted metaregression analysis to explore the statistical heterogeneity across studies, looking at age, age and baseline MMSE (as a proxy for disease severity). Only one graph showed a significant relationship; this was between baseline MMSE and functional outcomes at 24 weeks for all doses of donepezil giving a metaregression estimate of α = 1.456, β = –0.065, p = 0.009. However, owing to the small number of studies in each analysis and the fact that the data were assessed at a population level (which may not reflect the individual level) we felt that these results may be ambiguous and so have placed them in Appendix 7 in case they are of interest.

Summary: donepezil versus placebo

We found an additional five RCTs118–122 to add to the 14 previously found by Loveman and colleagues;2 none of the new studies was of good quality or had a follow-up of longer than 6 months.

Pooled cognitive outcomes showed a significant benefit from donepezil measured by the ADAS-cog and MMSE with greater benefit shown at 24 weeks [ADAS-cog: WMD = –2.90 (95% CI –3.61 to –2.18), p < 0.001; MMSE: WMD = 1.21 (95% CI 0.84 to 1.57), p < 0.001]. The pooled estimates of all cognitive outcomes likewise showed a benefit from donepezil at 24–26 weeks’ follow-up.

Only one new study looked at functional outcomes for this comparison; at 12 weeks this showed a significant gain for those taking donepezil [mean difference: I = 40.5 (SD 7.6), C = 49.5 (SD 6.3), p < 0.01)]. At 24 weeks there were data from only the 2004 assessment trials, and the results from all the studies reporting functional outcomes were pooled; this analysis again showed a significant benefit from taking donepezil.

None of the new studies measured behavioural outcomes; the pooled estimates from the previous assessment, using the NPI, failed to show a significant gain on behavioural outcomes at either 12 or 24 weeks.

Just one new study121 looked at global outcomes; this showed a benefit from taking donepezil on the CDR [I = 1.2 (SD 0.2), C = 2.0 (SD 0.2), p < 0.01)]. The pooled results for the CIBIC-plus scale were only from the previous TAR and they showed a significant advantage from donepezil at 12 and 24 weeks’ follow-up [24 weeks: WMD = –0.43 (95% CI –0.55 to –0.31), p < 0.001]. When both the global outcome measures were pooled at 24–26 weeks, the results again showed a significant benefit from donepezil.

None of the new studies provided additional data on QoL or safety under randomised conditions.

The new studies found have added to the body of evidence showing a benefit from donepezil compared with placebo for cognitive, functional and global outcomes. However, there is no new or pooled evidence to show a behavioural benefit from donepezil versus placebo in people with mild-to-moderate AD. All but two of the studies included in these meta-analyses calculated their missing data points using LOCF or OC methods, thereby potentially biasing their results in favour of donepezil.

Graphical summary of donepezil versus placebo

The first thing that is noticeable when comparing Figures 20–22 is that no new large studies have been undertaken since 2004 and that only one study was multicentre. Closer examination shows that the quality of trials has not improved and, with the exception of Peng and colleagues, studies measured only cognitive abilities. As in the previous review, these outcomes showed that patients benefited from taking donepezil. Although the new trials add to the precision of our knowledge of the effects of donepezil on cognitive measures in mixed mild/moderate AD populations, none of the new studies was in the mild AD population. This means that we are still dependent on one moderately sized RCT from the 2004 review that looked at cognitive outcomes in the mild AD population, which showed a cognitive benefit for this group from taking 10 mg/day donepezil.

FIGURE 20.

Summary of studies included in the 2004 review: donepezil vs placebo. ADCS-ADL, Alzheimer’s Disease Cooperative Study – Activities of Daily Living Inventory; ADCS-CGIC, Alzheimer’s Disease Cooperative Study – Clinical Global Impression of Change; Analy, analysis; be, behavioural; Blind, blinding; Char, characteristics of patients; cog, cognitive; DAD, Disability Assessment for Dementia Scale; F, female; func, functional; GDS, Global Deterioration Scale; glo, global; M, male; Rand, randomisation; SIB, Severe Impairment Battery.

FIGURE 21.

Summary of studies included in the 2004 review: donepezil vs placebo (cont.). ADCS-ADL, Alzheimer’s Disease Cooperative Study – Activities of Daily Living Inventory; ADCS-CGIC, Alzheimer’s Disease Cooperative Study – Clinical Global Impression of Change; Analy, analysis; be, behavioural; Blind, blinding; Char, characteristics of patients; cog, cognitive; DAD, Disability Assessment for Dementia Scale; F, female; func, functional; GDS, Global Deterioration Scale; glo, global; M, male; Rand, randomisation; SIB, Severe Impairment Battery. *Median value reported instead of mean.

FIGURE 22.

Summary of new studies included in the 2010 review: donepezil vs placebo. ADCS-ADL, Alzheimer’s Disease Cooperative Study – Activities of Daily Living Inventory; ADCS-CGIC, Alzheimer’s Disease Cooperative Study – Clinical Global Impression of Change; Analy, analysis; be, behavioural; Blind, blinding; Char, characteristics of patients; cog, cognitive; DAD, Disability Assessment for Dementia Scale; F, female; func, functional; GDS, Global Deterioration Scale; glo, global; M, male; Rand, randomisation; SIB, Severe Impairment Battery.

Identified evidence

For the pair-wise comparison between galantamine and placebo, the 2004 review included six RCTs. 123–129 However, it is apparent that two of these publications124,129 report the same population. Accordingly, we have entered only Rockwood and colleagues’ primary publication124 in syntheses, below.

We identified an additional three moderately good-to-poor-quality RCTs, meeting few of the quality criteria indicated above (see Critical appraisal strategy).

The primary publication of the GAL-INT-6 study, written by Erkinjuntti and colleagues in 2002,130 was correctly excluded from the 2004 review because it conflated participants with multiple forms of dementia. However, we were able to include a subsequent paper – Bullock and colleagues 2004101 – reporting the AD-specific subgroup of this trial. A summary of their design characteristics can be found in Table 8 and the interventions, comparators and baseline characteristics of the participants in Table 9. A summary of the markers of internal validity is presented in Table 10.

Evidence of clinical effectiveness

Cognition

The previous review2 summarised the evidence from cognition outcomes thus:

Six published RCTs showed that galantamine appears to confer a statistically significant benefit to participants on the ADAS-cog scale when compared to placebo, whether reducing the deterioration or leading to some improvement in their condition. The benefit varies depending upon the dose of galantamine. The galantamine–placebo differences in ADAS-cog for 8 mg/day was 1.3 points, 16 mg/day 3.1 points, 18 mg/day 1.7 points, 16 or 24 mg/day 2.5 to 2.8 points; 24 to 32 mg/day 1.7 to 3.4 points and 36 mg/day 2.3 points. The one unpublished RCT mirrored these positive effects of galantamine versus placebo. In addition, some 14–17% more of galantamine participants were classified as responders (improving by 4 or more points on the ADAS-cog) than those on placebo.

New data

The results from the three new studies96,101,131 show that overall those treated with galantamine had improved scores on the ADAS-cog, whereas those in the control group remained stable or declined. However, missing data were accounted for using LOCF and OC methods, leading to a potential overestimate of the treatment effect. 86 Table 11 summarises the cognitive results from the new studies.

TABLE 11 - Measures of cognition in included studies: galantamine vs placebo

A, absolute value; b.i.d., twice a day; MC, mean change; o.d., once a day.

Estimated from figure.

Student’s t-test (calculated by reviewer).

Galantamine prolonged release, once a day.

Galantamine twice a day.

Value given as ‘+2’ in paper; assumed to be a typographical error, as all other observations are negative, and corroborated by OC data, which are otherwise relatively consistent with LOCF figures.

Analysis of variance (ANOVA) with factors for treatment and pooled country (USA vs ex-USA).

Study	Subgroup	Outcome	Type	Arm	Galantamine		Placebo		p-value
					N	Mean, n (SD)	N	Mean, n (SD)
Bullock et al. (2004)101	OC population	ADAS-cog 6 weeks			148	–0.5 (4.62)a	85	0.15 (6.26)a	0.366b
		ADAS-cog 13 weeks	MC		148	–1.48 (4.32)a	85	0 (6.03)a	0.031b
		ADAS-cog 26 weeks	MC		147	–1.1 (5.79)	83	2 (5.56)	< 0.001b
			A		147	21.5 (10.5)	83	25.7 (12)	0.006b
Brodaty et al. (2005)96	LOCF analysis	ADAS-cog 8 weeks	MC	o.d.c	287	–1.5 (5.08)	293	0 (5.14)	–
				b.i.d.d	294	–1.7 (4.97)			–
		ADAS-cog 12 weeks	MC	o.d.c	290	–2f (5.28)	296	0.2 (5.33)	–
				b.i.d.d	296	–2.5 (5.16)			–
		ADAS-cog 26 weeks	MC	o.d.c	291	–1.3 (5.29)	296	1.2 (5.68)	< 0.001f
				b.i.d.d	296	–1.6 (6.19)			< 0.001f
	OC population	ADAS-cog 8 weeks	MC	o.d.c	284	–1.5 (5.06)	289	0 (5.1)	–
				b.i.d.d	286	–1.7 (5.07)			–
		ADAS-cog 12 weeks	MC	o.d.c	269	–2.2 (5.25)	275	0 (5.14)	–
				b.i.d.d	268	–2.6 (5.07)			–
		ADAS-cog 26 weeks	MC	o.d.c	240	–1.4 (5.27)	248	1.3 (5.67)	< 0.001f
				b.i.d.d	227	–1.8 (6.33)			< 0.01f
Rockwood et al. (2006)131	LOCF analysis	ADAS-cog 8 weeks	MC		62	–1.85 (4.18)	65	–0.25 (4.97)	–
		ADAS-cog 16 weeks	MC		62	–1.6 (5.38)	65	0.325 (5.49)	–

Synthesis with existing evidence base

We pooled data from the new studies96,101,131 with those of the 2004 assessment2 by random-effects meta-analysis using the ADAS-cog as the outcome measure for ≤ 24 mg/day at 12–16 weeks and 21–26 weeks post randomisation. No studies reported the MMSE. We also meta-analysed the data by > 12 mg/day and all doses combined; these results can be found in Appendix 5.

Alzheimer’s Disease Assessment Scale Cognitive Subscale

Two new studies were included in the meta-analysis. The overall pooled estimates showed a benefit from galantamine compared with placebo at 12–16 and 21–26 weeks: 12–16 weeks WMD = –2.39 (95% CI –2.80 to –1.97), p < 0.001; 21–26 weeks WMD = –2.96 (95% CI –3.41 to –2.51), p < 0.001 (Figures 23 and 24).

FIGURE 23.

Random-effects meta-analysis: ADAS-cog at 12–16 weeks (mean change from baseline) – galantamine (maximum dose ≤ 24 mg/day) vsplacebo. a, 18 and 24 mg/day arms pooled; b, once daily prolonged-release formulation and twice daily standard formulation pooled; c, 8, 16 and 24 mg/day arms pooled.

FIGURE 24.

Random-effects meta-analysis: ADAS-cog at 21–26 weeks (mean change from baseline) – galantamine (maximum dose ≤ 24 mg/day) vs placebo. a, 8, 16 and 24 mg/day arms pooled; b, once daily prolonged-release formulation and twice daily standard formulation pooled.

Pooled multiple outcome measures

Because ADAS-cog is the only outcome measure used to assess cognitive effect in the placebo-controlled trials of galantamine, it was not necessary to pool outcomes on a standardised level for this domain.

Functional

In 20042 the assessment group reported for functional outcomes that:

Three RCTs assessed mean changes from baseline on the DAD scale, all reporting statistically significantly slower deterioration for those receiving galantamine 24–32 mg/day compared to placebo. Two RCTs found that participants receiving 16 mg/day and/or 24 mg/day galantamine experienced a statistically significantly smaller deterioration on the ADCS/ADL compared to placebo.

New data

All three new RCTs measured functional outcomes. 96,101,131 They found that functional abilities had improved significantly more in the treatment group. Table 12 summarises the results.

TABLE 12 - Measures of functional effect in included studies: galantamine vs placebo

A, absolute value; b.i.d., twice a day; CR, clinician rated; D, dichotomous; GAS, Goal Attainment Scale; MC, mean change; o.d., once a day; PCR, patient and caregiver rated; VR, video rated.

Data extracted from publication using IPD in pooled analysis. 130

Test not specified.

Galantamine prolonged-release once a day.

Sample size not provided (must be ≥ no. of 26 weeks’ OCs); precision likely to be underestimated.

Analysis of variance (ANOVA) with factors for treatment and pooled country (USA vs ex-USA).

Galantamine twice a day.

Different values for SE given in table 2 (1.12) and figure 4 (0.57) of publication; latter used as closer to range of dispersion reported in other arms. 96

Clinician rated.

Analysis of variance.

Patient/caregiver rated.

Data extracted from secondary publication reporting subgroup with verbal repetition goals. 131

Mixed-effects model, with dementia severity and treatment assignment as fixed effects, and the patient as the random effect.

Study	Subgroup	Outcome			Type	Arm	Galantamine		Placebo		p-value
							N	Mean, n (SD%)	N	Mean, n (SD %)
Bullock et al. (2004)101	LOCF analysis	DAD		26 weeks	MC		188	–1 (15.8)a	97	–6 (14.5)a	< 0.01b
Brodaty et al. (2005)96	LOCF analysis	ADCS-ADL		26 weeks	MC	o.d.c	245d	0 (7.51)	258d	–2.7 (0.899)	< 0.001e
					MC	b.i.d.f	242d	–1 (0.778)			0.018e
	OC population	ADCS-ADL		8 weeks	MC	o.d.c	280	0.8 (6.86)	294	–0.7 (7.72)	–
					MC	b.i.d.f	292	0.9 (7.18)			–
				12 weeks	MC	o.d.c	276	0.4 (6.65)	281	–0.3 (7.71)	–
					MC	b.i.d.f	279	1.1 (7.85)			–
				26 weeks	MC	o.d.c	245	0 (8.61)	258	–2.4 (9.64)	0.003e
					MC	b.i.d.f	242	–1 (8.87)g			0.088e
Rockwood et al. (2006)131	LOCF analysis	GAS (CR)h		8 weeks	A		61	52.5 (9.12)	66	52.2 (6.97)	–
				16 weeks	A		61	54.8 (9.36)	66	50.9 (9.74)	0.02i
		GAS (PCR)j		8 weeks	A		61	54.6 (7.97)	66	52.5 (8.57)	–
				16 weeks	A		61	54.2 (10.8)	66	52.3 (9.12)	0.27i
	OC populationk	GAS-VR	Improved	16 weeks	D		20	14 (70.0)	30	8 (26.7)	< 0.01l
			Worsened	16 weeks	D		20	2 (10.0)	30	10 (33.3)	–
			No change	16 weeks	D		20	4 (20.0)	30	12 (40.0)	–

Synthesis with existing evidence base

The data from the new trials96,101,131 were pooled, by random-effects meta-analysis, with those of the studies found in 2004. The outcome measures considered were the Alzheimer’s Disease Cooperative Study – Activities of Daily Living (ADCS-ADL) Inventory and the Disability Assessment of Dementia (DAD) Scale.

Alzheimer’s Disease Cooperative Study – Activities of Daily Living

Results from the ADCS-ADL were pooled at 12–13 weeks’ follow-up and at 21–26 weeks’ follow-up. The overall pooled estimates showed functional benefit from galantamine compared with placebo: 12–13 weeks’ WMD = 1.39 (95% CI 0.59 to 2.20), p < 0.001; 21–26 weeks WMD = 2.23 (95% CI 1.33 to 3.14), p < 0.001 (Figures 25 and 26).

FIGURE 25.

Random-effects meta-analysis: ADCS-ADL at 12–13 weeks (mean change from baseline): galantamine (≤ 24 mg/day]) vs placebo. a, 8, 16 and 24 mg/day arms pooled; b, once daily prolonged-release formulation and twice daily standard formulation pooled.

FIGURE 26.

Random-effects meta-analysis: ADCS-ADL at 21–26 weeks (mean change from baseline): galantamine (≤ 24 mg/day]) vs placebo. a, 8, 16 and 24 mg/day arms pooled; b, once daily prolonged-release formulation and twice daily standard formulation pooled.

Disability Assessment for Dementia Scale

The results from the DAD were pooled at 21–26 weeks’ follow-up. They again showed a significant benefit from galantamine compared with placebo: WMD = 3.76 (95% CI 1.66 to 5.86), p < 0.001 (Figure 27).

FIGURE 27.

Random-effects meta-analysis: DAD at 21–26 weeks (mean change from baseline): galantamine (≤ 4 mg/day) vs placebo. a, 24 and 32 mg/day arms pooled.

Pooled multiple outcome measures

Two new studies were added to the meta-analysis of combined functional outcome measures at 21–26 weeks. The overall pooled estimate showed a significant functional benefit from galantamine compared with placebo: SMD = 0.27 (95% CI 0.18 to 0.34), p < 0.001 (Figure 28). The data set used in this meta-analysis can be found in Appendix 6.

FIGURE 28.

Random-effects meta-analysis: functional outcomes (SMD) at 21–26 weeks: galantamine (all dosages) vs placebo.

With only four data points in this evidence base, it would not be informative to perform metaregression.

Behavioural and mood

In 2004, Loveman and colleagues2 summarised the behavioural results from this comparison as:

Two published and one unpublished RCTs found that galantamine had some effect in improving or limiting further deterioration on the NPI scale compared to placebo. Differences in the mean change from baseline were statistically significant for doses of 16 mg/day or over in one of the three studies.

New data

Only one included study96 provided additional data on the effectiveness of galantamine in relieving behavioural symptoms of AD when compared with placebo. However, this failed to show any statistically significant benefit. Data are shown in Table 13.

TABLE 13 - Measures of behavioural effect and mood in included studies: galantamine vs placebo

b.i.d. twice a day; MC, mean change from baseline; o.d., once a day.

Galantamine prolonged-release formulation once a day.

Sample size not provided (must be ≥ no. of 26 weeks OCs); precision likely to be underestimated.

Analysis of variance (ANOVA) with factors for treatment and pooled country (USA vs ex-USA).

Galantamine standard formulation twice a day.

Study	Subgroup	Outcome	Type	Arm	Galantamine		Placebo		p-value
					N	Mean, n (%)	N	Mean, n (%)
Brodaty et al. (2005)96	LOCF analysis	NPI – 26 weeks	MC	o.d.a	245b	–0.6 (10.3)	258b	0.6 (9.96)	0.941c
				b.i.d.d	242b	–0.9 (11.4)			0.102c
	OC population	NPI – 26 weeks	MC	o.d.a	245	–0.6 (10.8)	258	0.1 (13.2)	0.451c
				b.i.d.d	242	–1.2 (12.9)			0.203c

Synthesis with existing evidence base

Neuropsychiatric Inventory

Only one new study added evidence to this meta-analysis. We looked for estimates of effectiveness at 13 and 21–26 weeks; at 13 weeks no significant benefit was found. However, at 21–26 weeks the overall pooled estimate favoured galantamine: WMD = –1.46 (95% CI –2.59 to –0.34), p = 0.012 (Figures 29 and 30).

FIGURE 29.

Random-effects meta-analysis: NPI at 13 weeks (mean change from baseline) – galantamine (all dosages) vs placebo. a, 8, 16 and 24 mg/day arms pooled.

FIGURE 30.

Random-effects meta-analysis: NPI at 21–26 weeks (mean change from baseline) – galantamine (all dosages) vs placebo. a, 8, 16 and 24 mg/day arms pooled; b, once daily prolonged-release formulation and twice daily standard formulation pooled.

Pooled multiple outcome measures

Because the NPI is the only outcome measure used in this domain of the evidence base, it was not necessary to pool outcomes on a standardised level.

Global effect

The previous assessment in 20042 summarised the effectiveness of galantamine on global outcomes:

Five published and one unpublished RCT assessed the effect of galantamine compared to placebo on the CIBIC plus, individually showing that higher proportions of participants receiving galantamine experience improvement in their condition compared to those on placebo (0% to 6.5% more participants). In contrast, a higher proportion of placebo participants tend to deteriorate (4% to 18% more participants). Also, a higher proportion of galantamine compared to placebo participants were considered to be responders to treatment with differences of between 4% (8 mg/day) and 17% (24 mg/day). When studies are pooled there are no statistically significant effects demonstrated.

New data

Two new studies were found that measured global outcomes. 96,131 Rockwood and colleagues131 found a benefit from galantamine measured by the CIBIC-plus compared with placebo at 13–16 weeks (Table 14).

TABLE 14 - Measures of global effect in included studies: galantamine vs placebo

A, absolute value; b.i.d., twice a day; D, dichotomous; NS, not statistically significant; o.d., once a day.

Galantamine prolonged-release formulation once a day.

Cochran–Mantel–Haenszel statistic using modified ridit scores, derived from rank score (the van Elteren test) and controlling for country effect (USA vs ex-USA).

Galantamine standard formulation twice a day.

Chi-squared test (Yates’ correction) (calculated by reviewer).

Test not stated.

Study	Subgroup	Outcome	Type	Arm	Galantamine		Placebo		p-value
					N	Mean, n (%)	N	Mean, n (%)
Brodaty et al. (2005)96	LOCF analysis	CIBIC-plus score – 26 weeks	A	o.d.a	291	4.21 (SD 1.1)	301	4.35 (SD 1.14)	NSb
				b.i.d.c	302	4.21 (SD 1.07)			NSb
		CIBIC-plus: markedly improved – 26 weeks	D	o.d.a	291	3 (1.0)	301	3 (1.0)	0.712d
				b.i.d.c	302	3 (1.0)			0.685d
		CIBIC-plus: moderately improved – 26 weeks	D	o.d.a	291	14 (4.8)	301	11 (3.7)	0.621d
				b.i.d.c	302	15 (5.0)			0.553d
		CIBIC-plus: minimally improved – 26 weeks	D	o.d.a	291	49 (16.8)	301	48 (15.9)	0.856d
				b.i.d.c	302	46 (15.2)			0.897d
		CIBIC-plus: no change – 26 weeks	D	o.d.a	291	114 (39.2)	301	111 (36.9)	0.623d
				b.i.d.c	302	127 (42.1)			0.224d
		CIBIC-plus: minimally worse – 26 weeks	D	o.d.a	291	81 (27.8)	301	80 (26.6)	0.802d
				b.i.d.c	302	78 (25.8)			0.907d
		CIBIC-plus: moderately worse – 26 weeks	D	o.d.a	291	24 (8.2)	301	41 (13.6)	0.050d
				b.i.d.c	302	30 (9.9)			0.201d
		CIBIC-plus: markedly worse – 26 weeks	D	o.d.a	291	6 (2.1)	301	7 (2.3)	0.951d
				b.i.d.c	302	3 (1.0)			0.336d
	OC population	CIBIC-plus score – 26 weeks	A	o.d.a	246	4.19 (SD 1.13)	259	4.36 (SD 1.15)	NSb
				b.i.d.c	240	4.21 (SD 1.11)			NSb
		CIBIC-plus: markedly improved – 26 weeks	D	o.d.a	246	3 (1.2)	259	3 (1.2)	0.728d
				b.i.d.c	240	3 (1.3)			0.751d
		CIBIC-plus: moderately improved – 26 weeks	D	o.d.a	246	14 (5.7)	259	9 (3.5)	0.327d
				b.i.d.c	240	24 (5.8)			0.298d
		CIBIC-plus: minimally improved – 26 weeks	D	o.d.a	246	43 (17.5)	259	41 (15.8)	0.705d
				b.i.d.c	240	36 (15.0)			0.895d
		CIBIC-plus: no change – 26 weeks	D	o.d.a	246	90 (36.6)	259	94 (36.3)	0.981d
				b.i.d.c	240	93 (38.8)			0.636d
		CIBIC-plus: minimally worse – 26 weeks	D	o.d.a	246	69 (28.0)	259	70 (27.0)	0.875d
				b.i.d.c	240	67 (27.9)			0.903d
		CIBIC-plus: moderately worse – 26 weeks	D	o.d.a	246	23 (9.3)	259	36 (13.9)	0.146d
				b.i.d.c	240	25 (10.4)			0.294d
		CIBIC-plus: markedly worse – 26 weeks	D	o.d.a	246	4 (1.6)	259	6 (2.3)	0.812d
				b.i.d.c	240	2 (0.8)			0.336d
Rockwood et al. (2006)131	LOCF analysis	CIBIC-plus score – 8 weeks	A		61	3.64 (SD 0.797)	65	4.17 (SD 0.905)	–
		CIBIC-plus score – 16 weeks	A		61	3.67 (SD 0.996)	65	4.12 (SD 0.987)	0.03e

Synthesis with existing evidence base

Clinician’s Interview-based Impression of Change

When the new studies’ data were pooled with the existing evidence base, the overall pooled estimates of the CIBIC-plus at 26 weeks showed a benefit from galantamine compared with placebo: WMD = –0.20 (95% CI –0.30 to –0.09), p < 0.001 (Figure 31).

FIGURE 31.

Random-effects meta-analysis: CIBIC-plus at 26 weeks – galantamine (maximum dose ≥ 24 mg/day) vs placebo. a, once daily prolonged-release formulation and twice daily standard formulation pooled.

Pooled multiple outcome measures

Because CIBIC-plus is the only outcome measure used to assess global effect in the placebo-controlled trials of galantamine, there was no reason to pool outcomes on a standardised level for this domain.

Summary: galantamine versus placebo

We found an additional three RCTs96,101,131 to add to the five reported in the 2004 review. 2

Overall cognitive results from the new studies using ADAS-cog showed improvements for those taking galantamine. When these studies were pooled with the existing evidence the benefit remained and increased with time, with greater benefit seen at 21–26 weeks [WMD = –2.96 (95% CI –3.41 to –2.51), p < 0.001] than 12–16 weeks [WMD = –2.39 (95% CI –2.80 to –1.97), p < 0.001].

All of the new studies reported functional outcomes. Those measured by the DAD and ADCS-ADL scales generally showed significant improvement; those measured by the Goal Attainment Scale (GAS) were rather more ambiguous. Pooled results of the ADCS-ADL and the DAD at 21–26 weeks continued to show benefit from galantamine compared with placebo [WMD = 2.23 (95% CI 1.33 to 3.14), p < 0.001; WMD = 3.76 (95% CI 1.66 to 5.86), p < 0.001, respectively]. When data from both these outcome measures were pooled, results still favoured galantamine.

Behavioural outcomes from one new study, measured by the NPI, failed to show a benefit from galantamine. This lack of benefit was also seen from the pooled results at 13 weeks from follow-up. However, when the new data were pooled with those of the previous assessment2 a significant difference favouring galantamine was found at 21–26 weeks [WMD = –1.46 (95% CI –2.59 to –0.34), p = 0.012].

Two of the new studies measured global outcomes; one found that it produced a significant benefit on the CIBIC-plus. When these data were pooled with the data from the previous review, significant benefit was found on the CIBIC-plus at 26 weeks’ follow-up: WMD = –0.20 (95% CI –0.30 to –0.09), p < 0.001, with doses of ≤ 24 mg/day.

No QoL data were reported in either the new or the old studies for this comparison. The main AEs found were gastrointestinal.

The evidence from the new studies confirmed the benefits from galantamine for cognitive outcomes and, although there was some ambiguity with functional measures, the overall pooled estimates were favourable. Although none of the new studies showed significant benefit for behavioural outcomes, when these data were pooled with existing evidence, gains were shown at later follow-up times. Again, although the new trial data varied in its significance, pooling suggested that there were gains to be made on global outcomes. However, it should be noted that in all these studies estimates for missing data were calculated using LOCF or OC methods and may therefore have inflated the benefits from galantamine.

Graphical summary of galantamine versus placebo

In contrast with the donepezil research, one large study has been conducted since 2004 (comparing two different methods of delivery versus placebo), and the quality of the studies has improved overall (Figures 32 and 33). As in the previous review, the larger studies are more likely to show a benefit from galantamine on cognitive outcomes. The evidence for an effect on functional outcomes continues to be mixed: previously there appeared to be a relationship to the size of the dose – in the new study by Brodaty and colleagues96 effectiveness at 24 mg/day was linked to mode of delivery, with the prolonged-release capsule (PRC) showing a significant gain over placebo, which the regular capsule did not. Similar to the 2004 review,2 the results on global outcomes were mixed; thus it remains unclear what effect galantamine has on these outcomes.

FIGURE 32.

Summary of studies included in the 2004 review:2 galantamine vs placebo. ADCS-CGIC, Alzheimer’s Disease Cooperative Study – Clinical Global Impression of Change; Analy, analysis; be, behavioural; Blind, blinding; Char, characteristics of patients; cog, cognitive; F, female; func, functional; GDS, Global Deterioration Scale; glo, global ; M, male; Rand, randomisation, SIB, Severe Impairment Battery.

FIGURE 33.

Summary of new studies included in the 2010 review: galantamine vs placebo. a. Significant only for ’clinician-rated’ outcome measure, not for ’patient/caregiver rated’. ADCS-CGIC, Alzheimer’s Disease Cooperative Study – Clinical Global Impression of Change; Analy, analysis; be, behavioural; Blind, blinding; Char, characteristics of patients; cog, cognitive; F, female; func, functional; GDS, Global Deterioration Scale; glo, global; M, male; Rand, randomisation; SIB, Severe Impairment Battery.

Identified evidence

The 20042 review included four RCTs. 134–137

Our searches identified three additional relevant trials,138–140 details of which are tabulated in Table 16. Table 17 contains information about studies’ interventions, comparators and baseline characteristics and Table 18 gives key markers of internal validity for the included studies, which were moderately good to poor quality.

Evidence of clinical effectiveness

Cognition

The 2004 assessment report by Loveman and colleagues2 summarised the evidence they found on cognitive outcomes as:

Statistically significant differences between the 6–12 mg/day treatment groups and placebo were reported by two of three published trials which reported ADAS-cog and MMSE. No statistically significant effects were seen in the low-dose treatment groups in these studies. However, sample sizes were very low (< 30 participants in each group) and this study presented no information on power calculations.

The unpublished studies both found statistically better mean changes from ADAS-cog baseline scores in participants taking rivastigmine compared with placebo groups, a statistically significantly higher percentage of participants receiving t.i.d. (× 3 daily). Rivastigmine showed an improvement of at least four points compared with the placebo group, but there was no statistically significant difference between the b.i.d. (× 2 daily) group and placebo. One of the studies also reported on ADAS-cogA, and found statistically significant differences in both mean change from baseline and percentage of improvers for both b.i.d. and t.i.d. treatment groups compared with placebo.

Both unpublished studies reported MMSE as an outcome measure and found a statistically significant improvement in participants receiving rivastigmine compared with those receiving a placebo, with the exception of the 9 mg/day rivastigmine group.

New data

In the three studies138–140 we found that had been published since 2004, comparing rivastigmine with placebo for mild-to-moderate AD, all studies reported benefits for the treatment group on cognitive outcome measures. These results appear to be dose dependent (as in the previous report), with doses ≥ 12 mg/day showing a greater effect (see Appendix 5). However, only one study measured missing outcomes with an ITT population. 138 Table 19 shows the summary results for cognition in the included studies. It should be noted that in the study by Winblad and colleagues140 only the 10-cm patch is currently licensed in the UK.

TABLE 19 - Measures of cognition in included studies: rivastigmine vs placebo

A, absolute value; b.i.d., twice a day; CGIC, Clinical Global Impression of Change; D, dichotomous; MC, mean change; NS, not statistically significant; t.i.d., three times a day; TMT, trail-making test; WMS, Wechsler Memory Scale.

Estimated from figure.

The t-test, using pooled error term from analysis of covariance (ANCOVA)/analysis of variance (ANOVA) (SAS type III analysis).

Mantel–Haenszel test blocking for centre.

Mantel–Haenszel test.

Publication does not explicitly state population in which outcomes were measured; description of withdrawals gives the impression that data may represent final OC population, which is what we have assumed.

Student’s t-test (two-tailed) (calculated by reviewer).

Post hoc Tukey test.

A 10-cm² rivastigmine patch – equivalent to 9.5 mg/day.

Two-way ANCOVA (explanatory variables: treatment, country and baseline scores).

A 20-cm² rivastigmine patch – equivalent to 17.4 mg/day.

Cochran–Mantel–Haenszel van Elteren test using modified ridit scores stratified by country.

Study	Subgroup	Outcome	Type	Arm	Rivastigmine		Placebo		p-value
					n 1	Mean 1 (SD)	n 2	Mean 2 (SD)
Feldman and Lane (2007)138	ITT population	ADAS-cog – 12 weeks	MC	b.i.d.	227	–1.9 (6.66)a	220	0.9 (5.93)a	< 0.001b
				t.i.d.	228	–0.8 (6.04)a			< 0.05b
		ADAS-cog – 18 weeks	MC	b.i.d.	227	–1.6 (6.66)a	220	1.8 (6.67)a	< 0.001c
				t.i.d.	228	–0.1 (6.79)a			< 0.001c
		ADAS-cog – 26 weeks	MC	b.i.d.	227	–0.2 (7.3)	220	2.8 (7.2)	≤ 0.001c
				t.i.d.	228	1.2 (7.2)			< 0.05c
		ADAS-cog: any improvement – 12 weeks	D	b.i.d.	227	68 (30.0%)a	220	36 (16.4%)a	≤ 0.001d
				t.i.d.	228	52 (22.8%)a			< 0.05d
		ADAS-cog: any improvement – 18 weeks	D	b.i.d.	227	75 (33.0%)a	220	28 (12.7%)a	≤ 0.001d
				t.i.d.	228	57 (25.0%)a			≤ 0.001d
		ADAS-cog: any improvement – 26 weeks	D	b.i.d.	227	52 (22.9%)a	220	28 (12.7%)a	NSd
				t.i.d.	228	41 (18.0%)a			< 0.05d
		ADAS-cogA – 26 weeks	MC	b.i.d.	227	–0.1 (7.9)	220	3.2 (7.8)	≤ 0.001c
				t.i.d.	228	1.5 (7.8)			< 0.05c
		MMSE – 26 weeks	MC	b.i.d.	227	0.3 (3.6)	220	–1.4 (3.6)	≤ 0.001b
				t.i.d.	227	–0.6 (3.6)			< 0.05b
	LOCF analysis	ADAS-cog – 26 weeks	MC	b.i.d.	209	–0.7 (6.9)	208	2.7 (6.8)	≤ 0.001c
				t.i.d.	199	0.8 (6.9)			< 0.05c
		ADAS-cogA – 26 weeks	MC	b.i.d.	209	–0.6 (7.5)	208	3.1 (7.4)	≤ 0.001c
				t.i.d.	199	1 (7.5)			< 0.05c
		MMSE – 26 weeks	MC	b.i.d.	193	0.4 (3.4)	198	–1.4 (3.5)	≤ 0.001b
				t.i.d.	186	–0.4 (3.5)			< 0.05b
	OC population	ADAS-cog – 26 weeks	MC	b.i.d.	180	–0.9 (6.8)	183	2.1 (6.8)	≤ 0.001c
				t.i.d.	173	0.9 (7)			NSc
Mowla et al. (2007)139	OC populationf	MMSE – 12 weeks	A		34	17.4 (3.7)	32	16 (3.7)	0.129f
			MC		34	1.1 (1.4)	32	–0.5 (0.5)	< 0.001g
		WMS-III – 12 weeks	A		34	8.7 (2.2)	32	7.5 (1.4)	0.011f
			MC		34	0.97 (1.7)	32	–0.66 (1.1)	< 0.001g
		CGIC-item 2 (cognitive) – 12 weeks	A		34	3.1 (0.96)	32	3.7 (0.67)	0.005f
Winblad et al. (2007)140	LOCF analysis	ADAS-cog – 16 weeks	MC	10h	248	–0.825 (6.3)a	281	0 (6.71)a	0.09i
				20j	262	–1.39 (6.47)a			< 0.05i
				Oral	253	–0.5 (6.36)a			NSi
		ADAS-cog – 24 weeks	MC	10k	248	–0.6 (6.4)	281	1 (6.8)	0.005i
				20j	262	–1.6 (6.5)			< 0.001i
				Oral	253	–0.6 (6.2)			0.003i
		MMSE – 24 weeks	MC	10i	250	1.1 (3.3)	281	0 (3.5)	0.002k
				20j	262	0.9 (3.4)			< 0.001k
				Oral	256	0.8 (3.2)			0.002k
		Ten-point clock-drawing test – 24 weeks	MC	10i	245	0.3 (3.4)	269	–0.1 (3.2)	0.08k
				20j	251	0.1 (3.1)			0.08k
				Oral	246	0.2 (2.9)			0.15k
		TMT – 24 weeks	MC	10i	241	–12.3 (55.1)	258	7.7 (56.6)	< 0.001i
				20j	238	–6.5 (55.9)			0.005i
				Oral	240	–9.8 (66.1)			< 0.001i

Synthesis with existing evidence base

The data from the new trials were synthesised with those of the previous assessment using random-effects meta-analysis. The measures of ADAS-cog and MMSE were first considered separately and then combined in a pooled multiple outcome measure analysis at ≥ 12 mg/day. We also meta-analysed the data by ≤ 10 mg/day, 4 mg/day and combined doses; results can be found in Appendix 5.

Alzheimer’s Disease Assessment Scale – Cognitive Subscale

The meta-analyses of ADAS-cog scores at 24–26 weeks showed a significant benefit from rivastigmine (≥ 12 mg/day) compared with placebo: WMD = –2.46 (95% CI –3.37 to –1.56), p < 0.001 (Figure 34).

FIGURE 34.

Random-effects meta-analysis: ADAS-cog at 24–26 weeks (mean change from baseline) – rivastigmine (≥ 12 mg/day) vs placebo. a, b.i.d. (twice a day) and t.i.d. (three times a day) arms pooled; b, the 17.4 mg/day patch and 12 mg/day capsules arms pooled.

Mini Mental State Examination

At 24–26 weeks’ follow-up the pooled estimate of effect showed a benefit from rivastigmine: WMD = 1.02 (95% CI 0.63 to 1.41), p < 0.001 (Figure 35).

FIGURE 35.

Random-effects meta-analysis: MMSE at 24–26 weeks (mean change from baseline) – rivastigmine (≥ 12 mg/day) vs placebo. a, b.i.d. (twice a day) and t.i.d. (three times a day) arms pooled; b, the 20-cm2 patch and 12 mg/day capsule arms pooled.

Pooled multiple outcome measures

When we pooled all the results for the cognitive outcomes from the new and existing studies, we found that the overall pooled estimate showed a significant benefit from rivastigmine compared with placebo: SMD = 0.28 (95% CI 0.14 to 0.42), p < 0.001 (Figure 36). The data set used in this meta-analysis can be found in Appendix 6.

FIGURE 36.

Random-effects meta-analysis – cognitive outcomes (SMD) at 24–26 weeks: rivastigmine (all dosages) vs placebo.

With only four data points in this evidence base, it would not be informative to perform metaregression.

Functional

In 2004 Loveman and colleagues2 reported that:

Two published studies reported the PDS as a functional outcome measure. One of these found a statistically significant improvement in participants treated with 6–12 mg/day rivastigmine compared with placebo, and the other reported that a statistically significantly higher percentage of these high dose participants than placebo participants showed an improvement of at least 10%.

New data

Two of the three new studies found since 2004 reported significant functional benefit from rivastigmine compared with placebo. These used PDS and ADCS-ADL as their outcome measures. A summary table of results can be found below in Table 20.

TABLE 20 - Measures of functional ability in included studies: rivastigmine vs placebo

A, absolute value; b.i.d., twice a day; MC, mean change; NS, not statistically significant; t.i.d, three times a day.

Mantel–Haenszel test blocking for centre.

Student’s t-test (two-tailed) (calculated by reviewer).

Post hoc Tukey test (note: t-test p < 0.001).

A 10-cm² rivastigmine patch – equivalent to 9.5 mg/day.

Data extracted from figure.

Two-way analysis of covariance (ANCOVA) (explanatory variables: treatment, country and baseline scores).

A 20-cm² rivastigmine patch – equivalent to 17.4 mg/day.

Study	Subgroup	Outcome	Type	Arm	Rivastigmine		Placebo		p-value
					n	Mean (SD)	n	Mean (SD)
Feldman and Lane (2007)138	ITT population	PDS – 26 weeks	MC	b.i.d.	227	–2.6 (11.1)	221	–4.9 (11.2)	≤ 0.001a
				t.i.d.	225	–1.5 (11.2)	221	–4.9 (11.2)	< 0.05a
	LOCF analysis	PDS – 26 weeks	MC	b.i.d.	207	–1 (11.4)	209	–4.7 (11.3)	≤ 0.001a
				t.i.d.	195	–2.3 (11.5)	209	–4.7 (11.3)	< 0.05a
Mowla et al. (2007)139	OC population	ADL – 12 weeks	A		34	25.3 (6.6)	32	27.1 (6.9)	0.283b
			MC		34	1.2 (2.6)	32	–0.68 (1.3)	0.58c
Winblad et al. (2007)140	LOCF analysis	ADCS-ADL – 16 weeks	MC	10d	247	–0.6 (9.43)e	281	–1.6 (7.96)f	NSg
				20g	263	0.4 (9.73)e	281	–1.6 (7.96)d	< 0.05e
				Oral	254	–0.4 (7.97)e	281	–1.6 (7.96)e	NSf
		ADCS-ADL – 24 weeks	MC	10e	247	–0.1 (9.1)	281	–2.3 (9.4)	0.01g
				20h	263	0 (11.6)	281	–2.3 (9.4)	0.02g
				Oral	254	–0.5 (9.5)	281	–2.3 (9.4)	0.04g

Synthesis with existing evidence base

Data from the existing evidence were synthesised with the new data in a meta-analysis of the PDS scores.

Progressive Deterioration Scale

The overall pooled estimate at 24–26 weeks showed a significant benefit from rivastigmine: WMD = 3.10 (95% CI 1.81 to 4.40), p = 0.001 (Figure 37).

FIGURE 37.

Random-effects meta-analysis – PDS at 24–26 weeks (mean change from baseline): rivastigmine (12 mg/day) vs placebo. a,b.i.d. (twice a day) and t.i.d. (three times a day) arms pooled.

Pooled multiple outcome measures

Two new studies were found to add to this combined meta-analysis of functional outcomes at 24–26 weeks. Again, the overall pooled estimate showed a benefit from rivastigmine compared with placebo: SMD = 0.21 (95% CI 0.12 to 0.29), p < 0.001 (Figure 38). The data set used in this meta-analysis can be found in Appendix 6.

FIGURE 38.

Random-effects meta-analysis: functional outcomes (SMD) at 24–26 weeks: rivastigmine (all dosages) vs placebo.

With only three data points in this evidence base, it would not be informative to perform metaregression.

Global effect

The evidence from the 2004 assessment2 was summarised thus:

Both of the published studies which included CIBIC-plus as a global outcome measure reported a statistically significant improvement in high dose participants (6–12 mg/day) compared with placebo participants. One study also reported a statistically significantly greater proportion of ‘responders’ among participants treated with rivastigmine compared against placebo participants. Another study reported that a greater proportion of high dose rivastigmine participants than placebo participants had a ‘successful’ CIGIC assessment, i.e. scoring one or two on the scale. Two trials found a statistically significant improvement on the Global Deterioration Scale (GDS) measure in participants treated with 6–12 mg/day of rivastigmine compared with placebo participants.

New data

The two new studies138,140 in this comparison that reported global outcomes had conflicting results. Feldman and Lane138 found mostly significantly favourable results with the CIBIC-plus and the Global Deterioration Scale (GDS), whereas Winblad and colleagues’ results were mostly non-significant140 (Table 22).

TABLE 22 - Measures of global effect in included studies: rivastigmine vs placebo

A, absolute value; b.i.d., twice a day; D, dichotomous; MC, mean change; t.i.d, three times a day.

Estimated from figure.

The t-test using pooled error term from analysis of covariance (ANCOVA)/analysis of variance ANOVA (SAS type III analysis).

The t-test using pooled error term from ANOVA (SAS type III).

Mantel–Haenszel test.

A 10-cm² rivastigmine patch – equivalent to 9.5 mg/day.

Data extracted from figure.

Cochran–Mantel–Haenszel van Elteren test using modified ridit scores stratified by country.

A 20-cm² rivastigmine patch – equivalent to 17.4 mg/day.

Chi-squared test (Yates’ correction) (calculated by reviewer).

Study	Subgroup	Outcome	Type	Arm	Rivastigmine		Placebo		p-value
					N	Mean, n (%)	N	Mean, n (%)
Feldman and Lane (2007)138	ITT population	CIBIC-plus score – 18 weeks	A	b.i.d.	215	4.1 (SD 1.03)a	213	4.5 (SD 1.02)a	≤ 0.001b
				t.i.d.	220	3.9 (SD 1.04)a			≤ 0.001b
		CIBIC-plus score – 26 weeks	A	b.i.d.	222	4.1 (SD 1.3)	216	4.5 (SD 1.3)	≤ 0.001c
				t.i.d.	222	3.9 (SD 1.3)			< 0.05c
		CIBIC-plus: any improvement – 12 weeks	D	b.i.d.	220	66 (30.0)a	213	34 (16.0)a	≤ 0.001d
				t.i.d.	215	62 (28.8)a			< 0.05d
		CIBIC-plus: any improvement – 18 weeks	D	b.i.d.	220	68 (30.9)a	213	40 (18.8)a	≤ 0.001d
				t.i.d.	215	47 (21.9)a			NSd
		CIBIC-plus: any improvement – 26 weeks	D	b.i.d.	220	68 (30.9)a	213	40 (18.8)a	< 0.05d
				t.i.d.	215	49 (22.8)a			NSd
		GDS – 26 weeks	MC	b.i.d.	227	0 (SD 0.7)	222	–0.3 (SD 0.7)	< 0.05b
				t.i.d.	229	–0.2 (SD 0.7)			NSb
	LOCF analysis	CIBIC-plus score – 26 weeks	A	b.i.d.	206	3.9 (SD 1.2)	205	4.5 (SD 1.2)	≤ 0.001c
				t.i.d.	198	4.1 (SD 1.2)			< 0.05c
		GDS – 26 weeks	MC	b.i.d.	195	0 (SD 0.7)	202	–0.3 (SD 0.7)	< 0.05b
				t.i.d.	188	–0.1 (SD 0.7)			NSb
	OC population	CIBIC-plus score – 26 weeks	A	b.i.d.	177	3.9 (SD 1.2)	179	4.4 (SD 1.2)	≤ 0.001c
				t.i.d.	167	4.1 (SD 1.2)			< 0.05c
Winblad et al. (2007)140	LOCF analysis	ADCS-CGIC: score – 16 weeks	A	10e	248	3.9 (SD 1.14)f	278	4.35 (SD 1.25)f	NSg
				20h	260	3.93 (SD 1.17)f			NSg
				Oral	253	4.25 (SD 1.11)f			NSg
		ADCS-CGIC: score – 24 weeks	A	10e	248	3.9 (SD 1.2)	278	4.2 (SD 1.3)	0.01g
				20h	260	4 (SD 1.3)			0.054g
				Oral	253	3.9 (SD 1.3)			0.009g
		ADCS-CGIC: markedly improved – 24 weeks	D	10e	248	5 (2.0)	278	2 (0.7)	0.361i
				20h	260	5 (1.9)			0.395i
				Oral	253	3 (1.2)			0.916i
		ADCS-CGIC: moderately improved – 24 weeks	D	10e	248	29 (11.7)	278	26 (9.4)	0.463i
				20h	260	32 (12.3)			0.334i
				Oral	253	29 (11.5)			0.513i
		ADCS-CGIC: minimally improved – 24 weeks	D	10e	248	43 (17.3)	278	50 (18.0)	0.937i
				20h	260	48 (18.5)			0.975i
				Oral	253	60 (23.7)			0.129i
		ADCS-CGIC: unchanged – 24 weeks	D	10e	248	105 (42.3)	278	91 (32.7)	0.029i
				20h	260	94 (36.2)			0.457i
				Oral	253	96 (37.9)			0.244i
		ADCS-CGIC: minimally worse – 24 weeks	D	10e	248	41 (16.5)	278	65 (23.4)	0.065i
				20h	260	50 (19.2)			0.285i
				Oral	253	30 (11.9)			< 0.001i
		ADCS-CGIC: moderately worse – 24 weeks	D	10e	248	22 (8.9)	278	36 (12.9)	0.177i
				20h	260	27 (10.4)			0.429i
				Oral	253	30 (11.9)			0.803i
		ADCS-CGIC: markedly worse – 24 weeks	D	10e	248	3 (1.2)	278	8 (2.9)	0.303i
				20h	260	4 (1.5)			0.448i
				Oral	253	5 (2.0)			0.696i

Synthesis with existing evidence base

Data from the new studies were pooled with the existing evidence in random-effects meta-analyses using the CIBIC-plus at 26 weeks and the GDS at 26 weeks. The results can be seen in Figures 39 and 40.

FIGURE 39.

Random-effects meta-analysis – CIBIC-plus at 26 weeks: rivastigmine (12 mg/day) vs placebo. a, b.i.d. (twice a day) and t.i.d. (three times a day) arms pooled.

FIGURE 40.

Random-effects meta-analysis – GDS at 26 weeks (mean change from baseline): rivastigmine (12 mg/day) vs placebo. a, b.i.d. (twice a day) and t.i.d. (three times a day) arms pooled.

Clinician’s Interview-based Impression of Change

The meta-analysis showed a significant benefit from rivastigmine at 26 weeks: WMD = –0.42 (95% CI –0.55 to –0.29), p < 0.001.

Global Deterioration Scale

This meta-analysis also showed a significant benefit from rivastigmine at 26 weeks: WMD = 0.20 (95% CI 0.12 to 0.27), p < 0.001.

Pooled multiple outcome measures

We then pooled the results from both outcomes; the results from this can be seen in Figure 41 and showed an overall pooled estimate of SMD = 0.23 (95% CI 0.16 to 0.31), p < 0.001. The data set that was used in this meta-analysis can be found in Appendix 6.

FIGURE 41.

Random-effects meta-analysis – global outcomes (SMD) at 24–26 weeks: rivastigmine (all dosages) vs placebo.

With only four data points in this evidence base, it would not be informative to perform metaregression.

Summary: rivastigmine versus placebo

Our update searches identified three new RCTs138–140 to add to the four134–137 included in the previous review. 2

All three studies138–140 showed benefits from rivastigmine on the ADAS-cog and MMSE, although these benefits were dependent on dose, with greater benefits seen at 12 mg/day than at 6 mg/day. When these data were pooled with the existing evidence, significant differences favouring rivastigmine continued to be seen on the ADAS-cog at 24–26 weeks (≥ 12 mg/day), WMD = –2.46 (95% CI –3.37 to –1.56), p < 0.001. However, the benefits from rivastigmine were not apparent on MMSE scores until 24–26 weeks’ follow-up [WMD = 1.02 (95% CI 0.63 to 1.41), p < 0.001]; this may be due to the MMSE’s difficulties with detecting change. When the outcomes from both cognitive measures were combined they continued to show an advantage from taking rivastigmine on cognitive outcomes.

Two of the three new studies reporting functional outcomes showed significant gains for these measures. When these new data were synthesised with existing evidence using the PDS, significant gains were shown at 24–26 weeks [WMD = 3.10 (95% CI 1.81 to 4.40), p = 0.001].

The data on behavioural outcomes from the new studies were unclear, with the smaller study139 showing a benefit from rivastigmine that the larger one140 did not. The existing evidence was too heterogeneous for meta-analysis, so the overall effectiveness of rivastigmine for behavioural outcomes is unknown.

The results for global outcomes were also mixed. Results from the CIBIC-plus were almost universally significant, whereas those measured by the Alzheimer’s Disease Cooperative Study–Clinical Global Impression of Change (ADCS-CGIC) were almost universally not; those using the GDS showed no significant gain from rivastigmine. However, when these data were pooled with the existing evidence, the overall estimates favoured rivastigmine on the CIBIC-plus [WMD = –0.42 (95% CI –0.55 to –0.29), p < 0.001] and the GDS [WMD = 0.20 (95% CI 0.12 to 0.27), p < 0.001]. When results from both these outcome measures were combined, the result continued to show significant benefit from rivastigmine, these results are based on a robust ITT population.

When rivastigmine patches were compared with capsules, the results showed that the 9.5 mg/day patch was similarly effective as the 12.5 mg/day capsule, but with fewer side effects.

None of the included studies in either the updated or the original review reported QoL outcomes.

As in the other AChEIs, the main AEs were gastrointestinal.

Overall, pooled estimates of cognitive benefits from rivastigmine were favourable, but were shown to be dose dependent as in the previous review. The results from functional and global outcomes also showed significant gains. However, the results from individual trials of behavioural outcomes were mixed (pooling was not possible owing to heterogeneity). The lower dose transdermal patch (9.5 mg/day) was shown to be as effective as the capsule (12 mg/day), but with fewer side effects.

Graphical summary of rivastigmine versus placebo

The graphical summary in Figure 42 shows that two large studies138,140 and one small study139 have been added to the evidence for rivastigmine since 2004. The figure illustrates how the new studies have added to the precision of our knowledge of the effects of rivastigmine in AD. Previously, the results for cognitive outcomes were ambiguous; however, the results from the new trials all show cognitive benefits. The smaller new study139 showed a gain on behavioural outcomes that had not been seen in the previous studies, but the results for functional and global outcomes continued to be mixed.

FIGURE 42.

Summary of all included studies in the 2004 and 2010 reviews: rivastigmine vs placebo. *Significant only for ’TMT’, not ’ten-point clock-drawing test’. Analy, analysis; be, behavioural; b.i.d., twice a day; Blind, blinding; Char, characteristics of patients; cog, cognitive; F, female; func, functional; glo, global; M, male; Rand, randomisation; SIB, Severe Impairment Battery; t.i.d., three times a day; TMT, trail-making test.

Differing approaches to pooling data

A key difference to our memantine findings and Lundbeck’s lies in our differing approaches to pooling data. In particular, Lundbeck have pooled together memantine plus AChEI versus placebo plus AChEI trials with memantine monotherapy versus placebo studies; we were not comfortable with this approach owing to the heterogeneity of the data. Nevertheless, we have followed this approach for completeness and present the results in Appendix 14. The effect of pooling data in this way is to show a more favourable response to memantine.

Identified evidence

The 2004 review2 lists two RCTs141,142 as investigations into the effectiveness of memantine in AD. However, one of those studies141 addressed the effectiveness of memantine in combination with donepezil; accordingly, this study is considered as part of our assessment of combination therapy (see Summary: head-to-head comparisons).

We identified one additional RCT143 of relevance to this comparison, details of which are presented in Table 24. This study’s interventions, comparators and baseline characteristics can be seen in Table 25 and the markers of internal validity in Table 26.

Evidence of clinical effectiveness

Cognition

The 2004 report by Loveman and colleagues2 found two studies of memantine; they summarised their results as follows:

Both studies used the SIB as a measure of cognitive outcome. Statistically significant differences in favour of the use of memantine over placebo were apparent in the two studies. MMSE scores deteriorated in both the memantine group and the placebo group and the degree of deterioration was not statistically significantly different between the two groups.

New data

The data from the new trial143 showed a significant effect only from memantine on one of six analyses. However, this was in an OCs-only analysis that may have biased the results; Table 27 summarises the results.

TABLE 27 - Measures of cognition in included studies: memantine vs placebo

MC, mean change.

Analysis of covariance (ANCOVA) (treatment group and centre as main effects; baseline score as covariate).

Estimated from figure.

Study	Subgroup	Outcome	Type	Memantine		Placebo		p-value
				n	Mean (SD)	n	Mean (SD)
Van Dyck et al. (2007)143	LOCF analysis	SIB – 24 weeks	MC	170	–2 (13)	165	–2.5 (12.8)	0.616a
	OC population	SIB – 4 weeks	MC	167	0.875 (7.43)b	164	–0.3 (6.4)b	0.146a
		SIB – 8 weeks		158	2.08 (7.86)b	155	0.375 (7.16)b	0.064a
		SIB – 12 weeks		146	1.65 (9.06)b	150	–0.825 (8.27)b	0.008a
		SIB – 18 weeks		140	0 (8.28)b	139	–2.12 (9.14)b	0.065a
		SIB – 24 weeks		131	–1.8 (12.6)	126	–2.4 (13.5)	0.617a

Synthesis with existing evidence base

Severe Impairment Battery

The data from the new trial were pooled with that of the existing studies in random-effects meta-analyses of the Severe Impairment Battery (SIB) at 12 weeks and 24–28 weeks. The results showed a significant effect at 12 weeks [WMD = 4.15 (95% CI 0.52 to 7.78), p = 0.025], but not at 24–28 weeks (Figures 43 and 44).

FIGURE 43.

Random-effects meta-analysis – SIB at 12 weeks (mean change from baseline): memantine vs placebo.

FIGURE 44.

Random-effects meta-analysis – SIB at 24–28 weeks (mean change from baseline): memantine vs placebo.

Functional

The previous assessment report2 summarised the findings about the effects of memantine on functional outcomes as:

Both studies demonstrated that memantine appears to show a statistically significant benefit to participants on the ADCS-ADL when compared to placebo, with a reduction in the level of deterioration.

New data

The results from the new study showed no significant benefit on functional outcomes for memantine compared with placebo (Table 28).

TABLE 28 - Measures of functional ability in included studies: memantine vs placebo

FAST, Functional Assessment Staging Tool; MC, mean change.

Analysis of covariance (ANCOVA) (treatment group and centre as main effects; baseline score as covariate).

Estimated from figure.

Study	Subgroup	Outcome	Type	Memantine		Placebo		p-value
				n	Mean (SD)	n	Mean (SD)
Van Dyck et al. (2007)143	LOCF analysis	ADCS-ADL – 24 weeks	MC	171	–2 (7.85)	165	–2.7 (7.71)	0.282a
		FAST – 24 weeks		151	0.3 (1.23)	141	0.6 (1.19)	0.093a
	OC population	ADCS-ADL – 4 weeks	MC	168	0.312 (4.37)b	164	0.512 (4)b	0.801a
		ADCS-ADL – 8 weeks		159	–0.0875 (5.2)b	156	–0.188 (4.84)b	0.665a
		ADCS-ADL – 12 weeks		147	0 (5.46)b	150	–0.488 (5.05)b	0.155a
		ADCS-ADL – 18 weeks		142	–0.688 (7.3)b	140	–1.38 (5.62)b	0.357a
		ADCS-ADL – 24 weeks		133	–1.3 (6.92)	127	–2.3 (6.76)	0.188a
		FAST – 24 weeks		133	0.3 (1.15)	127	0.6 (1.13)	0.074a

Synthesis with existing evidence base

The data from the new studies were synthesised with the existing evidence in a random-effects meta-analysis.

Alzheimer’s Disease Cooperative Study – Activities of Daily Living Index

Two studies provide data on functional effect as measured by ADCS-ADL; both report the modified ADCS-ADL₁₉ version of the instrument, consisting of 19 items that have been individually validated in cases of more severe dementia. The data were meta-analysed at 12 weeks’ and 24–28 weeks’ follow-up. The results were not significant at 12 weeks and barely significant at 24–28 weeks, especially considering that the population analysed were LOCF: WMD = 1.41 (95% CI 0.04 to 2.78), p = 0.044 (Figures 45 and 46).

FIGURE 45.

Random-effects meta-analysis – ADCS-ADL₁₉ at 12 weeks (mean change from baseline): memantine vs placebo.

FIGURE 46.

Random-effects meta-analysis – ADCS-ADL₁₉ at 24–28 weeks (mean change from baseline): memantine vs placebo.

Functional Assessment Staging Tool

Another meta-analysis was conducted with data from existing and new studies using the Functional Assessment Staging Tool (FAST) at 24–28 weeks’ follow-up. The overall pooled estimate showed a significant benefit from memantine compared with placebo, WMD = –0.34 (95% CI –0.55 to –0.13), p = 0.002 (Figure 47).

FIGURE 47.

Random-effects meta-analysis – FAST at 24–28 weeks (mean change from baseline): memantine vs placebo.

Behavioural and mood

The 2004 assessment report2 summarised the finding for behavioural outcomes comparing memantine with placebo as:

It appears that participants receiving memantine and already receiving a steady dose of donepezil have a statistically significantly lower NPI score than placebo. Those on memantine only however, showed no statistically significant difference compared to placebo.

New data

The study143 that was published after 2004 measured behavioural outcomes using the NPI and the Behavioural Rating Scale for Geriatric Patients (BGP). Neither measure showed a significant benefit from memantine (Table 29).

TABLE 29 - Measures of behavioural effect and mood in included studies: memantine vs placebo

MC, mean change.

Analysis of covariance (ANCOVA) (treatment group and centre as main effects; baseline score as covariate).

Study	Subgroup	Outcome	Type	Memantine		Placebo		p-value
				n	Mean (SD)	n	Mean (SD)
Van Dyck et al. (2007)143	LOCF analysis	NPI – 24 weeks	MC	161	1 (16.5)	154	1.1 (17.4)	0.963a
		BGP: total – 24 weeks		151	0.6 (6.14)	141	1.5 (7.12)	0.197a
		BGP: care dependency – 24 weeks		151	0.5 (4.92)	141	1.4 (4.75)	0.076a
	OC population	NPI – 24 weeks		133	0.5 (15)	127	1 (15.8)	0.782a
		BGP: total – 24 weeks		133	0.4 (6.92)	127	1.1 (6.76)	0.312a
		BGP: care dependency – 24 weeks		133	0.4 (4.61)	127	1.2 (5.63)	0.138a

Synthesis with existing evidence base

The NPI data from van Dyck and colleagues143 were pooled with the existing data at 24–28 weeks’ follow-up in a random-effects meta-analysis. This analysis also failed to show a significant gain from memantine compared with placebo (Figure 48).

FIGURE 48.

Random-effects meta-analysis – NPI at 24–28 weeks (mean change from baseline): memantine vs placebo.

Global effect

In 2004, Loveman and colleagues2 summarised the results for global measures comparing memantine with placebo as:

Both studies used the CIBIC-plus as a measure of global outcome, and, in both cases, memantine appeared to be effective.

New data

Van Dyck and colleagues143 also measured global outcomes with the CIBIC-plus; however, the differences they found were not significant (Table 30).

TABLE 30 - Measures of global effect in included studies: memantine vs placebo

C, continuous.

Cochran–Mantel–Haenszel statistic using modified ridit scores (van Elteren test) controlling for study centre.

Study	Subgroup	Outcome	Type	Memantine		Placebo		p-value
				n	Mean (SD)	n	Mean (SD)
Van Dyck et al. (2007)143	LOCF analysis	CIBIC-plus score – 24 weeks	C	171	4.3 (13.1)	163	4.6 (12.8)	0.182a
	OC population	CIBIC-plus score – 24 weeks	C	134	4.3 (12.7)	127	4.6 (11.3)	0.089a

Synthesis with existing evidence base

Clinician’s Interview-based Impression of Change

When the new data were pooled with the existing studies in a random-effects meta-analysis the overall pooled estimate showed a significant beneficial effect from memantine compared with placebo: WMD = –0.30 (95% CI –0.47 to –0.13), p < 0.001 (Figure 49).

FIGURE 49.

Random-effects meta-analysis – CIBIC-plus at 24–28 weeks: memantine vs placebo.

Summary: memantine versus placebo

One new moderate-to-poor-quality study was found to add to the existing evidence for memantine versus placebo.

The pooled results for cognitive abilities measured by the SIB showed a significant benefit from memantine at 12 weeks’ follow-up [WMD = 4.15 (95% CI 0.52 to 7.78), p = 0.025]. However, at 24 weeks the data pooled with that of the previous review showed no significant benefit.

Similar to the previous review,2 the new study143 found a significant benefit from memantine from the FAST functional outcome, although not with the ADCS-ADL at 12 weeks. When the FAST data from new and existing studies were pooled, a significant relationship was found between memantine and an improvement in scores [WMD = –0.34 (95% CI –0.55 to –0.13), p = 0.002]. A marginally significant benefit was seen from memantine when pooled ADCS-ADL data were measured at 24–28 weeks [WMD = 1.41 (95% CI 0.04 to 2.78) p = 0.044].

The results from behavioural outcomes in the new study,141 similar to the previous review,2 failed to show a significant benefit from memantine, either singly or when the data were pooled.

Although the results of the CIBIC-plus failed to show a significant gain from memantine, when these data were pooled with that from the previous review a significant effect was found [WMD = –3.00 (95% CI –0.471 to –0.129), p < 0.001].

No studies reported QoL outcomes. The main AEs were agitation and hypertension.

The meta-analysis of memantine versus placebo studies showed benefit from memantine at 12 weeks’ follow-up on the SIB. However, treatment gain, measured by functional outcome, depended on the type of instrument used, and no benefit was seen from behavioural outcomes. Nevertheless, pooled estimates of global outcomes showed a benefit from taking memantine at 24–28 weeks. Overall, the pooled results from these moderate-to-poor-quality studies showed inconclusive results for cognitive and behavioural outcomes. The results for functional outcomes were dependent on the measure used, but the pooled results of the new and existing evidence for global outcomes showed significant benefit from using memantine.

Graphical summary of memantine versus placebo

Figure 50, below, illustrates how little the evidence has changed for memantine versus placebo. The cognitive benefits found in 2004 failed to be replicated; indeed the new study143 favoured memantine on only one outcome measure. However, the quality of the new143 and existing study142 was not high; thus, these results cannot be considered conclusive.

FIGURE 50.

Summary of all included studies in the 2004 and current reviews: memantine vs placebo. Analy, analysis; be, behavioural; Blind, blinding; Char, characteristics of patients; cog, cognitive; F, female; func, functional; glo, global; M, male; Rand, randomisation.

Identified evidence

Alongside placebo-controlled trials, a certain amount of randomised evidence provides direct, head-to-head comparisons of two or more of the technologies under review. Three such RCTs were included in the 2004 review:2 donepezil versus rivastigmine,144 donepezil versus rivastigmine,145 and donepezil versus galantamine. 146

Our searches identified a further four RCTs of this type. 99,147–149 Details of the design of these trials are tabulated in Table 32, and a summary of treatments and baseline characteristics of participants can be found in Table 33. Two of the new RCTs99,147 compared all three AChEIs (although Nordberg and colleagues’ trial147 is of relevance to the current review only for its safety data). One trial investigated donepezil versus rivastigmine148 and the last was concerned with donepezil versus galantamine. 149