Lack of Evidence for the Efficacy of Memantine in Mild Alzheimer Disease

Memantine, indicated for moderate to severe Alzheimer disease (AD), is frequently prescribed off-label either alone or with a cholinesterase inhibitor for mild AD and mild cognitive impairment.¹ In 2006, it was prescribed to 19% of patients with mild AD in the United States.² In the mainly academic centers that comprise the Alzheimer Disease Neuroimaging Initiative, 45.7% of patients with mild AD were receiving memantine,³ and 25.0% of patients with mild AD who were registered in 2009 in the National Institutes of Health National Alzheimer Coordinating Center were receiving it.³ In mild to moderate AD clinical trials conducted from 2003 to 2009, 13.5% to 63.4% of patients received memantine.⁴ Similarly, 11.4% and 11.1% of patients with mild cognitive impairment who were in the Alzheimer Disease Neuroimaging Initiative and the National Alzheimer Coordinating Center, respectively, were receiving memantine.⁵ Nearly 40% of US neurologists surveyed reported prescribing memantine at least sometimes to patients with mild cognitive impairment.⁶

The European Medicines Agency approved memantine in May 2002 for use in “moderately severe to severe” AD, and the US Food and Drug Administration (FDA) approved it in October 2003 for “moderate to severe dementia of the Alzheimer type” (Table 1). The “moderate to severe” FDA indication is based on 2 clinical trials that included patients with AD on the basis of their receiving a Mini-Mental State Examination (MMSE)⁷ score of 14 or less as described in the “clinical studies” section of the prescribing information.⁸

In July 2005, the FDA did not approve a supplemental new drug application by the manufacturer to expand marketing approval for memantine from “moderate to severe” to “mild to severe” AD.^{9 - 10} The supplemental new drug application included 3 placebo-controlled clinical trials that enrolled patients with mild to moderate AD, defined on the basis of the patient having an MMSE score of 10 to 23.^{11 - 13} Two of the 3 trials did not show statistical significance on the primary end points. Because the trials included both patients with mild and moderate AD, specific evidence for memantine's efficacy in the mild AD subgroup is lacking.

In November 2005, the European Medicines Agency expanded its indication to “moderate to severe” AD.^{14 - 16} This expanded indication was based on pooled analyses of “moderate” subsets of patients with MMSE scores of 10 to 19; these pooled analyses were derived from the same 3 clinical trials of mild to moderate AD reviewed by the FDA.¹⁴

Subsequently, similar meta-analyses of memantine were published by the US manufacturer¹⁷ and the European manufacturer.¹⁴ Each included the same 6 trials, comprising the 3 trials of mild to moderate AD already mentioned and 3 other trials of moderate to severe AD. The European meta-analysis¹⁴ specifically excluded patients with mild AD who had MMSE scores of 20 to 23 and claimed only “clinically relevant efficacy in patients with moderate to severe AD.” Notwithstanding the FDA's prior rejection of a mild AD indication, the US authors claimed that memantine was beneficial for patients with AD “across the spectrum of the disease, from mild to severe cases.”¹⁷ These 2 meta-analyses, one a subset of the other and using the same data sets, allowed for the opportunity to directly assess the effects of memantine in mild AD and in moderate AD.

We sought all clinical trials of memantine vs placebo that included patients with mild AD. The 2 company-sponsored meta-analyses^{14 ,17} formed the basis for the search. We relied on the formal searches for memantine clinical trials of AD that we conducted for the comprehensive Cochrane Collaboration systematic reviews of memantine for dementia¹⁸ (R.M., unpublished data, 2011) to search for trials that included patients with mild AD. In particular, manufacturer-sponsored meta-analyses, registries, presentations, and publications were searched for randomized, placebo-controlled, parallel-group clinical trials of memantine in patients with mild to moderate AD. We sought additional information on the 3 mild to moderate AD trials included in the company-sponsored meta-analyses by searching corresponding publications, conference posters, company-sponsored Web sites, publications, and clinical trials registries and by querying the companies for additional data.

We extracted clinical and demographic characteristics and outcomes from the trial reports, including the Alzheimer Disease Assessment Scale–cognitive subscale (ADAS-cog),¹⁹ the Clinician's Interview-Based Impression of Change plus caregiver's input (CIBIC-plus),²⁰ the Alzheimer Disease Cooperative Study–activities of daily living (ADCS-ADL) scale,²¹ and the Neuropsychiatric Inventory (NPI),²² assessing cognition, global change, functional activities, and behavior, respectively.

Effects were calculated for each memantine-placebo contrast in each of the 3 trials, combined using inverse variance–weighted averages²² and expressed in meta-analytic summaries as mean differences with 95% confidence intervals (CIs) and P values using the Review Manager version 5.0 software.²³ We used the fixed-effects model that had been used in both company-sponsored meta-analyses²³ ; χ² tests and the I² statistics derived from the test statistics were used to establish the presence of heterogeneity or inconsistency among the effects,²⁴ with P ≤ .15 or I²≥50% taken as criteria for heterogeneity of effect sizes.

Summary statistics (sample sizes, means, and standard deviations in each treatment group) were available from observed cases of patients with mild AD and moderate AD combined (ie, with baseline MMSE scores of 10-23 from the US meta-analysis) and for the subset of patients with moderate AD (ie, with MMSE scores of 10-19 from the European meta-analysis¹⁴ ). Sample sizes for the subset of patients with mild AD (ie, with MMSE scores of 20-23) were obtained by subtraction, and sample means were obtained by recognizing the sample mean of the combined set as a sample size–weighted average of the sample means in the 2 subsets. Standard deviations for the subset of patients with mild AD were obtained by subtracting the sum of squared observations for the subset of patients with moderate AD from that for the combined set.²⁵ Sums of squares were computed from the standard deviations and means. Specifically, within each treatment group, the mean

and standard deviation (s_mild) of the subset of patients with mild AD were computed as

and

respectively, where n represents the sample size, the subscript “comb” refers to the combined set, and the subscript “mod” refers to the subset of patients with moderate AD.²⁵

We used observed cases because they were used in the European meta-analysis and preferred in the European Public Assessment Report,¹⁶ and because the samples of the last observation carried forward (LOCF) from one meta-analysis^{14 ,16} were not available.

A total of 16 clinical trials of 3-month duration or longer were identified, of which 6 were included in the industry-sponsored meta-analyses. Seven trials potentially included patients with mild AD. Four trials were excluded: No results have been published or made available from the 6-month Daiichi Sankyo trial, MA 3301 (listed as Japan Pharmaceutical Information Center clinical trials information no. 050079),²⁶ or from the 12-week Lundbeck trial 99817 of 47 patients conducted in Taiwan (e-mail communication, H/S Lundbeck, June 2006). Results from a 1-year, Merz-sponsored placebo-controlled trial of 37 patients with mild to moderate AD²⁷ (with MMSE scores of 15-21) were not available separately for the very few patients with MMSE scores greater than 19, nor were results of the 1-year, Lundbeck-sponsored trial 10112,²⁸ which was a placebo-controlled trial of 278 patients who mostly had moderate AD (with MMSE scores of 12-20). Both of these trials were brain-imaging trials that did not present clinical outcomes except for the MMSE scores in the latter, which showed no significant differences. We found 6 meta-analyses that included patients with mild to moderate AD.^{14 ,16 - 18 ,29 - 30} We included three 6-month trials that included patients with mild to moderate AD.^{11 - 13} These trials and 3 Web postings of the synopses,^{31 - 33} 2 meta-analyses,^{14 ,17} a European expert report,¹⁶ and Forest Laboratories provided data for the analyses.

The inclusion criteria for these trials were very similar with respect to AD diagnostic criteria, outpatient status, and minimum age (Table 2). MMSE inclusion criteria were scores of 10 to 22 for the 2 US trials and scores of 11 to 23 for the European trial. The overall percentage of female patients was 58%. One US trial, MD-12, allowed the concurrent use of cholinesterase inhibitors.¹² All 3 trials included random allocation to treatment, patients and investigators blinded to treatments, and outcome assessments performed while blinded to treatment assignment.

The number of randomly assigned patients who had 1 outcome assessed (ie, the LOCF samples of patients) was 1282, and 1128 of these patients (88.0%) completed the trials and constituted the observed cases. The number of patients with mild AD in the observed cases was 431, which is 38.2% of the total.

There was no evidence for the efficacy of memantine in the subset of patients with mild AD on any of the outcomes in any trial or when statistically combined (Figure 1). There were no significant effects on the ADAS-cog (−0.17 [95% CI, −1.60 to 1.26]; P = .82), the CIBIC-plus (−0.09 [95% CI, −0.30 to 0.12]; P = .40), the ADCS-ADL scale (0.62 [95% CI, −1.64 to 2.71]; P = .56), or the NPI (0.09 [95% CI, −2.11 to 2.29]; P = .94).

For the subset of patients with moderate AD (Figure 2), there was no significant difference between memantine and placebo on the ADAS-cog in any individual trial, although there was a significant effect when the 3 trials were statistically combined (−1.33 [95% CI, −2.28 to −0.38]; P = .006). There was a significant effect on the CIBIC-plus in 1 of the 3 trials and when combined (−0.16 [95% CI, −0.32 to 0.00]; P = .05). There was no significant effect on the ADCS-ADL scale (0.57 [95% CI, −1.75 to 0.60]; P = .34) or the NPI (0.25 [95% CI, −1.48 to 1.99]; P = .77) in any of the individual trials or when the trials were combined.

There was no evidence for heterogeneity or inconsistency for any of the outcomes in either the mild AD or moderate AD groups (Figures 1 and 2). Overall, there were no differences in all-cause discontinuation between memantine and placebo in the patients with mild to moderate AD (14.7% vs 13.0%). A subset of patients with mild AD was not available.

The authors of the US company–sponsored meta-analysis concluded that memantine is effective overall in mild to severe AD, although they acknowledged small effect sizes.¹⁷ The European company–sponsored meta-analysis was conducted specifically for the European regulatory review process to support an expanded indication from “moderately severe” to “moderate” AD that explicitly meant expanding top MMSE scores from 14 to 19. To this end, they removed the patients with mild AD from the analyses and obtained somewhat larger effect sizes than did the US meta-analysis.

These 2 meta-analyses, one a subset of the other and using the same data sets, gave us the opportunity to directly assess the effects of memantine in mild AD and in moderate AD separately. Contrary to the conclusions of the US meta-analysis, we found no effects for memantine on any of the 4 outcomes in mild AD and no significant effects for memantine on ADLs or behavior in the subset of patients with moderate AD. No individual comparison in either the mild or moderate subset of patients on any of the 4 outcomes was significant, except for the CIBIC-plus in the MD-10 trial (Figures 1 and 2). Moreover, the differences between memantine and placebo on the ADAS-cog and the CIBIC-plus in the subset of patients with moderate AD (comprising 697 patients), although statistically significant, were very small, −1.33 and −0.16, respectively, about half the effect sizes from the many trials of cholinesterase inhibitors.^{18 ,34 - 37}

The European meta-analysis excluded patients with MMSE scores greater than 19 and combined the subsets of patients with moderate AD (ie, with MMSE scores of 10-19) from these trials with the subsets of patients with moderate AD from 3 moderate to severe AD trials (ie, with MMSE scores of 10-14)^{38 - 40} in order to persuade European regulators that the drug should be indicated for a broader definition of moderate AD than previously approved. The subsets of patients with moderate AD from these trials not only used differently defined patients but also a different cognitive outcome (ie, the Severe Impairment Battery).⁴¹ The Severe Impairment Battery and ADAS-cog scores had to be standardized or transformed to z scores in order to be statistically combined.¹⁴ The results from this procedure were endorsed by the European Medicines Agency and European public experts¹⁶ and were included in memantine's Summary of Product Characteristics,¹⁵ thus expanding memantine's indication from “moderately severe” to the broader “moderate” AD range. In effect, the European meta-analysis was a procedure to gain marketing approval for the treatment of patients with MMSE scores of 15 to 19; otherwise, memantine would have continued to be constrained for or indicated only for the treatment of patients with MMSE scores of 14 or less (as it is in the US product labeling).⁸

A limitation of meta-analyses based on individual trials is that individual patient responses cannot be assessed. Thus, the authors of the US meta-analysis¹⁷ incorrectly claimed that memantine “demonstrated consistent benefits compared with placebo for cognitive measures, global clinical impression, and functional outcomes across the overall spectrum of AD severity,” and they further elaborated that their results “support the hypothesis that drugs indicated for AD are beneficial for all patients with clinically diagnosed disease, and that arbitrary divisions into severity groups might not be necessary for treatment or drug development in this field.” Our results, although not analyzed at the patient level, do not support these assertions, and they provide insight into why the FDA did not approve the manufacturer's application to market the drug for mild AD.⁹

The lack of evidence for the efficacy of memantine could come to light only because the European meta-analysis was a subset of the US meta-analysis. If the European meta-analysis had not excluded patients with higher MMSE scores, then it would not have been possible to distinguish the particular lack of evidence for the efficacy of memantine in the mild AD group, which comprised nearly 40% of the patients overall (ie, 431 patients), about the size of typical AD efficacy trials. Given the very small effects in the subset of patients with moderate AD (ie, with MMSE scores of 10-19), it is uncertain whether clinical effects would have been detected in the top half of the subset of moderate patients (ie, those with MMSE scores of 15-19), the group outside US labeling but within European Union labeling. The data released so far do not allow for this determination.

We limited our analyses to observed cases because (1) the European meta-analysis assessed only these samples, (2) this was preferred in the European Public Assessment Report,¹⁶ and (3) the LOCF samples from that report were unavailable to us. It is very unlikely, however, that a slightly larger LOCF sample, approximating an intent-to-treat sample and comprising only about 12% more patients, would have yielded significant results in the subset of patients with mild AD because (1) there were no trends for efficacy whatsoever, (2) the discontinuation rate in each treatment arm was very similar, and (3) effect sizes in LOCF samples tend to be smaller.

Each memantine clinical trial was designed to be analyzed as a whole and not planned to assess subgroups. Post hoc and subgroup analyses from individual trials are generally underpowered and unreliable. Meta-analyses gain statistical power by combining clinical trials that individually were not significant on most of the outcomes. However, such meta-analyses may obscure important effects owing to differences among patients within the trials.^{42 - 43}

There were differences in the direction of change in the placebo groups between subsets of patients with mild AD and moderate AD. In the moderate subset, the cognitive, global, and functional scores tended to worsen over 6 months. By comparison, the cognitive measure in the mild subset tended to improve, and the global scores did not change. Thus, the analyses for the patients with mild AD may have been underpowered to detect cognitive change, or the outcomes were not sufficiently sensitive for relatively short 6-month trials. The effects of a drug that might maintain, but not improve, function would be more difficult to determine earlier in the course of illness when little cognitive or functional deterioration can be expected. On the other hand, these outcomes are used in nearly all mild to moderate AD trials, and it is also likely that, as in trials of cholinesterase inhibitors, the larger clinical ratings effects are seen in the more moderately impaired patients compared with the more mildly impaired.^{44 - 46} One cholinesterase inhibitor trial that selected patients with mild AD who had MMSE scores of 21 to 26 showed efficacy with the drug-treated group improving and the placebo group not declining, the latter consistent with our observations from the present analysis.⁴⁷

The trials were designed to show symptomatic effects, and it is possible that memantine may have longer-term or disease-modifying effects that become evident in more severe and rapidly deteriorating patients, or it may have cognitive effects mainly in patients with behavioral disturbances. The evidence for this, however, is also lacking because no properly designed, longer-term trials have been undertaken. Neither of the two 52-week placebo-controlled trials of mainly moderate AD (37 patients with MMSE scores of 15-21 in one trial²⁷ and 278 patients with MMSE scores of 12-20 in the other^{28 ,48} ) reported significant improvements on the ADAS-cog, the MMSE, or the NPI with memantine. It is evident that the statistical effect on the behavioral rating scale is greater in the moderate to severe AD trials^{38 - 40} than the null effects are on the behavioral rating scale in the mild to moderate AD trials. To further assess this, there are 2 ongoing trials of patients with moderate to severe AD who have behavioral disturbances: the Memantine for Agitation in Dementia trial⁴⁹ and Lundbeck trial 10158.⁵⁰

The results provide another argument that results of randomized clinical trials of marketed drugs should be made available soon after trials are completed. Timely release combined with the willingness to openly examine subsets of patients and individual patient outcomes would have allowed earlier recognition of some limitations in memantine's broader efficacy (Table 1). As already mentioned, in the United States, nearly half of the patients with mild AD and a substantial proportion of patients with mild cognitive impairment are receiving memantine despite a lack of evidence that the drug is helpful and some evidence that it is not.⁵¹

In summary, we describe the evidence base and lack of evidence for memantine in mild AD. Results may explain the FDA decision not to approve memantine for mild AD and to restrict labeling to MMSE scores of 14 or less. In view of the very small effects on cognition and global change in the subset of patients with moderate AD, remaining questions include whether clinical effects of memantine are observable in patients with MMSE scores of 15 to 19, which are above the moderate level indicated in the United States but within the European Union's range, and whether or not memantine's efficacy in clinical trials is in fact confined to more severe AD. These results, however, do not substitute for the need for adequate and well-controlled trials specifically assessing memantine's short- and long-term efficacy in mild AD and mild cognitive impairment.