The population to be studied was adults with AD diagnosed on the basis of standardized criteria of the Diagnostic and Statistical Manual of Mental Disorders
, 4th edition,1
or the National Institute of Neurological and Communicative Disorders and Stroke/Alzheimer's Disease and Related Disorders Association.33
Treatment included therapeutic doses for at least 12 weeks (the minimum period needed to see a treatment effect) of any of the available second-generation ChEIs. Cognitive outcomes must have been measured, on any validated scale. We accepted only original reports (not secondary publications of previously reported data) of randomized, double-blind, placebo-controlled, parallel-group clinical trials.
We searched the English-language literature, using MEDLINE and EMBASE, from January 1980 to May 2002, since the earliest publication concerning clinical use of a second-generation ChEI appeared in the 1990s.34
Key words were cholinesterase inhibitor AND Alzheimer, and the limits were randomized controlled trials, English and human. Searches were also conducted for individual ChEIs (key words donepezil, E2020 or Aricept; rivastigmine, ENA 713 or Exelon; galantamine or galanthamine; AND Alzheimer). The Cochrane databases were searched from inception. Recent review articles and published reports of clinical trials were manually cross-referenced, as were all references and bibliographies from retrieved articles.
“Differential” photocopying was used to blind raters as to authors and their location and as to date and journal of publication to reduce potential bias. First, 2 raters reviewed the Methods section of all articles identified. Articles meeting the inclusion criteria were then rated on quality by 2 raters using the Jadad scale,35
which is simple to use and has been validated.35,36,37
Disagreements regarding inclusion and quality were settled though consensus discussion.
From the Results section of the included articles 3 raters extracted the numbers of patients in the following categories: responding or not responding to treatment, reporting any adverse event, discontinuing treatment (“dropping out”) for any reason and dropping out because of adverse events. Discrepancies were managed though consensus discussion among all the reviewers.
Data relating to responders were extracted with the use of 2 definitions. Global responders
were defined as subjects rated as “improved” (i.e., excluding “unchanged” but including “minimal improvement” and better) on a global assessment scale (Clinical Global Impression of Change [CGIC]38
or Clinician Interview-Based Impression of change plus caregiver input [CIBIC+]39
); the intent-to-treat (ITT) population was the denominator for proportions. This meta-analysis focused on global improvement since it is an important outcome and a regulatory requirement that includes treatment effects not captured on strictly cognitive scales, and it measures clinically relevant change.28,40 Cognitive responders
were defined as subjects with a 4-point or greater improvement on the Alzheimer's Disease Assessment Scale–cognitive portion (ADAS–cog);41
the ITT population was the denominator for proportions. This is the standard definition of responder, as first defined by the US Food and Drug Administration.42
The denominators for the proportions of subjects reporting any adverse event, dropping out for any reason or dropping out because of adverse events were also the ITT population. Manufacturers were contacted for data missing from the published reports.
For the main analyses, we identified the numbers of responders and nonresponders in each of the 2 groups within each study, calculated response rates for the treated patients (Rt) and placebo recipients (Rp), then calculated an effect size for each study: the difference in response rates (Di = Rt – Rp).
values) were pooled across the studies with the random-effects meta-analytic model developed by Cochran,43,44
which essentially weights each study's effect size by its sample size and by the between-study variance. This model yields a pooled mean point estimate and a 95% confidence interval (95% CI). Thus, it generally creates wider confidence intervals than other methods.45
However, because it incorporates between-study differences, it tends to mitigate discrepant results when there is a great deal of variation. Such variation is to be expected because of the wide variations found in this disease and in its response to treatment.
This procedure was followed for global response, cognitive response and other outcomes of interest, which included adverse events, dropout for any reason and dropout because of adverse events. For this research, adverse events were defined as any adverse event that emerged during treatment, as reported by the original authors.
To evaluate publication bias, we generated a funnel plot comparing effect size with sample size46
and evaluated the results with the Begg and Mazumdar adjusted rank correlation test.47
The number needed to treat (NNT) and the number needed to harm (NNH) were calculated according to the method of Cook and Sackett.48
The NNT is the reciprocal of the risk difference when the outcome is positive, and the NNH is the reciprocal of the risk difference when the outcome is negative. NNTs were based on the proportion of global responders and NNHs on the proportion of patients reporting adverse events. For studies in which the difference between treatment groups is not statistically significant, CIs may cross zero and, as such, are difficult to characterize; we used the method described by Altman49
to overcome this difficulty.
We performed subanalyses to assess the impact of ethnicity (Asian v. predominantly white patients), dose, drug, duration of treatment and CGIC definition. Compared with white patients, the Japanese require lower doses of many psychotherapeutic medications50
and may have a higher rate of response to ChEI therapy.14
Dosages were grouped according to common prescribing practice or analysis of the literature, or both, as follows: subtherapeutic (donepezil, 1 to 3 mg/d; galantamine, 8 mg/d), low (donepezil, 5 mg/d; rivastigmine, 3 to 6 mg/d), high (donepezil, 10 mg/d; galantamine, 16 to 24 mg/d; rivastigmine, 9 to 12 mg/d) or above that recommended (galantamine, 32 mg/d); low-dose and high-dose groups were compared. Studies were grouped by duration of treatment, with shorter term defined as 12 to 14 weeks and longer term as 24 to 52 weeks. The ChEIs were also grouped by definition of CGIC. The CGIC scale indicates degrees of change from baseline as follows: 1, marked improvement; 2, moderate improvement; 3, minor improvement; and 4, no change. Thus, CGIC1–4
includes no change and CGIC1–2
is the strictest definition.