Skip navigation

New dementia drug

Donepezil (Aricept) is an expensive new drug for the treatment of dementia, costing about £1,000 per patient per year. Its recent introduction has raised questions not only about whether the drug is effective and cost-effective, but also about the extent of evidence that should be available before a new agent is licensed by the regulatory authorities.


Dementia is perhaps the psychiatric disorder of greatest public health importance, with an approximate UK prevalence of 5% at age 65, rising to 20% in the over 80s. It is a progressive, global loss of mental function (concentration, memory and orientation) occurring in clear consciousness. Most cases are due to the degenerative process of Alzheimer's disease (loss of brain cells generally, especially those containing acetylcholine) or to cerebrovascular disease (multi-infarct dementia).

Drug treatment

Various types of drug including antidepressants and antipsychotics are used for symptomatic treatment of dementia. As for specific treatment, reversing the degeneration of nerve cells continues to seem a tall order, and drug research has concentrated on boosting the presumed failure of acetylcholine function in the brains of Alzheimer sufferers, by blocking the enzyme which breaks down acetylcholine. The problem here is that cells containing acetylcholine are found throughout the body, not just in the brain, so side effects are likely.


There is currently only one published trial of this acetylcholine esterase inhibitor, a "preliminary dose-ranging study" [1] from the clinical science department of the manufacturer. It compared donepezil at three doses (1, 3 and 5 mg) with placebo. It was randomised and double-blind for 12 weeks, followed by a two-week single-blind open-label washout. The main analyses were at the "endpoint", which is not defined clearly. There were 40 patients per group, but as it is sold as 5 mg and 10 mg tablets, the main comparison involved only 80 patients.


The primary outcome measure was the group mean score on the 70-point ADAS-cog, an observer-rated scale measuring cognitive function, in which higher scores are better. Other scales were used, including quality of life indicators.


You will search long and hard to find a definition of a clinically useful outcome, the number of patients who benefited, and whether that benefit was sustained in those patients. So no NNTs for donepezil. What we do have is lots of statistics in a vacuum.

There was a small average decline in the ADAS-cog score in those receiving placebo (of 0.7 on a 70-point scale, with a range of change from a fall of 7 points to an increase of 14.5 points), and a small average increase in those on 5 mg donepezil (2.5 on a 70-point scale, with a range of change from a fall of 8 points to an increase in 7 points). The difference was statistically significant. The way in which these results are presented can over-emphasise the differences, as the two graphs show. The top graph shows the whole story, with the 70-point ADAS-cog scale. The bottom graph is roughly how the results were presented in the paper [1].

There were a number of secondary outcomes, some of which showed changes. There was no effect of the drug on the quality of life as assessed by the carers.

A cold and fishy eye

Anyone looking at this study with a cold and fishy eye would be understandably sceptical. Extrapolating a useful clinical response to a new drug from a single small trial of limited duration and demonstrating limited efficacy is impossible. We should ask whether this really amounts to evidence of sufficient strength on which to change practice. There are, though, two unpublished randomised, double blind, placebo controlled studies of 5 mg and 10 mg donepezil, each with about 150 patients per group. According to the UK Drug Information Service [2], both doses of donepezil show overall changes of smaller magnitude than the published study.

It might be argued that because when donepezil is stopped, the small difference in scores disappears, it could maintain improvements over longer periods. But there is little evidence for this.

The appropriate question is different. Are there any patients who do very well with this drug? Again, there is no convincing evidence. There is little consensus on what change on an ADAS-cog score indicates clinical significance. The lowest value seems to be 4 points. Even on that basis the majority of patients will not have any clinical benefit (as the quality of life results judged by the carers indicates). If clinical benefit was associated with a larger change on the scale than 4 points, then only a small minority would benefit.

Licensing new drugs: have we got it right?

Prescribers are in an unenviable position. On the one hand, they will face pressure from patients, relatives and heavy promotion to try a new treatment in a condition of otherwise grim prognosis. On the other hand, the evidence that the drug is effective appears scanty. There is an additional ethical dimension in that many patients will not be able to give truly informed consent about whether they wish to accept these uncertainties.

The UK licensing procedure requires a new drug be manufactured to an acceptable quality, have an acceptable safety record, and be efficacious. Should it be extended, so that an independent agency assesses on behalf of the NHS, whether it is sufficiently well-researched and has acceptable cost-utility for it to be routinely available in the NHS?

Bandolier cannot, of course, answer this question, but the debate will continue, in these pages and elsewhere, notably on the evidence-based-health mailing list [3], where many of these ideas have been aired.

A comment and a suggestion

The conclusion is that donepezil probably produces an improvement corresponding to "turning the clock back" by about three or six months. The benefits are small, and the weight of evidence is unconvincing.

Perhaps this is an opportunity for industry and health service to come together to ask some sensible questions about how we do things. Bandolier suggests setting some prior standard, agreed by consensus, of what constitutes an acceptable and worthwhile change on the ADAS-cog score. Then manufacturers should be challenged to make the individual patient data from the three randomised trials available for single-patient meta-analysis against that consensus. We might then have a clearer picture of what the "true" effect is, and even see if it is possible better to predict patients in whom this benefit will be found. The alternative will be a slanging match with patients and carers in the middle.


  1. SL Rogers, LT Friedhoff. The efficacy and safety of donepezil in patients with Alzheimer's Disease. Dementia 1996 7: 293-303.
  2. Donepezil. UK Drug Information Services. Monograph 4/97/12

previous or next story in this issue