Commentary

The review by Birks et al is a rare example of an independent analysis of a drug development programme. The authors used careful, modern methodology, and the included trials used recognised outcome measures in appropriate patient groups. The results of the meta-analysis should therefore be valid. On the basis of trials, rivastigmine has been approved by regulatory authorities in many countries. Perhaps surprisingly, however, clinicians need to be cautious in their interpretation of these rivastigmine data. What has gone wrong?

The main problem is that the data for 2 large trials of rivastigmine, including those for almost half of the patients in the phase III trials, were not available for this review. Obviously, these missing data could have a major effect on the result of the quantitative analysis. As they stand, the results are liable to an unknown degree of bias and must be considered to be unreliable. It is unknown if the data from these missing trials were made available to the Federal Drug Administration and other regulatory authorities when they were assessing this drug.

This raises 2 important issues. Firstly, drug companies should be urged to release all the trial data on their drugs.^{1, 2} Increasingly, as clinicians try to base their treatment decisions on unbiased information, they will be reluctant to use treatments where there appear to be gaps in the available evidence. Secondly, reviewers need to clearly report how much existing trial data were unavailable for the review—and the current reviewers are to be commended on this point. However, when a high proportion of existing data are unavailable, and when this is likely to produce a biased estimate of the effectiveness of the treatment (as is possibly the case here), then reviewers should hesitate before undertaking a statistical meta-analysis.

As this evidence currently stands, clinicians using the Cochrane Library to inform their decisions are likely to prefer alternative anticholinesterases to rivastigmine.