Selegiline, or the problem of early termination of clinical trials

Parkinson’s disease is the second most common neurodegenerative disorder, after dementia. About 1.4% of people aged 55 years or over have Parkinson’s disease,¹ and because of the aging of Western populations the absolute number of patients is rapidly increasing. Until now, treatment has been mainly symptomatic, but much effort is being put into developing neuroprotective agents that may stop progression or even cure the disease. Clearly, unrecognised adverse effects of such treatments may potentially affect large numbers of patients and any suggestion of such effects needs thorough investigation.

Selegiline has probably become the most controversial drug in Parkinson’s disease during the past decade. Its presumed efficacy was initially ascribed to neuroprotection due to inhibition of monoamine oxidase-B, then to a symptomatic effect, and more recently again to neuroprotection, this time due to inhibition of apoptosis. The greatest controversy, however, occurred because selegiline caused the early termination of the intervention arms of two large multicentre studies—for completely different reasons. In the DATATOP study of the US Parkinson Study Group subjects randomised to receive selegiline did better than those randomised to placebo or tocopherol in that they reached the endpoint (start of levodopa treatment) significantly later.² In contrast, in the trial of the Parkinson’s Disease Research Group in the United Kingdom mortality among patients receiving selegiline in addition to levodopa (plus decarboxylase inhibitor) was higher than in those taking only levodopa (plus decarboxylase inhibitor).³

This latter finding was totally unexpected and generated much debate,⁴,⁵ but it did cause sales of selegiline to drop considerably.⁶ In this week’s BMJ Ben-Shlomo et al provide supplementary information from the UK trial, including further follow up data and more detail about patient characteristics and causes of death, in an attempt to explain their previous findings (p 1191).⁷

The selegiline plus levodopa arm of the UK trial was terminated at the end of September 1995 as a result of an increased risk estimate based on deaths till the end of 1993. Consequently, all end of trial analyses are biased. The more valid and unbiased estimate of the risk of mortality based on the five year follow up shows a hazard ratio for these patients compared with those receiving only levodopa of 1.38 (95% confidence interval 0.95 to 2.04). Another unbiased estimate comes from the subjects who were re-randomised from the bromocriptine arm (hazard ratio 1.54; 0.83 to 2.87). Although this is a highly selected group, the authors suggest, plausibly, that because of the randomisation these data can be viewed as if they came from an independent trial.

The difference in mortality seemed to be highest in the third and fourth year of follow up, after which it diminished. This finding was previously criticised because of the lack of a biological explanation.⁵ The authors dismiss the possibility that had the combined treatment arm of the trial continued, the mortality difference might have diminished even further. Their dismissal may not, however, be justified, for at least two reasons. Firstly, notwithstanding the similar findings in the bromocriptine group, the data are compatible with a “randomly high” increased mortality at 24 to 48 months. This holds in particular for the “as treated” analysis, which seems the more informative here. Secondly, one can question the implicit assumption that had the mortality ratio further decreased this would have meant that the findings of increased mortality were due to chance. Hardly any effective therapy is entirely harmless. Even if selegiline in combination with levodopa reduces mortality in the long run, the net effect on mortality might be unfavourable in situations where there is little to gain—for example, at the beginning of the disease.

The cause specific mortality rates do not suggest a specific cause for the excess deaths. An excess of deaths from Parkinson’s disease was reported, but this probably reflects lack of further information since in almost half the cases where the panel could not reach a diagnosis the diagnosis on the death certificate was Parkinson’s disease. Also the comparison of clinical characteristics in those who died between those who took only levodopa and those on combination treatment yielded no clear clues to explain the extra deaths in the combined group as all differences were far from significant.

What therefore can we conclude about the hazards of combination treatment with levodopa and selegiline? Firstly, the UK trial finds no significant increased mortality due to combined treatment with levodopa and selegiline. Secondly, because of the interim analyses, doubts have been raised, where previously we had no indication that selegiline had any detrimental effect whatsoever. Any new evidence incriminating selegiline will weigh heavily. The possibility that the net effect of positive and adverse effects of combination therapy with selegiline depends on background risk or severity of disease should be considered. Thirdly, the recommendation to avoid combined treatment among patients with more advanced Parkinson’s disease and postural hypotension, frequent falls, confusion, and dementia is not based on unequivocal results from the trial but rather on clinicians’ beliefs.

What does this mean for the treatment of patients with Parkinson’s disease? Unfortunately, the early termination of the arms involving selegiline in both the US and the UK trials has limited the evidence on the long term effects of selegiline alone and in combination with levodopa and diminished the possibility that this can ever be validly obtained. Appreciation of the effect of combination treatment on morbidity is almost impossible in the UK trial, because only the results of the intention to treat analyses have been reported, whereas more than half of the study population had withdrawn from their original randomised treatment. On the basis of the limited evidence available, a cautious recommendation seems to be not to start combination treatment with selegiline and levodopa in patients with newly diagnosed Parkinson’s disease. At the moment, however, there is little evidence to advise people who have been using both drugs for years and seem to be doing fine to change their treatment.

We commissioned two editorials on the paper by Ben-Shlomo et al on p 1191. The first, by Monique Breteler, discusses the clinical implications. The second, by Keith Abrams, discusses the difficulties of interpreting interim analyses of trials