Effect of adding selegiline to levodopa in early, mild Parkinson's disease. Formal systematic review of data on patients in all relevant trials is required.

Editor—The increased death rate with selegiline in the Parkinson’s Disease Research Group’s trial highlights the problem of interpreting the results of a small trial in isolation. Ben-Shlomo et al emphasised the internal consistency of this finding, which is not surprising because much of the original data were reanalysed. The hazard ratio was lower than in the original report (1.32 v 1.57), which suggests that the trial might have stopped on a random high. There is no formal systematic review of all unconfounded randomised trials comparing long term selegiline with control treatment in patients with early Parkinson’s disease. One attempt by the manufacturers of selegiline did not say which trials were included. I found three such trials in the Cochrane controlled trials register, with follow up ranging from 2.5 to 8.2 years—that is, after the increase in deaths with selegiline started in the Parkinson’s Disease Research Group’s trial. Two trials did not start levodopa treatment for the first few months, unlike the Parkinson’s Disease Research Group’s trial, but most patients had been taking levodopa for several years at the end of follow up. 4 In one trial all patients were switched to selegiline treatment after 1.5 years, which would dilute any treatment effect. I performed a meta-analysis of the number of deaths at the last available follow up (figure). This is simplistic because of unequal follow up periods in the selegiline and control arms in some trials, but the results did not change after accounting for this. The results are based on under 2000 patients and show a non-significant 16% increase in the relative risk of death with selegiline (lower than that seen in the Parkinson’s Disease Research Group’s trial). None of the other trials showed a trend against selegiline, but the confidence intervals were wide and compatible with an excess of deaths. There was no evidence of heterogeneity among the trials. The Parkinson’s Disease Research Group’s trial found more falls and possible dementia in those who died in the selegiline arm, but this is difficult to interpret without similar data for survivors. Dementia is a poor prognostic factor and is probably associated with a greater risk of falling. Rather than selegiline causing dementia, it might be that, by chance and despite randomisation, more patients in the selegiline arm had a propensity to develop dementia. This crude meta-analysis suggests that selegiline is unlikely to reduce mortality, but whether it increases mortality remains unclear. A formal systematic review of all relevant trials using data on individual patients data to assess survival, disability, and the risk of dementia is needed. This will require close collaboration between all the trialists. Unfortunately, a recent metaanalysis considered only survival and did not include the two largest trials. 3