Investigation by Parkinson's Disease Research Group of United Kingdom into excess mortality seen with combined levodopa and selegiline treatment in patients with early, mild Parkinson's disease: further results of randomised trial and confidential inquiry

Statistical analysis

The death rates in arms 1 and 2 were compared using the log rank test and Kaplan-Meier survival curves. The relative mortality hazard ratio and 95% confidence intervals were calculated using Cox's proportional hazards model, which enabled adjustment for possible prognostic factors. The adequacy of the proportional hazards model was tested using a log-time interaction with treatment group to check whether the hazard ratio changed with follow up.14

Codes for specific causes were grouped under the standard classification headings except for the more common conditions such as ischaemic heart disease (410-414) and cerebrovascular disease (430-438). Comparison of categorical and continuous variables were analysed using the χ² test or Fisher's exact test for categorical variables and the t test for continuous variables.

Results

Mortality

Our previous paper reported the results of the interim analysis of December 1994.1 This analysis included deaths only before the end of 1993 because of the delay in notification of deaths from the NHS central register. This report provides data on mortality up to the end of September 1995, when arm 2 was terminated, providing an additional 21 months of follow up (average 6.8 years) and new results on 104 patients randomised from the bromocriptine arm to either arm 1 (53 patients) or 2 (51 patients).

Death rates were similar in arms 1 and 2 during the additional follow up (table 1). They were higher in patients who were rerandomised to arm 2 (levodopa and selegiline) than in those rerandomised to arm 1 (levodopa alone) (hazard ratio 1.54 (95% confidence interval 0.83 to 2.87)). The overall hazard ratio for mortality in arm 2 (levodopa and selegiline) compared with arm 1 (levodopa alone) when subjects who had been randomised a second time were included was 1.33 (1.02 to 1.74) (P=0.038 in log rank test); this was little altered when subjects who had been rerandomised were excluded. The confidence intervals have not been adjusted to take account of the early termination of arm 2; the inclusion of additional information means that this updated result is much less affected by the decision to stop treatment early. After adjustment for other baseline factors—age, sex, duration of Parkinson's disease, disability before treatment, year of entry to trial—the hazard ratio was 1.30 (0.99 to 1.72). Analysis based on patients receiving treatment (“on treatment analysis”) gave a hazard ratio of 1.39 (0.94 to 2.05).

Figure 2 shows the updated Kaplan-Meier curve. Although a test of departure from the assumption of proportional hazards was not significant, the excess mortality was greatest in the third and fourth years of follow up and was more apparent for the on treatment analysis (table 2).

As entry to the trial stopped in September 1990, information on mortality was complete for the first 5 years of follow up, so the results for the first 5 years were unaffected by the early termination of arm 2. The hazard ratio for the first 5 years for arm 2 compared with arm 1 was 1.38 (0.95 to 2.04).

Cause of death

Up to December 1993, 120 patients died (44/249 (17.7%) in arm 1 and 76/271 (28.0%) in arm 2). Twenty four cases had information from a postmortem examination. As information was not available for 21 cases because notes had been destroyed or lost, we relied only on information from the trial assessments, which may not have had information about the terminal event. The kappa coefficient15 for the 20 cases classified by the panel on the two occasions was 0.76 for the underlying cause of death, 0.73 for the confidence rating, and 1.0 for the diagnosis of Parkinson's disease (kappa >0.75, excellent; 0.40 to 0.75, fair to good; >0.40, poor). The panel reached a diagnosis in 90 cases. It decided that information was insufficient for the remaining 30 cases to be certain of the cause of death. The pattern of cause specific mortality based on the panel's classification was similar to that previously reported (table 3). Only Parkinson's disease showed an excess of deaths (hazard ratio 2.50 (1.32 to 4.73), whereas for other causes combined the hazard ratio was 1.21 (0.76 to 1.93).

When arms 1 and 2 were compared there were no significant differences between any of the clinical features and the mode of death (table 4). Patients in arm 2 were more likely to have possible dementia, and within three months of death they were more likely to have falls, postural dizziness, and shortness of breath. However, the proportion of cases with a revised diagnosis—for example, multiple system atrophy, progressive supranuclear palsy, Alzheimer's disease, and cerebrovascular disease—was slightly greater in arm 1.

The mean confidence score was slightly worse in arm 2 (3.00 v 2.59, P=0.19) and the postmortem rate was slightly lower. In arm 2 more deaths were classified as sudden, but the proportion of unexpected deaths and the proportion of deaths that occurred at home were similar. There was no evidence that patients in arm 2 had greater cardiovascular comorbidity as assessed by clinical disease or cardiac drugs, and patients in arm 2 were less likely to be taking antidepressants before death.

Disability subscales

Previous analyses showed that progression of disability was similar in the two treatment groups. As this contrasts with the excess mortality from Parkinson's disease seen in arm 2, we examined this further by analysing the disability rating before death (table 5). There were no significant differences in any of the subscales between the two arms. Fifty two per cent of the disability ratings (91/176) were made within the year preceding death, 20% (35/176) within 1-2 years before death, and 28% (50/176) more than 2 years before death. This distribution was similar for arms 1 and 2.

Drugs

Ascertainment of the type of drugs taken around the time of death is important if the excess mortality observed with selegiline and levodopa is due to some acute toxic mechanism. We obtained drug information on 91 of the 120 patients who died, which was similar for arm 1 (31/44 (71%)) and arm 2 (60/76 (79%)) Almost all of the patients were taking levodopa 3 months before death (arm 1, 30/31 (97%); arm 2, 59/60 (98%)). In contrast, 23% (7/31) of the patients in arm 1 were no longer in the arm of treatment to which they had been randomised compared with 87% (52/60) taking selegiline in arm 2.

Discussion

The updated results show a relatively increased mortality for the combined levodopa and selegiline treatment compared with treatment with levodopa alone of around 35%, equivalent to one excess death per 75 patients treated for 1 year. The mortality ratios were remarkably consistent regardless of whether all deaths, deaths of patients who were rerandomised, or deaths in the first 5 years were considered. These estimates are lower than previously reported1 and are more realistic, as the previous result was based on an interim analysis. Although our confidence intervals are comparatively narrow, they are all around unity so that some results are significant while others are not. Had arm 2 of the trial continued, it is possible that the mortality would have diminished further, and both previous and current results could simply reflect chance. However, the similarity of the size of the effect in patients rerandomised from arm 3 provides an independent replication of the findings seen for the main group allocated to arms 1 and 2. Whereas subjects withdrawn from arm 3 may be unrepresentative, the randomisation process ensures that the internal comparison is valid and can be viewed as a separate trial. In addition, the complete mortality results based on the first 5 years of follow up were not affected by the decision to terminate arm 2 of the trial and so are less likely to represent a random high value.

We emphasise that this trial fails to support the hope that combined treatment might be associated with reduced mortality or improvement in disability rating scales. Unfortunately, no data were collected on quality of life or mood, so we cannot comment on whether combined treatment may have benefited these measures.

Possible explanations

One problem with the observed excess mortality is the lack of a clear reason for this observation. Other conditions which mimic Parkinson's disease are difficult to diagnose as atypical features often develop only after several years16 and they have a worse prognosis than Parkinson's disease. We did not, however, find a higher rate of revised diagnosis in arm 2 compared with arm 1 (11% v 15%). Another criticism was that an intention to treat analysis was inappropriate because of the comparatively large number of patients who withdrew.9 Ideally, we would like to have had accurate drug data on all of the patients, including those who withdrew at the time arm 2 was terminated. For the subgroup of patients who died for whom data were available, most patients in arm 2 were still receiving combined treatment while only a fifth of patients in arm 1 had selegiline added to their drug regimen before they died.

Since the original publication two studies have shown that selegiline diminishes autonomic responsiveness and increases risk of orthostatic hypotension. We postulated that if this mechanism was clinically important we should observe more sudden or unexpected deaths, hypotensive episodes, falls, and possibly a higher postmortem rate in arm 2. Our findings provide limited support for this hypothesis, though none of the differences were significant. However, retrospective analysis of clinical notes is likely to significantly underestimate the true rate of any hypotensive effect of selegiline and levodopa. The most marked difference in clinical characteristics between the two arms was for falls before death and possible dementia. Falls commonly occur among patients with severe Parkinson's disease because of postural instability and akinesia as well as any autonomic effect. A randomised controlled trial of selegiline, α tocopherol, or placebo for Alzheimer's disease noted a significant increase in falls and syncope in patients receiving selegiline in combination with α tocopherol.21

Dementia is not uncommon in association with Parkinson's disease and is a poor prognostic factor. Selegiline and levodopa treatment may directly result in increased confusion. Alternatively, dementia may be a marker for general frailty and increased risk of adverse drug effects.24

One explanation could be that selegiline and levodopa contribute to hypotensive episodes which increase the risk of either a heart attack or stroke, especially in elderly patients with pre-existing atherosclerotic disease. However, both our analyses of cause specific mortality and of comorbidity did not support this notion. If combined treatment actually accelerated disease progression, and hence death from Parkinson's disease, subjects in arm 2 would be expected to have worse disability scores and to be more disabled or bedbound before death. The data do not, however, support this hypothesis either. The use of cardiac or antidepressant drugs was no greater in arm 2 than arm 1, although we cannot rule out the possibility of a drug interaction because some of the patients' records were destroyed.

One remaining possibility is that combined treatment is harmful to a subgroup of patients. This might explain why the greatest comparative mortality ratio was seen for the analysis of patients on allocated treatment between 2 and 4 years. If susceptible subjects are selectively removed from arm 2 the mortality ratios would be expected to return to unity with further follow up because only non-susceptible subjects would then remain in the study.

Clinical implications

Despite uncertainties there are some clear clinical implications from these results. There is no evidence that combined treatment with levodopa and selegiline confers advantages over levodopa treatment alone in terms of mortality or morbidity in patients with early, mild Parkinson's disease. There seems little logic in giving patients with newly diagnosed disease combined treatment, although treatment might be started with selegiline alone and then withdrawn if levodopa treatment was indicated. Clinicians should determine whether the addition of selegiline for severely disabled patientd provides additional symptomatic benefit. In these patients quality of life is generally more important than quantity of life, and each case should be reviewed on its merits. However, if patients have clinically significant orthostatic hypotension, cardiac arrhythmias, confusional states, hallucinations, or deteriorating cognitive function, gradual and slow withdrawal of selegiline over 4 to 6 weeks should be seriously considered.