Objective: To determine whether the excess mortality observed in patients who received both levodopa and selegiline in a randomised trial could be explained by revised diagnosis of Parkinson's disease, autonomic or cardiovascular effects, more rapid disease progression, or drug interactions.

Design: Open randomised trial and blind comparison and reclassification of the cause of death of patients who were recruited from 93 hospitals between 1985 and 1990 and who had died before December 1993 in arms 1 and 2.

Setting: United Kingdom.

Subjects: 624 patients with early Parkinson's disease who were not receiving dopaminergic treatment and a subgroup fo 120 patients who died during the trial.

Interventions: Levodopa and a dopa carboxylase inhibitor (arm 1), levodopa and a dopa decarboxylase inhibitor in combination with selegiline (arm 2), or bromocriptine alone (arm 3).

Main outcome measures: All cause mortality for 520 subjects in arms 1 and 2 and for 104 subjects who were randomised into these arms from arm 3. Cause specific mortality for people who died in the original arms 1 and 2 on the basis of the opinion of a panel, revised diagnosis and disability ratings, evidence from clinical records of either autonomic or cardiovascular episodes, other clinical features before death, and drug interactions.

Results: After extended follow up (mean 6.8 years) until the end of September 1995, when arm 2 was terminated, the hazard ratio for arm 2 compared with arm 1 was 1.32 (95% confidence interval 0.98 to 1.79). For subjects who were randomised from arm 3 the hazard ratio for arm 2 was 1.54 (0.83 to 2.87). When all subjects were included the hazard ratio was 1.33 (1.02 to 1.74) and after adjustment for other baseline factors it was 1.30 (0.99 to 1.72). The excess mortality seemed to be greatest in the third and fourth year of follow up. Cause specific death rates showed an excess of deaths from Parkinson's disease only (hazard ratio 2.5 (1.3 to 4.7)). No significant differences were found for revised diagnosis, disability rating scores, autonomic or cardiovascular events, other clinical features, or drug interactions. Patients who died in arm 2 were more likely to have had possible dementia and a history of falls before death compared with those who died in arm 1.

Conclusion: The results consistently show excess mortality in patients treated with combined levodopa and selegiline. Revised diagnosis, autonomic or cardiovascular events, or drug interactions could not explain this finding, but falls and possible dementia were more common in arm 2. The results do not support combined treatment in patients with newly diagnosed Parkinson's disease. In more advanced disease, combined treatment should perhaps be avoided in patients with postural hypotension, frequent falls, confusion, or dementia.