We re-reviewed 257 patient records previously reviewed for an incidence study of dementia in Parkinson's disease (PD) to determine the frequency, date of death, and cause of death. We posited that if disease duration is shortened when dementia occurs, then dementia may be far more common than reflected in prevalence studies. There were 17 deaths among 65 demented patients and 28 deaths among 168 nondemented patients. When we matched a subset of the nondemented patients to the demented patients by age and disease duration distributions, the demented subjects had significantly more deaths (p less than 0.02), and survival among demented subjects was decreased (p less than 0.05). Dementia was a significant predictor of death in this sample. We conclude that dementia reduces survival in patients with PD. Incidence is a much better measure of dementia in PD than prevalence because shortened duration makes it less likely to detect dementia in prevalence surveys.