Use of surrogate respondents in studies of stroke and dementia