Reliability of information collected by proxy in family studies of Alzheimer's disease.

The study evaluated the reliability of data obtained from proxy informants. The index subjects in this study were 81 nondemented participants in the Multi-Institutional Research in Alzheimer Genetic Epidemiology (MIRAGE) study. These index subjects and 159 proxy informants, identified by the index subjects, participated in the study. The kappa statistic with multiple raters per subject (for dichotomous variables) and the intraclass correlation coefficient (for continuous variables) were used to measure reliability. Among proxy respondents who provided answers, there was excellent agreement between proxy responses and the responses of the index subjects (0.7 < or = kappa < or =0.9), with the exception of questions about head injury (kappa = 0.4). A large proportion (>90%) of the proxy informants in this study were able to provide information on most items. Higher nonresponse rates (as high as 30%) were observed for medication history and women's health questions. This study supports the reliability of proxy responses for most categories of questions that are elicited in typical epidemiological studies, including the MIRAGE study.