Analysis of case-control data derived in part from proxy respondents.

In the conduct of case-control studies, it has been a general experience that some cases are identified only after death, and that others cannot be personally interviewed because of extreme morbidity. As a way of dealing with the dilemma of obtaining data on unavailable cases, investigators have sometimes chosen to interview a relative or friend (a proxy respondent) whose responses are based on his or her knowledge of the lifestyle, medical history, occupational history, etc., of the unavailable index subject. The same procedure may be used for controls who are unable to respond personally to the questions of the investigator. Some methodological inquiries have been carried out to assess the agreement between index (i.e., personally interviewed) and proxy respondents. In these studies, the proxy respondents have usually been the spouses of the index respondents, and the questions have most often concerned the frequency and/or quantity with which foods were consumed (1-4). Results from these studies show a high concordance between group level aggregate measures, such as mean nutrient intake or mean frequency of consumption, estimated from index and proxy responses. On an individual level, however, dietary information obtained from proxy respondents appears to have