Background: Use of surrogate responders often needs to be considered in case-control studies with a high case fatality.
Methods: Agreement between 98 colon cancer patients and their closest relative was expressed as a percentage of the exact agreement and by Kappa coefficients and intra-class correlation coefficients.
Results: The percentage of "don't know" answers was higher for surrogates than for index cases and the highest percentage was seen for questions on early events like childhood diseases. Agreement was best for responses to dichotomous questions on smoking and for prevalent or chronic diseases like diabetes or psoriasis, and lower (54-64%) when a quantitative response of, e.g., smoking was requested. The next-of-kin reported fewer job periods than the study person, 4.5 and 2.8, respectively, and there was a higher agreement for the latest job held than for the longest held job. We found an overall agreement between 91% and 100% for responses to ever having worked in a specific type of industry or occupation.
Conclusions: Use of next-of-kin data will often be a better alternative than excluding severely ill or deceased cases, if the exposure under study correlates with disease progression.
Copyright 2003 Wiley-Liss, Inc.