Validation of interview-based disease classifications—A mail survey of physicians

It was found that due to attrition of the sample population, absence of a designated physician and physician non-response, only about two-thirds of the initial study panel were covered, and it was demonstrated that failure of coverage was not random, but selective in relation to characteristics of both doctor and patient.

Abstract

Abstract Diagnostic information obtained in the third round of a longitudinal household interview study of retirement-age couples was submitted to the respondents' physician for confirmation, and the rates of coverage of the survey population and of corroboration of diagnosis were examined. It was found that due to attrition of the sample population, absence of a designated physician and physician non-response, only about two-thirds of the initial study panel were covered. It was further demonstrated that failure of coverage was not random, but selective in relation to characteristics of both doctor and patient. These included a broad range of sociodemographic variables in the respondent, and of professional and personal characteristics of the physician. Differences in corroboration rates among the population subgroups, however, were not great enough to show significance, and accordingly we have concluded that selective though the panel losses were, they probably did not heavily impair the representativeness of the responses which were subject to validation. An exception to this is the fact that confirmation rates did vary by sex, with women's reports somewhat less valid than men's in the opinion of their physician. Corroboration also showed a considerable variation by diagnosis, with the milder and less specific items generally less well confirmed. Understandably, if the physician chosen was not the one involved in management of a particular condition, he very often pleaded ignorance. Over-all, 90 per cent of diagnostic reports were confirmed when the contact doctor was evidently the appropriate one. For all responding doctors (80 per cent of those questioned) the corroboration rate was about 80 per cent. The fact that our levels of corroboration are higher than those seen in many other such studies may be due to attrition involving poor reporters selectively, and to the progressive refinement of information inherent in a longitudinal study.