Issues of selection bias and limitations of the use of family history in the study of disease familial aggregation.

摘要

Familial aggregation of disease is an important issue in genetic epidemiology. It refers to the increased disease occurrence among relatives of a case compared to that among relatives of a control. This dissertation focuses on the question of selection bias and the limitation of the use of family history in epidemiologic studies of familial aggregation.; Selection bias in studies of familial aggregation occurs when study subjects are selected through affected or unaffected family members. In some published studies, family members are counted only once as exposed or unexposed even multiple cases or controls have been identified from that family. In other studies, the way of selecting family members through cases or controls were not specified. This study illustrates the potential bias of the selection method that was used in some published studies by using a reconstructed cohort study design. It is shown that the method in the published studies can cause severe bias under some conditions and in order to get an unbiased estimate of effect measure, a family should be counted once each time a case or a control is selected from the family.; Family history has been used as an exposure variable to indicate one's genetic susceptibility in epidemiology literatures. Family history is affected by several factors such as one's family structure and disease occurrence in the general population. Hypothetical sibship data was used to study the dependency of family history on age and time in a follow-up study. It is shown that family history depends on the age of the study subject and the age of his/her relatives when the proportional hazard model is assumed for the effect of the susceptible genotype. Also it is shown that family history depends on disease occurrence and mortality from competing risks. The implication of this study is that recurrence risk ratio, which measures disease familial aggregation, should be interpreted as an age-specific measure and analysis of family data should reflect this dependency.