Semiparametric Methods for Secondary Outcomes in Case-control Studies

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Modern case-control studies typically involve the collection of data on a large number of variables, often at considerable logistical and monetary expense. These data are of potentially great value to subsequent researchers, who although not concerned with the disease that was the subject of the original study, may want to use the available information towards an analysis of the effects of an exposure on a secondary outcome other than the disease that defined the original case series. A difficulty with re-using data in this way is that the case-control sampling scheme used in the original study will likely induce bias in estimates of log odds ratios and other parameters in the secondary study, if conventional analytical approaches are used. In this proposal, we plan to develop novel statistical methodology for making robust and highly efficient inferences on the effects of an exposure on a secondary outcome under case-control sampling on a different outcome. An important advance is that the proposed methodology applies whether or not sampling probabilities are known to the investigator, and is particularly useful when, as in most observational studies, one needs to adjust for a moderate to large number of confounders. We will use the proposed new methodology to accurately evaluate the effects of lead exposure on the risk of osteoporosis and on decline in cognitive function, using data from the Nurses' Health Study case-control study of lead and hypertension. PUBLIC HEALTH RELEVANCE: Modern case-control studies typically involve the collection of data on a large number of variables, often at considerable logistical and monetary expense. These data are of potentially great value to subsequent researchers, who although not concerned with the disease that was the subject of the original study, may want to use the available information towards an analysis of the effects of an exposure on a secondary outcome other than the disease that defined the original case series. A difficulty with re-using data in this way is that the case-control sampling scheme used in the original study will likely induce bias in estimates of regression parameters in the secondary study, if conventional analytical approaches are used. In this proposal, we plan to develop novel statistical methodology for making robust and highly efficient inferences on the effects of an exposure on a secondary outcome of a categorical or continuous nature, under case-control sampling on a different outcome. An important advance is that the proposed methodology applies whether or not sampling probabilities are known to the investigator. Furthermore, the methods are particularly useful when, as in most observational studies, one needs to adjust for a moderate to large number of confounders, and data on some key covariates may be missing for a subset of subjects. The new methodology will be evaluated via extensive simulation studies, and will be used to accurately assess the effects of lead exposure on the risk of osteoporosis and on decline in cognitive function using data from the Nurse's Health Study case-control study of lead and hypertension.