STATISTICAL METHODS FOR REPRODUCTIVE EPIDEMIOLOGY

点击翻译按钮获取中文摘要

Currently reproductive epidemiology studies are performed to identify hazards to human reproduction, both in terms of occupational and environmental risk factors as well as host risk factors. The entire pregnancy history for a couple is taken in such a study. However, within a family, pregnancy outcomes are not independent events. For instance, some studies have suggested that within a family the occurrence of a fetal loss is associated with greater chance of subsequent fetal loss. In addition, this nonindependence is also true for outcomes such as preterm delivery and for low birth weight offspring. Thus, statistical procedures which assume independent observations may not give the correct interpretation to the results. The dependence of observations made in the same family has long been recognized in the analysis of data from animal studies and methods to deal with the problem in this context have been proposed. However, few methods have been proposed for use in the analysis of data from human reproductive studies. Methods proposed for analysis of animal data, where multiple conceptuses are exposed simultaneously, may not be appropriate for the analysis of human reproductive data in which exposure or other variables may vary from pregnancy to pregnancy. One such method has been proposed by Kissling (1981) and was applied to data from an occupational reproductive study. This work will examine four broad classes of models for use in the analysis of human reproductive outcomes. Methods for parameter estimation and hypothesis testing will be developed that can be used when observations are not independent, as is the case in reproductive studies. The properties of these estimators and tests will be extensively studied through a multitude of Monte Carlo simulation studies. These models will be applied to an existing occupational reproductive epidemiology data set examining the effects of styrene exposure on reproductive function. The results of analyses for the different models as well as for the analysis of this data set using the simple logistic model, which ignores the independence issue, will be compared and contrasted to each other.