Statistical Methods In Epidemiology--general

Environmental Health Sciences

Investigators

Linked publications & trials

Abstract

Methodologic work progressed in two areas: (1) When data are clustered, e.g. based on litter in a toxicology experiment, or sibship in a human study, conditional logistic regression is often used to allow for dependencies where inherent susceptibility to the endpoint under study varies across clusters. If there are unmeasured factors that vary across clusters and that influence susceptibility to effects of the exposure under study, then there may be residual dependency, which will invalidate such analyses. These factors are called "effect modifiers" in epidemiology. We developed a statistical test for unmeasured effect modification. (2) We have shown that when studying a continuous marker of health, such as blood pressure, one can markedly improve the efficiency of a study (over what would be achieved with random sampling) by our proposed design, which oversamples observations at the extremes, i.e. people with unusually high or low values of the outcome. The analytic strategy is being further developed and applied to studies of neurodevelopmental scores in relation to pesticide exposure. (3) We have shown that the auxiliary covariate data collected in biomedical studies when the primary exposure variable is only assessed on a subset of paritipants can be used to enhance statistical power and improve the accuracy of the effect estimation. We have investigated methods for the generalized linear mixed model (GLMM) with a continuous auxiliary variable, in the presence of a validation subset. We use a kernel smoother to handle continuous auxiliary data. The method can be used to deal with the missing covariate or mismeasured covariate problems in a variety of applications. The proposed method will be applied to an environmental epidemiology study on the relationship between maternal serum DDE level and preterm births.