|
Witte Lab
Last updated: September 11, 2003 |
Hierarchical ModelingHierarchical (multilevel) modeling is a statistical approach that generally outperforms a conventional (one-level) analysis by using multiple level models. Specifically, hierarchical models can provide more reasonable and stable parameter estimates than conventional analytical approaches. In addition, this technique deals with problems of multiple comparisons and allows one to adequately model multilevel data within a hierarchical framework. We are extending this technique to cancer research and genetic epidemiology through a combination of application, methodological development, and simulation studies. We provide an extensive application of this approach in analyzing case-control data on diet and breast cancer. Here, hierarchical modeling gives more reasonable and precise relative risk estimates for food-effects by using a second model to pull conventional estimates toward each other when the corresponding variables have similar levels of nutrients (Witte et al., Epidemiology 1994;5:612-621). We have also undertaken a simulation study showing that hierarchical modeling generally gives more accurate effect estimates than standard techniques (Witte and Greenland, Statistics in Medicine 1996;15:1161-1170). In addition, we are advancing this valuable technique by showing how one can use hierarchical modeling in genetic epidemiologic research to evaluate multiple candidate genes (see, for example, Witte, Genetic Epidemiology 1997;14:1137-1142), and by providing to the scientific community software for undertaking hierarchical modeling (Witte et al., Epidemiology 1998;9:563-566; Witte et al. Epidemiology 2000;11:684-688; Conti and Witte, Am J Hum Genet 2003;72:351-363 ). This research is supported by the National Institutes of Health. |