|
Abstract:
|
Understanding the human health effects of exposure to a toxicant is complicated when the toxicant affects multiple endpoints. Furthermore, the endpoints may cluster into broad domains such that similarity of the exposure effects on outcomes within a domain is expected. An example of this situation is a study of the effects of prenatal methylmercury (MeHg) exposure on children in Seychelles, where the 21 neurodevelopmental outcomes measured at nine years cluster within seven domains. We discuss results of a Bayesian model that allows the effect of MeHg exposure to vary across outcomes, while shrinking the outcome- and domain-specific effects to the extent warranted by the data. Our model allows us to estimate these effects as well as an overall exposure effect. By analyzing several outcomes simultaneously, one gains power to detect small, but potentially important, health effects.
|
