An overarching aim of the Seychelles Child Development Study (SCDS) is to evaluate the effects of prenatal methylmercury exposure from fish consumption on childhood neurodevelopment. A cohort of 779 mother-child pairs was recruited in 1989-90 and the children were examined at multiple ages from 6 months to 24 years. Motivated by SCDS data, we developed models to examine exposure effects on multiple outcomes fit within a single model. I will start by motivating and describing a model to jointly estimate exposure effects on multiple continuous outcomes at a single age, where the outcomes are manifestations of different outcome classes or "domains." In the second part of the talk, I will use a similar modeling framework in a longitudinal model to estimate exposure effects on multiple cognitive outcomes measured at three ages. Our Bayesian hierarchical models include overall domain or time-specific slopes with outcome-specific deviations, while accounting for correlations between multiple test outcomes measured on the same subjects. The more complex residual correlation structure we use in the longitudinal setting enables a borrowing of strength across elements of the residual covariance matrix, where the amount of shrinkage may differ for outcome pairs measured at the same versus different ages. Compared to fitting separate models for each outcome, advantages of our framework include the ability to estimate domain-specific or time-specific exposure and covariate effects, and increased power for estimating outcome-specific effects.
