Anticonvulsant teratogenesis: 2. Statistical methods for multiple birth outcomes

The purpose of this paper is to demonstrate the application of generalized estimating equations to assess an exposure effect using multiple birth outcomes. This multivariate approach provides the flexibility of regression modeling and improved power, as compared to series of univariate analyses or collapsing the multiple end-points to a single indicator of affectedness. Motivating the discussion will be a large cohort study designed to assess the effects of anticonvulsant medications on a variety of birth outcomes, including major malformations, and growth and weight parameters, as well as a broad spectrum of minor physical anomalies. Because the study is still in progress, the aim here is not to present a definitive analysis, but to present and describe the application of these recently developed statistical methods to analyze studies with multiple outcomes. For simplicity, we will focus on the control and drug-exposed groups only from that study (ignoring the seizure history group), and we will concentrate on an analysis of minor physical anomalies.