Scientists are connecting massive environmental pollution data with birth registries to identify birth defect risk factors. It is a significant challenge to identify associations between maternal exposure to toxic chemicals and malformations in offspring in this type of study. We propose a novel visual analytics approach to addressing this challenge. The approach consists of a visual analytics pipeline where analysts can gradually and interactively refine the set of potential risk factors. The approach tightly integrates the following techniques: (1) Statistical analysis methods such as Point-Biserial Correlation analysis and Logistic Regression; (2) a risk pattern mining technique; and (3) visualization techniques such as Parallel Coordinates and the Value and Relation display. We demonstrate the usefulness of the approach using a dataset from an ongoing case-control study in Texas.