Study Design in the Study of Adverse Birth Outcomes: Clinic-Based Versus Representative Samples

Barbara Laraia, University of California, San Francisco

To understand the predictors of birth outcomes, it is advantageous to study these in representative samples with rich clinical and biological data. However, this is never possible. Researchers must make difficult design choices, exchanging strengths in some domains for others. We compare data from the Pregnancy, Infection and Nutrition study to area birth records, matching 83.7% of the PIN participants to their birth records. Using a two stage model, we will directly estimate the bias between the two samples and provide a correction if bias exists. We will use distance to delivery hospital, and differential distance between delivery hospital and other area hospitals, as instruments to predict who is in PIN in the first stage equation, then estimate predictors of several outcomes collected on PIN women. The estimation and evaluation of the two-stage model will speak directly to potential for bias and provide a correction if such a bias exists.