Steven Schwager, Department of Biological Statistics and Computational Biology and Department of Statistical Science, Cornell University

Censored Geometric Regression with Application to Human Subfertility

Abstract: The geometric distribution describes the number of failures observed in a Bernoulli sequence until the first success. We propose a model for geometric distributed data in the framework of generalized linear models, which can incorporate right-censored responses. Covariates can be included to describe the parameters of this distribution. Diagnostics for this model include a test for overdispersion. These methods are applied to a model of the success of pregnancy induced by intrauterine insemination in 200 women who were treated for subfertility at Yale. The number of these insemination services performed before success (pregnancy) is assumed to follow a geometric distribution. Right censoring occurs when a woman and her doctor decide to discontinue before success, for any of a variety of reasons. We discuss and develop an EM algorithm to estimate the parameters for the geometric model and the probability of discontinuing treatment. A Bayesian model is also developed to estimate the number of treatments expected before pregnancy is achieved. This is joint work with Daniel Zelterman at Yale University.