Predicting probabilities of pregnancy and multiple gestation from in vitro fertilization-a new model

Objective To develop a statistical model that adjusts for variation between patients and adequately predicts the observed distribution of pregnancies among singletons and multiple gestations of various orders. Methods All in vitro fertilization (IVF) cycles from the inception of the IVF program at Women and Infants' Hospital on May 26, 1988, until December 31, 1993, were evaluated using logistic regression in selected subsets. Results A new cycle-one specific model uses three different probabilities: P1, the probability of pregnancy (predicted by age and total embryo score); P2/P1, the conditional probability of finding a second implantation in those who had become pregnant with at least one (predicted by total embryo score); and P3/P2, the conditional probability of finding a third implantation in those who had become pregnant with at least two (with no significant predictors). This is the first model to use these three adjusted probabilities. Conclusion P1 increases with increasing total embryo score but decreases with increasing age. P2/P1 increases with increasing total embryo score but does not depend on age. Embryo scoring is useful because the total embryo score is a better predictor of P1 and P2/P1 than the number of embryos alone. By using patient-specific information (age and total embryo score) and cycle-specific tables, an estimate of the probability of pregnancy and multiple gestation can be provided before embryo transfer.