The performance of in vitro fertilization-embryo transfer (IVF-ET) programs is summarized typically as the average probability of achieving pregnancy per cycle. Variation in conception probabilities across women reduces the usefulness of such an aggregate measure. More relevant is the conditional probability of achieving pregnancy on a given cycle following a number of failed IVF-ET attempts. The authors construct a model that accurately describes 1,257 treatment cycles performed at Yale over 571 different women. The model assumes a split population, where some women can never conceive via IVF-ET, while the remaining women have identical and constant per cycle probabilities of conception. This model produces estimates that are highly consistent with the data, and suggests that continuing treatment beyond some threshold number of cycles is not efficacious. Recognizing this, the authors determine cutoffs beyond which treatment should not continue for IVF-ET programs with fixed capacities. They also consider cutoff policies where program participants may belong to one of several different split populations, detailing the case of two groups. Finally, the authors show how one may reduce the average time in treatment (including waiting time) considerably with minimal impact on the probability of achieving pregnancy.
