In the controlled ovarian hyperstimulation (COH) cycle of the in vitro fertilization-embryo transfer (IVF-ET) therapy, the clinicians observe the patients' responses to gonadotropin dosages through closely monitoring their physiological states, to balance the trade-off between pregnancy rate and ovarian hyperstimulation syndrome (OHSS) risk. In this paper, we model the clinical practice in the COH treatment cycle as a stochastic dynamic program, to capture the dynamic decision process and to account for each individual patient's stochastic responses to gonadotropin administration. We discretize the problem into a Markov decision process and solve it using a slightly modified backward dynamic programming algorithm. We then evaluate the policies using simulation and explore the impact of patient misclassification. More specifically, we focus on patients with polycystic ovary syndrome (PCOS) or potential, that is, the patients that tend to be more sensitive to gonadotropin administration.
