Sequential predictions of menstrual cycle lengths.

Forecasting the length of the menstrual cycle and of its phases is an important problem in infertility management and natural family planning. Using repeated measurements of the length of the entire cycle and of the preovular phase provided by a large English database, we describe a Bayesian hierarchical dynamic approach to the problem. A state-space process is used to model the temporal behavior of the series of lengths for each woman. The individual processes are then embedded into a multivariate system through a Bayesian hierarchy in which model parameters are allowed to vary across subjects according to a specified probability distribution. The most interesting features of the suggested method are (a) it takes into account explicitly the temporal nature of the available data and (b) if combined with a fecundability model, it can be used to forecast the probability of conception in future cycles as a function of any intercourse behavior.