Abstract:
|
In toxicological studies, data on female rodent reproductive cycling are often used as an outcome variable. For example, some National Toxicology Program (NTP) studies routinely collect vaginal cytology slides from rats and mice, which can be used to determine daily estrous cycle stage. Analyses are typically based on summary statistics of daily data, including proportion of animals regularly cycling, average cycle length, proportion of time animals spend in each estrous stage, and number of cycles in a fixed period of time. We propose a first order non-stationary Markov chain model with transition probabilities depending on unobserved stage lengths. Estrous cycle data have four distinguishable stages: D, P, E, and M. The set of daily status (in one of these four stages) is the state space of our Markov process. We assume stage lengths are independent. A Bayesian approach is used for inference on dose effects on mean stage lengths. Gibbs sampler (implemented using BUGs software) is used to fit the model.
|