|
Abstract:
|
In animal carcinogencity studies, tumors are only observable at time of natural death or terminal sacrifice. Thus, time-to-tumor is subject to current status observation. When the observation time is independent of the event time, current status data can be analyzed using standard methods for interval censored data. However, time-to-death is likely related to time-to-tumor making this assumption unreasonable. To further complicate matters, multiple types of tumors are observed and analysis models should account for the relationships between these tumors to avoid potential biases. In this presentation, we will present a Bayesian method which models the times to each type of tumor using latent Wiener processes which fail when they hit a threshold for the first time. We incorporate shared random effects into the drifts of the latent processes to account for relationships in the tumor risks. Informative observation time is accounted for by modeling time-to-death using a latent Wiener process whose time scale is affected by the occurrence of each tumor.
|
