| Population-based studies and clinical trials provide many interesting methodological problems that render important policy implications as well as better explainations of disease progression processes. This thesis is to answer three such questions. Trends in United States cancer survival motivated a statistical method for survival data that may be subject to dependent censoring in disease populations that may contain a portion of long-term cancer surviors. Prostate cancer trends motivated work on a survival model for populations that may have long-term survivors and that exhibit a change-point effect in important covariates or predictors. Finally, a clinical trial on childhood acute lymphoblastic leukemia motivated work on a survival model for clustered data that explicitly models the correlation of failure times but also allows for population-level interpretation of survival parameters. |
