The analysis of clustered survival data with dependent censoring.

In survival studies with right censored failure times, it is common that censoring is correlated with failure. When subjects in the study are independent of each other, the correlation between failure and censoring is not identifiable. However, this thesis shows that, when subjects in the study are clustered and subjects in the same cluster are similar, the observed replication of correlated survival and censoring times among subjects within each cluster can be used to identify dependent censoring. This is implemented by a frailty model with informative censoring. In this model, the censoring and failure risks share a common component at the cluster level. This model allows flexibility in the direction and degree of dependence between failure and censoring. It includes the survival frailty model with independent censoring as a special case. It allows censoring due to several causes to be analyzed as informative, while treating censoring due to other causes as non-informative. It can also be used to analyze competing risks data. To fit the model, the EM algorithm is used with Markov Chain Monte Carlo simulations in the E-steps. Simulation studies and analysis of data for kidney disease patients are provided. Similar to other models for dependent censoring, the proposed frailty model for informative censoring is computationally intensive. If censoring were independent of failure, such complicated models need not be used. Therefore, a simple test of independent censoring is desirable to determine the significance of the more complicated dependent model. This thesis proposes such a test using martingale residuals. Simulation studies evaluate the performance of the test under different circumstances. Various other aspects of the proposed model are also considered. Consequences of incorrectly assuming non-informative censoring are investigated. Sensitivity analysis for model assumption is conducted. Related topics and future work are discussed.PhDBiological SciencesBiostatisticsUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/132055/2/3057966.pd