| Stochastic processes are important for modeling transitions among multiple events in natural history data. These multiple events can be the same, different, or mixed types. For instance, an end-stage renal disease (ESRD) patient may be on dialysis or get transplanted, and dynamic changes of these therapies occur over time. Analyzing dialysis and transplant survival separately, commonly used for ESRD data, is biased and inefficient.;The proposed model was applied to ESRD data, where dialysis death and receiving transplantation are two negatively dependent competing events. We found that age and vascular disease are positively associated with patient mortality and negatively correlated with the likelihood of receiving transplantation. This confirms the negative dependence of these two events. Our findings agree well with clinical observations and provide important insights on issues of access and waiting times for kidney transplantation.;The purpose of this research was to develop stochastic models and analytical methods for multiple event data, with applications to ESRD data. We investigated appropriateness of Markov, semi-Markov and marginal modeling approaches for ESRD data. Furthermore, we evaluated methods for comparing ESRD patient survival on different modalities. Within the multistate framework, one particular situation of interest was that there exist two dependent competing events in a population, where a proportion of subjects are "immune" to one event, and the remaining subjects are at risk for both events. We proposed a family of dependent competing risks model for such latent heterogeneous structure, through a bivariate frailty distribution... |
