Bayesian analysis of multi-type recurrent events with dependent termination
In clinical and epidemiological studies, recurrent events occur frequently, such as such as repeated lung infections in people with cystic fibrosis, recurrent shunt failures in children with hydrocephalus, and recurrent strokes in older adults. An important feature of recurrent events is that the event times are correlated, including within-subject correlation, event-specific dependence, between event-type correlation, etc. And, a scientific objective is to examine covariate effects on the risks of recurrent events. In this dissertation, we develop two Bayesian regression models to analyze the recurrent event data where the mutual correlations have been accounted for. We first consider two types of correlation among recurrent event times which is subject-specific heterogeneity and event-specific dependence. Subject-specific heterogeneity represents unmeasured variables which induce the within-subject correlation among event times. Alternatively, correlation may be induced by recurrent event processes where the incidence of events may elevate or reduce the risk of future recurrent events. The methods developed in this these can inform researchers how the risk of future disease events and the covariate effects change as the number of past events accumulate. The proposed model has been compared with several commonly used models with Monte Carlo simulations and applied to the motivating lipid-lowering trial (LLT) component of the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT) (ALLHAT-LLT). In the context of multi-type recurrent event, we propose a model to jointly analyze the multi-type recurrent events and dependent terminal event with nonparametric covariate functions. The recurrent events and dependent terminal event are linked together via shared random effects, which represent the subject-specific heterogeneity in the hazard functions. It is very common that the true underlying covariate effects are nonlinear, rather than simple linear function. For example, age is the largest risk factor for cardiovascular diseases and researchers are interested to explore the functional form of age on the hazard of cardiovascular diseases. Extensive simulation results suggest that misspecifying nonparametric covariate functions may introduce bias in parameter estimation. This method development has been motivated by and applied to the lipid-lowering trial (LLT) component of the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT).
Lin, Li-An, "Bayesian analysis of multi-type recurrent events with dependent termination" (2015). Texas Medical Center Dissertations (via ProQuest). AAI10027730.