Recurrent events are frequently of interest in observational studies, and serve as a useful means for evaluating treatments.When treatment is initiated after the start of follow-up, traditional methods may be biased or give undesirable interpretations.The methods proposed in this dissertation aim to address these issues, and are applied to data from the Adult-to-Adult Living Donor Liver Transplantation Cohort Study.In the second chapter we propose to estimate the effect of a rare treatment on the recurrent rate using a 2-stage modeling approach and an extension of the method of sequential stratification.In the first stage, we model the pre-treatment recurrent event rate, adjusted for time-dependent predictors including the event history.Using results from this model we calculate a prognostic score for each patient and match each treated patient to similar yet-untreated controls.Within each stratum patients are followed from the index patient’s time of treatment initiation.Within each matched set control subjects are censored if they subsequently receive treatment, which can induce dependent censoring when treatment is not rare.In the third chapter we extend the method to accommodate frequently assigned treatments by introducing a variation of Inverse Probability of Censoring Weighting (IPCW).We also introduce multiple treatment states, with the comparison of interest being between the experimental and conventional courses of treatment.In the fourth chapter we consider the setting where the recurrent event process is potentially stopped by a correlated terminal event.We use a frailty model to estimate prognostic scores for each patient, and match based on this score as well as the estimated frailty.Sequential stratification is again used to fit final models for the recurrent event rate and terminal event hazard, adjusted for residual distance in the prognostic score and estimated frailty.
【 预 览 】
附件列表
Files
Size
Format
View
Sequential Stratification Methods for Estimating Effects of Time-Dependent Treatments on Multivariate Survival Outcomes.