Analysis of Marked Recurrent Events in the Presence of a Terminating Event.

Recurrent Event;Marks;Terminating Event;Inverse Weighting;Public Health;Health Sciences;Biostatistics

Ma, YuSen, Ananda ;

University of Michigan

关键词: Recurrent Event;    Marks;    Terminating Event;    Inverse Weighting;    Public Health;    Health Sciences;    Biostatistics;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/96129/rickma_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

In many biomedical studies where the event of interest is recurrent (e.g., hospital admission), marks are observed upon the occurrence of each event (e.g., medical costs, length of stay). In Chapter II, we propose novel methods which contrast group-specific cumulative means, influenced by the recurrent event rate and survival probability. Our proposed methods utilize a form of hierarchical modeling: a proportional hazards model for the terminating event; a proportional rates model for the conditional recurrent event rate given survival; and a generalized estimating equations approach for the marks, given an event has occurred. Group-specific cumulative means are estimated (as processes over time) by averaging fitted values from the afore-listed models, with the averaging being with respect to the marginal covariate distribution. Large sample properties are derived, while simulation studies are conducted to assess finite sample properties. We apply the proposed methods to data obtained from the CANADA-USA Peritoneal Dialysis Study. Typically in observational studies, it is necessary to account for measured and un-measured heterogeneity across study subjects. This is often accomplished through model covariates (for measured factors) and frailty variates (to account for unmeasured predictors). In Chapter III, we propose fully parametric models and estimation is carried out simultaneously. Residual correlation across the terminating event, recurrent event and mark process is captured by a Normal frailty variate. Maximum likelihood based estimation is carried out via a Gaussian Quadrature technique for integration. Through simulation, the methods are shown to work well for practical sample sizes. In Chapter IV, we develop inverse weighting methods to contrast group-specific cumulative means. Both the underlying data structure and the target estimands are the same as those from Chapter II. However, we avoid constructing semi-parametric or parametric models for each process to achieve consistent estimates in Chapter IV. We further take into account of treatment imbalance and unobserved censoring times by combining Inverse Probability Treatment Weighting (IPTW) and Inverse Probability Censoring Weighting (IPCW). Efficiency is compared with the procedure proposed in Chapter II.