Advisory Committee Chair
Gary R Cutter
Advisory Committee Members
Charity J Morgan
Meredith Kilgore
David Naftel
Jeffery Szychowski
Document Type
Dissertation
Date of Award
2017
Degree Name by School
Doctor of Philosophy (PhD) School of Public Health
Abstract
Unobserved individual heterogeneity, that is frailty, is a major concern in the application of analysis for competing risks data. Unlike the study of clustered competing risks data using multivariate frailty models, we have not found a study of non-clustered competing risks data using univariate frailty model. This dissertation focuses on developing frailty models for the analysis of non-clustered competing risks data. The estimation of the effects of covariates by a parametric univariate frailty model could be a useful approach when heterogeneity must be considered in the non-clustered competing risks setting. In paper 1, we compared the performance of the Cox, Exponential and Fine-Gray models for the estimation of a HR. The data were simulated using both latent failure time and bivariate random variable approaches. No differences between these models were observed in power, type I error and standard error. However, the Fine-Gray model showed greater bias and lower interval coverage compared to the other two models. Both Cox and exponential models perform better for the estimation of the HRs as compared to the Fine-Gray model. However, the Fine-Gray model gives unbiased estimates of the time-averaged effect. In paper 2, we provide a PUFM for non-clustered competing risks data. Assuming different events may have the same frailty due to unobserved covariates from the same cause, the PUFM is a cause-specific proportional hazards frailty model. A sensitivity analysis of Cox PH model, Weibull AFT model and proposed PUFM for estimating the covariate coefficients of interest event shows that the proposed method performs very well for competing risks data with an assumption of an exponential distributed time and gamma, lognormal or Weibull distributed frailties. In paper 3, to address the heterogeneity in the individuals enrolled in the NARCOMS registry and to describe the cause-specific mortality among the participants in this registry, we extend the PUFM to estimate the effect of age and gender on death due to different causes. The results showed that, considering the frailty among the NARCOMS participants, older age in the study was associated with higher mortality risk from MS and infection compared to risks from other causes.
Recommended Citation
Liu, Yuliang, "Univariate Frailty Model For Competing Risks Data Analysis" (2017). All ETDs from UAB. 2316.
https://digitalcommons.library.uab.edu/etd-collection/2316