First Advisor

Lalonde, Trent

Second Advisor

Schaffer, Jay

Third Advisor

Schaffer, Jay

Document Type

Dissertation

Date Created

12-9-2015

Abstract

Arbitrarily interval-censored survival data refer to the situation where the exact time of the occurrence of an event of interest is only known to have occurred within some two consecutive examinations. External time-dependent covariates refer to those whose values change during the periodic follow-up, and whose value at a particular time does not require individuals to be under direct observation. Regression modeling of survival data usually either handles arbitrarily interval-censored data alone (Farrington, 1996) or external time-dependent covariates alone (Cox, 1972; Therneau & Grambsch, 2000). In the current research, an adjustment has been made to the data augmentation used in Farrington’s estimation method for arbitrarily interval-censored data to accommodate external time-dependent covariates. The three approaches, regression analysis of arbitrarily interval-censored survival data by Farrington (1996), the extended Cox model (Cox, 1972; Therneau & Grambsch, 2000) for handling external time-dependent covariates, and the proposed model for handling both arbitrarily interval-censored data and external time-dependent covariates, were compared in terms of hypothesis testing performance. The simulation results revealed that the proposed model was more powerful than the other two models, and the type I error rate from the proposed model fluctuated around the nominal level .05, and was comparable to that from the extended Cox model. Moreover, the proposed model gave the smallest absolute relative bias of parameter estimates, and always gave the correct direction of the effect from the significant external time-dependent covariate. As such, the proposed model depicted the survival experience of subjects regarding the timing of the occurrence of an event more realistically. According to the results of the current research, the proposed model can be used in practice as an alternative to the popular extended Cox model (Cox, 1972; Therneau & Grambsch, 2000) for investigating what factors influence the survival times of subjects.

Keywords

Statistics

Extent

246 pages

Local Identifiers

Fang_unco_0161D_10453.pdf

Rights Statement

Copyright is held by the author.

Share

COinS