Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

Semiparametric Estimators in Competing Risks Regression

Yabes, Jonathan (2012) Semiparametric Estimators in Competing Risks Regression. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

[img]
Preview
PDF
Primary Text

Download (570kB) | Preview

Abstract

Clinical trials and cohort studies that collect survival data frequently involve patients who may fail from one of multiple causes (failure types). These causes are called competing risks. The cumulative incidence function (CIF), or subdistribution, is a commonly reported quantity that describes the crude failure type-specific probability of the study population. The proportional subdistribution hazards model has been widely applied to study the effects of covariates on the CIF. In practice however, the time of failure may be recorded but the cause may be unknown or missing. To avoid bias, we developed two semiparametric estimators of covariate effects: the inverse probability weighted (IPW) estimator and the augmented inverse probability weighted (AIPW) estimator. We showed that these estimators are consistent and asymptotically normal. Their finite sample size properties and robustness were demonstrated through simulations. In many situations, investigators are interested in the marginal survival distribution of latent failure times, rather than the CIF. Because of the identifiability problem in competing risks, we derived an estimator of covariate effects in the Cox proportional hazards model by incorporating the random signs censoring (RSC) principle, which assumes that the main event failure time is independent of the indicator that the main event precedes the competing event. Unlike identifying assumptions that are typically imposed in practice, RSC is verifiable via stochastic ordering in the observed data. We further relaxed the RSC assumption by positing that independence is achieved conditional on some covariates. We showed that the resulting estimator is not only easy to implement but also has desirable asymptotic properties. We evaluated the estimator's finite sample size performance through simulations. Medical datasets were used to illustrate the proposed methods.
Public Health Significance: Biomedical and public health studies with time-to-event endpoint are abundant and often influence regulatory decisions. Trustworthiness of the research results not only relies on the design quality, but also on the soundness of the analytical approach used. The methodologies we propose account for two potential sources of bias in the conduct of such studies -- competing risks and missing data.


Share

Citation/Export:
Social Networking:
Share |

Details

Item Type: University of Pittsburgh ETD
Status: Unpublished
Creators/Authors:
CreatorsEmailPitt UsernameORCID
Yabes, Jonathanjgyabes@gmail.com
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairChang, Chung-Chou H.changj@pitt.eduCHANGJ
Committee MemberJeong, Jong-Hyeonjeong@nsabp.pitt.eduJJEONG
Committee MemberTang, Gonggot1@pitt.eduGOT1
Committee MemberUnruh, Markmarkunruh@gmail.com
Committee MemberWeissfeld, Lisalweis@pitt.eduLWEIS
Date: 24 September 2012
Date Type: Completion
Defense Date: 25 July 2012
Approval Date: 24 September 2012
Submission Date: 23 July 2012
Access Restriction: 3 year -- Restrict access to University of Pittsburgh for a period of 3 years.
Number of Pages: 81
Institution: University of Pittsburgh
Schools and Programs: Graduate School of Public Health > Biostatistics
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: Competing risks; cumulative incidence function; doubly-robust; inverse probability weighting; missing cause of failure; marginal survival function; proportional subdistribution hazards; random signs censoring.
Date Deposited: 24 Sep 2012 19:01
Last Modified: 15 Nov 2016 14:00
URI: http://d-scholarship.pitt.edu/id/eprint/13043

Metrics

Monthly Views for the past 3 years

Plum Analytics


Actions (login required)

View Item View Item