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

Joint Modeling of Time-to-Event Data with Competing Risks

Fu, Bo (2013) Joint Modeling of Time-to-Event Data with Competing Risks. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

Primary Text

Download (447kB) | Preview


When analyzing time-to-event data, informative dropout due to competing risks is one prac- tical aspect that researchers should take into account. If we fail to account for the association between the event of interest and informative dropouts, unknown amplitude bias may be en- countered when identifying the effects of potential risk factors related to time to the main cause of failure. A joint modeling approach of time to the main event and time to the competing events is proposed, to capture the dependence between the main event and the informative dropout due to competing risks via a set of random terms. Two fundamental likelihood functions with different structures of the random terms are provided, which may be combined in practice. We used three methods to optimize the corresponding likelihood functions in order to estimate the unknown covariate effects: Gaussian quadrature method, Bayesian Markov Chain Monte Carlo method, and hierarchical likelihood method. Four bias reduction correction methods for the h-likelihood estimation approach are explored. These methods were aimed to improve the accuracy of parameter estimation. The performances of the three methods were compared via simulations. We applied proposed methods to identify risk factors for dementia.

Time-to-event data have been widely investigated from clinical trials and from obser- vational studies. The proposed joint modeling method is significantly meaningful to public health research because informative dropout commonly exists for the time-to-event data. Methods that have been currently used either fail to adjust for the association between the main event and the informative dropout due to competing events or the methods used to adjust for the association are not easy to implement. In this dissertation, we showed that the proposed joint modeling approach provides less bias estimates on the effect of a risk factor and has fairly straightforward implementation, which will lead to benefits for medical research.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Fu, Bobof5@pitt.eduBOF5
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairChang, Chung Chouchangj@pitt.eduCHANGJ
Committee MemberWeissfeld, Lisa Alweis@pitt.eduLWEIS
Committee MemberGanguli, MaryGanguliM@upmc.eduGANGULIM
Committee MemberPike, Francisfrp3@pitt.eduFRP3
Committee MemberLi, Ruosharul12@pitt.eduRUL12
Date: 27 September 2013
Date Type: Publication
Defense Date: 16 April 2013
Approval Date: 27 September 2013
Submission Date: 3 April 2013
Access Restriction: 5 year -- Restrict access to University of Pittsburgh for a period of 5 years.
Number of Pages: 66
Institution: University of Pittsburgh
Schools and Programs: School of Public Health > Biostatistics
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: Bayesian-Markov chain Monte Carlo; Competing risks; Gaussian quadrature; Hierarchical likelihood; Informative dropout; Joint modeling; Bias reduction
Date Deposited: 27 Sep 2013 14:42
Last Modified: 01 Sep 2018 05:15


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item