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

Bayesian hierarchical joint modeling of repeatedly measured mixed biomarkers of disease severity and time-to-event

Buhule, Olive D. (2014) Bayesian hierarchical joint modeling of repeatedly measured mixed biomarkers of disease severity and time-to-event. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

Submitted Version

Download (1MB) | Preview


In many clinical follow-up studies, patients are observed at irregular intervals for more than one biomarker of disease severity. Although these biomarkers are often meant to measure the same disease severity, they may differ due to the instruments or reagents used as well as the scale of measurements. They could show different patterns for treatment because clinicians prescribe medications based on the severity of disease. Moreover, if these markers
are modeled separately to determine the factors that are associated with disease progression over time or to predict the event of interest given different treatments, they may yield misleading or inefficient results. Joint modeling of correlated biomarkers alone or with time-to-event data leads to efficient results, hence better clinical decisions.

In this study, we have first developed a joint model to analyze multivariate unbalanced repeatedly measured outcomes of mixed types, in particular, continuous and ordinal outcomes. Secondly, we have extended the first model to include time-to-event data. The postulated models assumes that the outcomes are from distributions that are in the exponential family and hence modeled as a multivariate generalized linear mixed effects model linked through random effects. The Markov Chain Monte Carlo (MCMC) Bayesian approach is used to approximate the posterior distribution and draw inference on the parameters. These joint models provide a flexible framework to account for the hierarchical structure of the highly unbalanced data as well as the association between the multiple mixed types of outcomes and time-to-event. Moreover, the simulation studies show that estimates obtained from the joint models are consistently less biased and more efficient than those obtained from the separate models. We applied our models to diabetes data from an observational study.

Diabetes and its associated complications such as heart attack and stroke are of serious public health concerns across the globe. Proper treatment can help control and prevent the development of these complications and hence improve the quality of life of millions of people.
This work proposes to efficiently estimate the treatment effect by introducing state-of-the-art statistical methods. This will help researchers identify effective treatments that can slow down the disease progression.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Buhule, Olive D.odb3@pitt.eduODB3
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairYouk, Ada O.youk@pitt.eduYOUK
Committee CoChairWahed, Abdus S.wahed@pitt.eduWAHED
Committee MemberArena, Vincent C.arena@pitt.eduARENA
Committee MemberWeeks, Daniel E.weeks@pitt.eduWEEKS0000-0001-9410-7228
Date: 29 September 2014
Date Type: Publication
Defense Date: 28 May 2014
Approval Date: 29 September 2014
Submission Date: 6 June 2014
Access Restriction: 5 year -- Restrict access to University of Pittsburgh for a period of 5 years.
Number of Pages: 171
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: Diabetes; Generalized Linear Mixed Effects Models; Hierarchical Modeling; Joint Modeling; Mixed Biomarkers; MCMC; Time-to-Event
Date Deposited: 29 Sep 2014 19:47
Last Modified: 30 Jun 2022 15:53


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item