Graduation Year

2022

Document Type

Dissertation

Degree

Ph.D.

Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Public Health

Major Professor

Janice Zgibor, RPh, Ph.D.

Co-Major Professor

Jongphil Kim, Ph.D.

Committee Member

Ellen Daley, Ph.D.

Committee Member

Jason Beckstead, Ph.D.

Committee Member

Amy Alman, Ph.D.

Keywords

Bayesian Inference, Multilevel Item Response Theory Model, Multivariate Skew Distribution, Nonlinear Mixed-Effects Joint Models, Partially Linear Mixed-Effects Models, Longitudinal-Survival Data

Abstract

Clinical trials have tended to collect both survival information and longitudinal biomarkers, as well as other covariates. In order to better assess the severity of diverse diseases, we need to collect various longitudinal outcomes. Furthermore, longitudinal data could consist of a number of different measurements of varying types. The multilevel item response theory (MLIRT) model has been widely used in several fields such as public health and health sciences for multivariate longitudinal outcomes. Joint models combining the longitudinal and survival processes, as well as the relation between them, have been developed to minimize bias and improve the efficiency of estimates. The majority of joint models, however, focus on a single longitudinal outcome. Additionally, the assumption of normality may also produce inaccurate results if skewness is present for continuous outcomes. Furthermore, the trajectories are not linear and are observed in a nonlinear manner as we normally observe in time. This means that partial linear regression is more appropriate to consider. Challenges remain in the interpretation of such complicated and long-term survival data due to data attributes, including a mix of longitudinal outcomes, measurement errors, and skewness. Ignoring these characteristics of the data could lead to biased conclusions being drawn.

We are aware of only a few studies investigating the joint models based on semi-parametric MLIRT models, and applying them to longitudinal clinical trial data even when they have multiple features. Therefore, the following two MLIRT-based models were developed in this dissertation by using a Bayesian approach, including: (i) MLIRT-based partially linear mixed-effects models for longitudinal data with heterogeneous characteristics; (ii) MLIRT-based partially linear mixed-effects joint models for longitudinal and time-to-event data with heterogeneous characteristics. This dissertation presents semi-parametric MLIRT-based joint models which are applied to analyze the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial. A joint model based on MLIRT was developed for analyzing the data on Primary Biliary Cirrhosis (PBC). Moreover, simulation studies were also conducted to evaluate the performance of the proposed methods under different scenarios. In spite of this study being a biostatistical methodology study, we also discovered several clinical findings that are noteworthy.

Download

Share

COinS