Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


42858.pdf (7.71 MB)

ETD Abstract Container

Abstract Header

Univariate and Multivariate Joint Models with Flexible Covariance Structures for Dynamic Prediction of Longitudinal and Time-to-event Data.

Palipana, Anushka
http://orcid.org/0000-0001-5237-1397
http://rave.ohiolink.edu/etdc/view?acc_num=ucin1657796561545003

Abstract Details

2022, PhD, University of Cincinnati, Arts and Sciences: Mathematical Sciences.
Joint modeling of noisily measured biomarkers alongside time-to-event outcomes, such as those used for evaluating disease progression over time and dynamic prediction of survival, has revolutionized statistical science. It is also becoming increasingly clear with the advent of non-stationary Gaussian processes applied to medical monitoring studies that typical random effects within a longitudinal sub-model do not properly reflect complex fluctuations in biological processes. In this body of work, we propose a novel, flexible five-component longitudinal sub-model for univariate and multivariate joint modeling frameworks. The most noteworthy development is our introduction in these modeling frameworks of scaled integrated fractional Brownian Brownian motion (IFBM), a more generalized version of the integrated Brownian motion stochastic process that has been demonstrated to accurately represent biological processes observed with uncertainty. While integrated Brownian motion has been investigated as a method for developing target functions capable of properly predicting threshold changes in disease progression markers, no applications have used joint longitudinal-survival modeling or assessed the utility of IFBM for such purposes. As the event sub-model, the Cox proportional hazards model is used. For Bayesian posterior calculation and inference, we employ Markov chain Monte Carlo (MCMC) techniques. Using data from national patient registries, we use this novel method to assess lung function trajectories and mortality in two distinct rare lung diseases-lymphangioleiomyomatosis and cystic fibrosis. We investigate clinically important target functions with the goal of predicting rapid lung function decline in each disease, including environmental exposures and community characteristics in the application of cystic fibrosis. Each application includes a comparison of IFBM, with literature-based integrated Ornstein-Uhlenbeck (IOU), and random intercepts and slopes modeling. Numerous longitudinal outcomes are collected in order to conduct a thorough investigation of the multidimensional impairment caused by a disease. When a patient's lung function rapidly deteriorates, it alerts health care professionals to monitor the patient's illness development and make informed and timely medical decisions, such as preparing the patient for lung transplantation. Cystic fibrosis is a multisystem condition. As a result, we present a multivariate joint model that employs a flexible covariance structure to effectively capture the highly variable nature of these longitudinal processes while modeling lung function, and growth simultaneously. This model is expected to improve the prediction accuracy of survival probability when compared to a univariate joint model that employs random-slopes random-intercepts model to depict the longitudinal process. Along with developing a dynamic prediction framework for predicting the future outcome trajectories and predictive probabilities of CF patients using clinically relevant target functions, we also developed a method for calculating dynamic event-free probabilities using multivariate longitudinal data. Our suggested model is assessed using simulations and is applied to the US Cystic Fibrosis Foundation–Patient Registry (CFF-PR) data set.
Seongho Song, Ph.D. (Committee Member)
Xuan Cao (Committee Member)
Xia Wang, Ph.D. (Committee Member)
Rhonda Szczesniak, Ph.D. (Committee Member)
Siva Sivaganesan, Ph.D. (Committee Member)
120 p.
Statistics
Target functions; Medical Monitoring; Real-time prediction; Univariate Joint Models; Multivariate Joint Models; Integrated Fractional Brownian Motion

Recommended Citations

Citations

  • Palipana, A. (2022). Univariate and Multivariate Joint Models with Flexible Covariance Structures for Dynamic Prediction of Longitudinal and Time-to-event Data. [Doctoral dissertation, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1657796561545003

    APA Style (7th edition)

  • Palipana, Anushka. Univariate and Multivariate Joint Models with Flexible Covariance Structures for Dynamic Prediction of Longitudinal and Time-to-event Data. 2022. University of Cincinnati, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1657796561545003.

    MLA Style (8th edition)

  • Palipana, Anushka. "Univariate and Multivariate Joint Models with Flexible Covariance Structures for Dynamic Prediction of Longitudinal and Time-to-event Data." Doctoral dissertation, University of Cincinnati, 2022. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1657796561545003

    Chicago Manual of Style (17th edition)

ucin1657796561545003
65
© 2022, all rights reserved.
This open access ETD is published by University of Cincinnati and OhioLINK.
Release 3.2.12