Off-campus Michigan Tech users: To download campus access theses or dissertations, please use the following button to log in with your Michigan Tech ID and password: log in to proxy server

Non-Michigan Tech users: Please talk to your librarian about requesting this thesis or dissertation through interlibrary loan.

Date of Award

2026

Document Type

Campus Access Dissertation

Degree Name

Doctor of Philosophy in Statistics (PhD)

Administrative Home Department

Department of Mathematical Sciences

Advisor 1

Kui Zhang

Committee Member 1

Qiuying Sha

Committee Member 2

Xiao Zhang

Committee Member 3

Hairong Wei

Committee Member 4

Weihua Zhou

Abstract

This dissertation consists of three complimentary methodological projects for analyzing heterogeneous biomedical data. First, it proposes a method for the joint detection of shared and condition-specific hub genes in gene regulatory networks, Second, it develops a two-stage hip fracture risk prediction Approach that integrates initial screening and Dual-energy X-ray Absorptiometry (DXA)-based refinement to improve prediction accuracy. Third, it introduces and evaluates domain adaptation methods to enhance the generalizability of hip fracture prediction across diverse cohorts. Together, these contributions advance robust statistical learning approaches for heterogeneous biomedical data settings.

Project 1 proposes the use of the stability selection for the joint detection of shared and condition-specific hub genes across multiple gene regulatory networks. Extensive simulation studies demonstrate that the proposed method achieves high sensitivity in detecting both shared and condition-specific hub genes while maintaining strong control of false-positives.

Project 2 presents a novel sequential two-stage model for predicting hip fracture risk, leveraging data from the initial screening and follow up data from Dual-energy X-ray Absorptiometry (DXA). In Stage 1, clinical and functional variables are used for screening, while Stage 2 incorporates DXA-derived imaging information to refine predictions for individuals with uncertain risk profiles. This stepwise approach improves sensitivity compared with conventional tools such as the T-score and the Fracture Risk Assessment Tool (FRAX).

Project 3 addresses the domain-shift problem in clinical risk prediction by aligning source and target cohorts through unsupervised domain adaptation methods. It evaluates methods including Maximum Mean Discrepancy (MMD), Correlation Alignment (CORAL), Domain-Adversarial Neural Network (DANN), and their combinations under outcome-free model selection. Results demonstrate that domain-adaptation methods consistently improve predictive performance, with the combined MMD + CORAL + DANN approach achieving the highest area under the curve (AUC) of 0.96 for females and 0.88 for males.

Share

COinS