Date of Award

Fall 8-2020

Degree Name

Doctor of Philosophy (PhD)


Ruosha Li

Second Advisor

Jing Ning

Third Advisor

Han Chen


Incorporating promising biomarkers to improve risk assessment and prediction is the central goal in many biomedical studies. Cost-effective designs and longitudinal designs are often utilized for measuring biomarker information, but they pose challenges to the data analyses. Statistical analyses for these kinds of data are routinely performed using parametric models. When the model assumptions are violated, parametric models may lead to substantial bias in parameter estimation, risk evaluation and prediction. In this dissertation, we will develop robust, exible statistical methods for risk assessment for matched case-control, nested case-control, and case-cohort designs, as well as a dynamic prediction tool for longitudinal data. In the first aim, we will develop a distribution-free method for identifying an optimal combination of biomarkers to differentiate cases and controls in matched case-control data. In the second aim, we will develop a semiparametric regression model with minimal assumptions on the link function for data from two-phase sampling designs with binary outcomes. In the third aim, we will develop a model-free dynamic prediction method for a survival outcome that provides dynamically updated risk scores using longitudinal biomarker(s).