Statistical evaluation of surrogate markers in randomized clinical trials

Thesis (Ph.D.)--Boston University === PLEASE NOTE: Boston University Libraries did not receive an Authorization To Manage form for this thesis or dissertation. It is therefore not openly accessible, though it may be available by request. If you are the author or principal advisor of this work and wo...

Full description

Bibliographic Details
Main Author: Miao, Xiaopeng
Language:en_US
Published: Boston University 2018
Online Access:https://hdl.handle.net/2144/31591
Description
Summary:Thesis (Ph.D.)--Boston University === PLEASE NOTE: Boston University Libraries did not receive an Authorization To Manage form for this thesis or dissertation. It is therefore not openly accessible, though it may be available by request. If you are the author or principal advisor of this work and would like to request open access for it, please contact us at open-help@bu.edu. Thank you. === In many randomized clinical trials, the primary endpoints are clinical measurements of disease process. It usually requires an extremely long follow-up period and a considerable sample size to assess the treatment effect on such clinical endpoints. Therefore surrogate markers that can predict the treatment effects on the clinical endpoints would be extremely useful in accelerating the drug development process and depicting the mechanisms of drug action. Although candidate surrogate markers are generally proposed based on biological considerations, their validations largely depends on statistical methods. There are two major statistical frameworks of evaluating candidate surrogate markers in a single-trial setting: one is called the statistical surrogate (Prentice, 1989) and the other is referred to as the principal surrogate (Frangakis and Rubin, 2002). Both frameworks define surrogacy based on the treatment effect on the clinical endpoint that is mediated through the surrogate marker. For the evaluation of statistical surrogates, most existing methods are based on parametric regression analyses, which might provide spurious interference in the presence of model misspecification. For the evaluation of principal surrogates, the applications of existing methods are restricted to some simplified contexts (e.g., HIV vaccine trials) or limited types of clinical endpoints (e.g., binary). In this dissertation, we develop two novel approaches for the evaluation of surrogate markers in randomized clinical trials. In the framework of statistical surrogacy, we develop a nonparametic testing procedure based on measure of divergence and random permutation. The proposed method is robust to model misspecification and influential points, and is applicable to a variety of setting. In the framework of principal surrogacy, we propose a multiple imputation approach based on the incorporation of baseline predictors. The proposed approach can accomodate different types of clinical endpoints, and can be used to evaluate principal surrogates in a general setting where most existing methods are not applicable. Extensive simulation studies are conducted to examine the performance of the proposed methods. The usefulness of the method is further illustrated by real examples in clinical trials. === 2031-01-01