Intermediate markers: Surrogacy assessment using principal stratification and multi-state models
Year: 2013
Type: dissertation
Abstract: Intermediate markers can be useful in clinical trials as either surrogate markers intended to replace the true outcome of interest or as auxiliary variables intended to improve efficiency in the analysis of the true outcome. We explore methods pertaining to both of these roles of intermediate markers. First, we propose methods for assessing the validity of a potential surrogate marker. Working under the principal stratication approach for surrogacy validation proposed by Frangakis and Rubin (2002), we propose quantities to evaluate surrogacy when the joint distribution of the potential surrogate and final outcomes is multivariate normal. The multivariate normality assumption is then relaxed and a Gaussian copula model is used to model the joint distribution of surrogate and final outcomes, and quantities are derived from this model to determine surrogacy. For both the multivariate normal model and the Gaussian copula model, a Bayesian estimation strategy is used and, as some parameters are not identifiable from the data, we explore the use of informative priors that are consistent with reasonable assumptions in the surrogate marker setting to aid in estimation. Methods for utilizing an intermediate marker as an auxiliary variable to improve efficiency in the analysis of the true outcome are then considered. A multi-state model with an incorporated cured fraction is used to model recurrence and death in colon cancer. The model is used to assess how individual covariates affect the probability of being cured of disease and the transition rates between the various disease states. Once parameter estimates from the model are obtained, survival probabilities can be estimated with gains in efficiency obtained as compared to Kaplan-Meier estimates. The model is then used in a multiple imputation strategy which imputes death times for censored subjects. By using the joint model, recurrence is used as an auxiliary variable in predicting survival times. We explore the use of a hierarchical model and model adaptations that can be made to potentially further the efficiency gains obtained through the multiple imputation procedure. We demonstrate the potential use of the proposed methods in shorting the length of a trial and reducing sample sizes.
Author: A. S. C. Conlon
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W105672414