TY - JOUR TI - Two-stage clinical trial designs with survival outcomes and adjustment for misclassification in predictive biomarkers DO - https://doi.org/doi:10.7282/t3-xkda-y788 PY - 2020 AB - Oncology indispensably leads us to personalized medicine, which allows an individual approach to be taken with each subject. Personalized oncology is based on pharmacogenomics and the effect of genetic differences in individuals. Biomarkers detected using molecular biology tools allow the molecular characterization of cancer signatures and provide information relevant for personalized treatment. The key to success of these targeted therapy is to identify a "predictive biomarker" and validate the "predictive biomarker" through randomized clinical trials. In this dissertation, we focus on biomarker based two-stage clinical trial designs with survival outcomes. In Part I of this dissertation, we assume that there is no misclassification of biomarker and we design a two-stage adaptive enrichment clinical trial, based on a binary "predictive" biomarker. At the interim analysis, based on the statistics observed from the biomarker negative strata, a decision is made to either continue enrolling both biomarker positive and biomarker negative subjects or enrich the remaining number of subjects only to biomarker positive subjects. In Part II, we address the issue of misclassification of biomarker which is common in determining the predictive biomarker status. A two-stage stratified study design is proposed and evaluated. We use the information obtained from both marker appeared-positive strata and marker appeared-negative strata, to solve the adjusted log rank statistics for true marker positive and true marker negative group. No additional distributional assumption is needed for this stratified design. In Part III, we extend the biomarker misclassification adjustment method to the two-stage enrichment designs proposed in Part I. With some additional distributional assumption (exponential distribution assumption for survival times), we can use the information obtained from interim analysis, to help obtain the adjusted log rank statistics for the true marker positive group, even though the marker appeared-negative group was discontinued after interim analysis and no marker appeared-negative subjects are enrolled in Stage II. Family-wise type I error control is achieved by considering correlation of log rank statistics from the same and/or different stages. R-code is developed to calculate critical values, to achieve specified global power, or specified marginal power, and to calculate sample size as well. KW - Predictive biomarker KW - Public Health LA - English ER -