Development of microarray genomic biomarkers and their application in clinical trials

Many human diseases, especially cancer, are known to be heterogeneous with regard to their molecular and genomic properties, and many molecularly targeted drugs benefit only a subset of treated patients. Therefore, it is very important to develop methods that predict whether a patient is likely to benefit from a targeted agent. The goals of this research were to systematically investigate statistical issues involved in prospective genomic biomarker clinical trials by assessing and developing genomic biomarker classification methods and developing new approaches for incorporating gene information into phase III clinical trials.;In the first part of this dissertation, we proposed a new version of the modified linear discriminant analysis (MLDA) approach and we considered modified LDA approaches to high-dimensional microarray classification for the development of classifier genomic biomarkers. We showed that the modified LDA approaches can be used as effective classification tools in high-dimensional microarray classification problems such as the development of genomic biomarkers from microarray data. We also performed a large-scale comparison study of different methods for prediction error estimates in high-dimensional microarray classification, which is very important for evaluating the performance of the developed classifier genomic biomarker. Our results show that classification algorithms, feature set size, and sample size of the training set all impact on the performance of prediction error estimation methods.;In the second part of this dissertation, we describe extensive simulation studies to investigate whether the power of phase III clinical trials can be improved by incorporating classifier genomic biomarkers developed from phase II data or part of phase III data using MLDA. We showed that the power of phase III trials can be greatly improved by incorporating such biomarkers when the biomarker positive and biomarker negative groups had different responses to the treatment. Finally, we compared several prospective genomic biomarker phase III designs and extended the enrichment design to a two-stage group sequential setting. We concluded that the group sequential enrichment design can be used as an efficient prospective genomic biomarker design in consideration of the overall power and average sample size as well as for economic and ethical reasons.;The results of this research provide important guidelines with respect to (1) development of genomic biomarkers from microarray data to identify sensitive patients and (2) use of classifier genomic biomarkers to improve the efficiency of phase III trials.