Statistical Methods for Cancer Detection Using Biomarkers

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT In cancer research, precision medicine hinges on the development of valid biomarkers for cancer diagnosis, dis- ease prognosis, and prediction of response to specific therapeutic interventions. Fueled by the rapid recent ad- vances in the scientific knowledge of molecular biology and high-throughput omics technologies, a large num- ber of candidate biomarkers for various cancers have been or are being identified. Statistical and computational methods play a critical role in rigorously evaluating these biomarkers and further developing clinically relevant prediction rules to ultimately improve and advance cancer treatment and patient management. However, most existing methods, for continuous biomarkers, target diagnostic accuracy measures dictated by mathematical con- venience rather than clinical utility. Particularly, a screening or diagnostic test in many clinical contexts needs to maintain a high sensitivity (or specificity) and thus specificity at a controlled sensitivity level (or sensitivity at a controlled specificity level) is a clinically desirable accuracy metric. Yet, statistical and computation methods for this metric are mostly lacking, or suboptimal even when available as in limited circumstances. To address this ur- gent analytic need, this proposed project will develop novel and efficient statistical and computational methods specifically targeting this accuracy metric of clinical interest. When a single biomarker is under consideration or compared with another biomarker, Aims 1 and 2 will provide statistical tools for the inference and for covariate adjustment. On the other hand, multiplex prediction rules that prudently combine multiple biomarkers hold the promise to achieve improved diagnostic accuracy, since many cancers are heterogeneous. For optimal multiplex rule formulation, Aims 3 and 4 will develop computation algorithms and statistical inference methods with both linear combination and, often biologically and clinically motivated, logic combinations. These proposed ana- lytic methods will be thoroughly investigated through rigorous asymptotic studies and extensive simulations. They will be applied to a number of our prostate cancer biomarker studies, which motivated this project, from the Early Disease Research Network (EDRN). User-friendly computer software will be made available to the re- search community. These proposed methods will facilitate more effective biomarker research for cancer as well as other diseases.