Discovery, Evaluation and Clinical Decision Making Based on Non-Monotone Biomarkers for the Early Detection of Disease

点击翻译按钮获取中文摘要

This project will develop new statistical methodologies to facilitate the discovery of biomarkers that are missed by current statistical methods and naively ignored as uninformative, in spite of the great discriminatory ability they may exhibit. Our work will focus on high-throughput technologies for the early detection of disease using blood-based biomarkers. For a given assay, these technologies allow us to have multiple biomarkers that are commonly ranked based on traditional measures such as the area under the receiver operating characteristic curve (AUC) or sensitivity/specificity at a given level of specificity/sensitivity depending on the clinical setting. Traditional criteria assume that a higher biomarker level increases the suspicion of the presence of the disease (or vice versa). There are cases, however, in which this single-directionality is severely violated. Having such markers in a large pool of candidates typically leads investigators to rank them by AUC, sensitivity, specificity, or partial AUC (pAUC), depending on the clinical setting, and to focus on only the top candidates. All these traditional metrics cannot reveal an appropriately ranked list of promising biomarkers. As a result, these biomarkers and their behavior cannot be further explored/validated by clinicians and biologists, simply because the statistical analysis does not make such markers available to them. Thus, these potentially excellent biomarkers are missed with current statistical techniques. This projects aims to develop new metrics that allow violations of the aforementioned directionality. We will provide a full framework that allows for discovery of new biomarkers (regardless of their directionality), evaluation of these biomarkers, assessment of their clinical utility and construction of a cutoff-based decision-making process.