Project Leader: Martin McIntosh PhD
The 25% of ovarian cancer cases diagnosed with their cancer confined to the ovary have a five-year survival rate over 90%, but the majority of cases , diagnosed at late stage, have a five-year survival rate of only 20%. Thus, early detection of ovarian cancer may substantially reduce ovarian cancer mortality. Combining imaging technology (such as transvaginal sonography and color Doppler imaging) and biomarker technology (such as CA 125) in screening has great potential to achieve high rates of early detection, but routine screening requires higher sensitivity than has currently been achieved. One approach to improving sensitivity is to discover and use new markers that can complement CA 125 (HER-2/neu, LPA, etc.), and many investigators have shown a substantial gain in sensitivity when using a panel of markers in diagnostic tests for distinguishing benign and malignant ovarian tumors. Another approach improves sensitivity by screening with only CA 125 using a sophisticated algorithm that captures information from longitudinal measurements (the ROM algorithm). Our goal is to combine these two approaches and develop novel algorithms for screening using a panel of longitudinally measured biomarkers. Substantial methodologic issues arise when confronting this problem. One issue is selecting which of the dozens of new and existing markers should be used in a panel. Unlike diagnostic testing, adding markers to a panel used for screening can actually decrease sensitivity, and so we must have statistical methods that can select which markers are mutually complementary and best improve sensitivity when used together. How a marker performs depends on how it is used, so that ranking markers requires we that specify algorithms for their use. Our approach decomposes markers' behavior into within and between subject components of variability then generates screening rules using the logic of hypothesis testing. We intend to rank markers by estimating their ability to improve the ROC curves that characterize their performance. The ultimate goal of this project recommends a panel of biomarkers and an algorithm that may be tested in a clinical trial for screening.
Project Contact:
