Stuart Baker

NCI, DCP, BRG

PLCO (Learn more about this study)

2011-0195

Sep 22, 2011

Ovarian Cancer Biomarkers: Serial CA-125 + Other Markers + Clinical Variables

There is great interest in the Division of Cancer Prevention in trying to find a combination of ovarian cancer biomarkers in PLCO from stored specimens that improves classification of women as developing ovarian cancer or not. Previous classification rules have involved either serial CA-125 measurements combined with clinical variables (such as age) or a single-time measure of CA-125 combined with measures of other markers. The proposal here is to create a classification rule that considers serial CA-125, single time measures of other markers, and clinical variables. The method would involve splitting the data into training and test samples; fitting the model in the training sample and evaluating performance in the test sample. For serial CA-125 data three CA-125 markers will be considered last value, last slope (velocity), and last difference in slopes (by analogy called acceleration). Classification rules will involve the method of Swirls-and-Ripples (Baker, 2010). Swirls-and-Ripples uses either diagonal discriminant analysis to obtain linear or curved classification boundaries, called Ripples or a modification that yields smooth highly nonlinear classification boundaries, called Swirls, that sometimes outperforms Ripples. The emphasis is on finding simple classification rules. Classification performance in the test sample will be summarized by ROC and relative utility curves. Software is available at http://prevention.cancer.gov/programs-resources/groups/b/software/swirls website