4/24/2015

Post date: May 4, 2015 9:41:56 PM

Title: A Framework for Quantifying Risk Stratification from Diagnostic Tests

Speaker: Hormuzd A. Katki, National Cancer Institute, NIH

A test or biomarker that stratifies disease risks allows clinicians to only intervene on only those who have or will develop disease. We propose a general framework for risk stratification by introducing the risk stratification distribution, which is the distribution of the changes in disease risk indicated by each possible test result. The mean of this distribution, the mean risk stratification (MRS) is the average amount of extra disease (or deficit of disease) that a test reveals for an individual patient. The MRS is a function of not only the risk difference, but also marker positivity, demonstrating that a big risk difference does not imply good risk stratification for markers that are rarely positive. The MRS is also a function of Youden's index and disease prevalence, demonstrating that a large Youden's index does not imply good risk stratification if disease is too rare. We demonstrate that the net expected benefit of a diagnostic test is a function of test characteristics solely through the MRS. Reasoning based on MRS enforces rational decision-making based on the principle of "equal management of equal risks". We discuss examples from the presenter's experience serving on the guidelines committee for introducing HPV testing into cervical cancer screening.