American Statistical Association
A biomarker is clinically important if it can identify populations with meaningfully different risks of disease. The measure of a biomarker to stratify risk has evolved from relative risks to diagnostic accuracy and AUC statistics. Using examples from our work studying biomarkers for cancers of the cervix and lung, we show that evaluating risk stratification based on relative risks or the AUC can be misleading, and advocate using absolute risks and risk differences to quantify risk stratification. Risk differences are valuable because they tell us how many more people are classified at high/low risk by the biomarker. Risk differences also highlight situations where small relative risks can provide potentially meaningful risk stratification. However, using absolute risks requires interpretation, and I discuss strategies we have used to interpret absolute risk for our scientific and clinical colleagues, such using principles of consistency to treat equal disease risks equally, and to develop analogies to closely-related diseases that use risk thresholds. We also discuss statistical methodology and software we have developed to estimate absolute risk from case-control studies conducted within cohorts that we used to estimate lung cancer risk.
|Date:||Monday, April 25, 2011|
|Time:||4:00 - 5:00 P.M.|
Memorial Sloan-Kettering Cancer Center
Department of Epidemiology and Biostatistics
307 East 63rd Street
(between First and Second Avenues)
New York, New York
Note: To gain access to the building, please follow the directions by the telephone in the foyer.