Model Prediction of the Length of Cancer Prior to Diagnosis with Application to Cancer Registry Data

• Activity Number: 122
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Diana L. Nadler*+ and Igor Zurbenko
• Companies: University at Albany and The State University of New York at Albany
• Keywords: Weibull distribution ; conditional survival analysis ; parametric statistics ; cancer ; carcinogenesis ; delay in diagnosis

Abstract

Survival analysis statistics in cancer research are often reported in terms of individual survival from the time of diagnosis; however, when utilizing cancer registry data the true time malignant cancer cells developed in the body is unknown. It has also been noted some types of cancer do not present symptoms in early stages and are diagnosed at late stages of disease, significantly reducing the chance of survival. This suggests there is a window of time allowing for earlier detection and increased long-term cancer survival which has globally become a dire public health concern. Here, we propose a conditional Weibull model that accounts for the assumption the individual survived up to the time of diagnosis that allows us to estimate the period between the occurrence of the first malignant cancer cell and the date of cancer diagnosis. When the hazard function decreases over time, the Weibull distribution does not possess the memoryless property of the closely related exponential distribution and retains a memory of prior survival times. We also consider an application of this method to a real dataset.