333 – Survival Methods for Risk Estimation/Prediction
Estimating Prostate Cancer Net Survival and Cancer-Specific Death
Christopher Morrell
Loyola University Maryland
Michael W. Kattan
Cleveland Clinic
Roy MacKintosh
VA Sierra Nevada Health Care System
Stephen Van Den Eeden
Kaiser Permanente
Thomas B. Neville
Soar
Estimates of the risk of death from prostate cancer are valuable for making screening decisions. Knowing a man's risk of death from prostate cancer can help him select the best PSA threshold for biopsy. Estimates of All-Cause Death (ACD) risk are limited to the population studied. Estimates of Cancer-Specific Death (CSD) risk are more useful than ACD risk. There are two methods of estimating CSD risk: 1) Using records of the cause of death, and 2) Net survival methods that remove other causes from ACD risk. Net survival methods are superior to relying on cause of death because they do not depend on judgment; but they depend heavily on estimates of No-Cancer Death (NCD) risk. Here we present approaches to estimating CSD risk for datasets that may not have records of cause of death. We use data for prostate cancer from the U.S. Department of Veterans Affairs (VA) with data for over 14 million men and 33 million PSA tests. VA has superb death record data for veterans but does not have good records of cause of death. Therefore, net survival methods are the only way to estimate CSD risk for VA data, which puts a premium on accurate estimates of NCD risk.