Predicting tumor failure in prostate carcinoma after definitive radiation therapy: limitations of models based on prostate-specific antigen, clinical stage, and Gleason score.

Abstract

In this report, we use new patient data to test three popular models developed to predict the outcome of definitive radiation therapy. The data come from 240 men with localized prostate cancer and who were treated with definitive radiation therapy at a community hospital. All three models tested were based on the three commonly available variables of pretreatment prostate-specific antigen (PSA), Gleason score, and tumor stage, and we used the Cox proportional hazards model and the logistic regression model to relate these variables to outcome. We discovered that in our data, the optimal way to use pretreatment PSA was as natural log(PSA), the optimal way to use T stage was in three categories: T1 and T2, T3, and T4, and that the optimal use of Gleason score was as <7 versus > or =7. Nevertheless, models confined to the optimal use of these three variables leave much uncertainty about important outcomes, such as the probability of relapse within 5 years.