PURPOSE:To assess the efficacy of a variety of prognostic models in the definition of intermediate-risk prostate cancer and to compare them to our own empiric model. METHODS AND MATERIALS:Two hundred fifty-six consecutive men with prostate adenocarcinoma treated with external beam radiotherapy alone were studied. Biochemical failure (defined as 3 consecutive PSA rises or the initiation of androgen deprivation therapy) was examined using univariate, multivariate, and recursive partitioning analyses. The risk classification model used in our department was then compared to a number of published models to assess the relative performance of each in discriminating risk groups. RESULTS:At a median follow-up of 62.4 months, the 5-year Biochemical failure-free survival (bFFS) was 46.8% for the overall group. This relates to 5-year bFFS of 77.8%, 51.1%, and 33.8% based on our institutional criteria for low-, intermediate-, and high-risk features, respectively. All the models examined showed an outcome group with a comparatively similar poor outcome when applied to our data. Large variation was seen in the intermediate-risk groups, with 5-year bFFS ranging from 38.1% to 51.1%. Good risk categories had similar large variations. All published models showed inability to delineate three significantly different outcome groups. Recursive partitioning analysis derived categories based on combinations of PSA (with cutpoints at 42.4, 20, and 10.6 ng/mL) and Gleason score (with cutpoints at 2-6 and 7-10) only. CONCLUSIONS:Large variations in the relative performance of a number of prognostic models are shown when applied to our local data. The prognostic efficacy of PSA and biopsy Gleason score is reiterated, although other factors will need to be explored to further improve the performance of prognostic models, particularly in defining the intermediate-risk subset of prostate cancer.