Prediction of aggressive prostate cancer on biopsy using classification and regression tree analysis (CART)

Abstract

4720 Background: Prostate cancer screening allows for the early detection of disease; however, not all cancer patients are candidates for intervention due patient co-morbidities and the indolent nature of some tumors. Distinguishing patients who are likely to harbor aggressive prostate cancer could significantly reduce the number of prostate biopsies performed in men that would be offered treatment only in the presence of high-grade prostate cancer. Methods: Data were collected on 1,563 consecutive referred men with a serum prostate specific antigen (PSA) of ≤ 10 ng/ml who underwent an initial prostate biopsy. Predictors of aggressive cancer (defined as Gleason sum ≥ 7) were identified using CART analysis that considered demographic variables (age, race, reason for referral, history of vasectomy, and family history), digital rectal exam findings, serum prostate–specific antigen levels (PSA), TRUS findings (prostate volume, hypoechoic lesion) and PSA density (serum PSA/prostate volume). The model building was carried out on 70% of the data (randomly selected training set) and study validation was carried using the remaining 30% of the data (validation set). Results: 408 (26.1%) of men had prostate cancer. Of these, 129 (31.6%) were determined to be Gleason ≥ 7. CART analysis created a decision tree that identified the following groups to be at high-risk for the detection of Gleason ≥ 7 prostate cancer: (1) PSAD > 0.165 ng/ml/cc; and (2) PSAD ≤ 0.165 ng/ml/cc and age > 57.5 years, and prostate volume > 22.7 cc. The performance of the decision tree is shown below: Conclusions: CART analysis successfully identified groups at high risk for the diagnosis of Gleason ≥ 7 prostate cancer. Application of this model to clinical practice would result in a 33.5% reduction in unnecessary biopsies when a diagnosis of high-grade prostate cancer would be the primary determinant of subsequent treatment decisions. No significant financial relationships to disclose.