Decision tree analysis for prostate cancer prediction

Abstract

Introduction/Objective. The use of serum prostate-specific antigen (PSA) test has dramatically increased the number of men undergoing prostate biopsy. However, the best possible strategies for selecting appropriate patients for prostate biopsy have yet to be defined. The aim of the study was to develop a classification and regression tree (CART) model that could be used to identify patients with significant prostate cancer (PCa) on prostate biopsy in patients referred due to abnormal PSA, digital rectal examination (DRE) findings, or both, regardless of the PSA level. Methods. The data on clinicopathological characteristics regarding prebiopsy assessment collected from patients who had undergone ultrasound-guided prostate biopsies included the following: age, PSA, DRE, volume of the prostate, and PSA density (PSAD). The CART analysis was carried out using all predictors identified by univariate logistic regression analysis. Different aspects of predictive performance and clinical utility risk prediction model were assessed. Results. In this retrospective study, significant PCa was detected in 92 (41.6%) out of 221 patients. The CART model had three splits based on PSAD, as the most decisive variable, prostate volume, DRE, and PSA. Our model resulted in an 83.3% area under the receiver operating characteristic curve. Decision curve analysis showed that the regression tree provided net benefit for relevant threshold probabilities compared with the logistic regression model, PSAD, and the strategy of biopsying all patients. Conclusion. The model helps to reduce unnecessary biopsies without missing significant PCa.