Construction of a Preoperative Radiologic-Risk Signature for Predicting the Pathologic Status of Prostate Cancer at Radical Prostatectomy.

Abstract

We developed a radiologic-risk signature (RRS) that serves as a surrogate for the pathologic status of prostate cancer (PCA) and investigated its ability to predict disease-free survival. This study included 631 patients with localized PCA who underwent prostatic multiparametric MRI before prostatectomy. Images from 426 training datasets were structurally interpreted and correlated to a postoperative Memorial Sloan Kettering Cancer Center (MSKCC) score by a stepwise partial least-squares regression analysis. The developed RRS, compared with a preoperative Kattan nomogram, was validated in a cohort of 205 patients with 3-year follow-up data after prostatectomy in terms of calibration, discrimination, and clinical usefulness. Statistical tests were performed by AUC analysis, Kaplan-Meier test, and decision curve analysis. The RRS, which consists of 12 preoperative variables, faithfully represented postoperative MSKCC score in 426 training (r = 0.75; p < 0.001) and 205 validation (r = 0.79; p < 0.001) datasets. For patients in the validation group, RRS showed better discriminative power (C-index, 0.859; 95% CI, 0.779-0.939; p = 0.013) than did the preoperative Kattan nomogram (C-index, 0.780; 95% CI, 0.701-0.859) for predicting 3-year biochemical recurrence and showed higher net benefits for a probability threshold of greater than 10%. Characteristics of RRS can faithfully represent the tumor pathologic status and predict accurately the disease postoperative outcome before prostatectomy.