INTRODUCTION/ BACKGROUND: The prediction of histology of SRM could be essential for their management. The RNN is a statistical tool designed to predict malignancy or high grading of enhancing renal masses. In this study we aimed to perform an external validation of the RNN in a cohort of patients who received a PN for SRM. MATERIALS AND METHODS: This was a multicentric study in which the data of 506 consecutive patients who received a PN for cT1a SRM between January 2010 and January 2013 were analyzed. For each patient, the probabilities of malignancy and aggressiveness were estimated preoperatively using the RNN. The performance of the RNN was evaluated according to receiver operating characteristic (ROC) curve, calibration plot, and decision curve analyses. RESULTS: The area under the ROC curve for malignancy was 0.57 (95% confidence interval [CI], 0.51-0.63; P = .031). The calibration plot showed that the predicted probability of malignancy had a bad concordance with observed frequency (Brier score = 0.17; 95% CI, 0.15-0.19). Decision curve analysis confirmed a poor clinical benefit from use of the system. The estimated area under the ROC curve for high-grade prediction was 0.57 (95% CI, 0.49-0.66; P = .064). The calibration plot evidenced a bad concordance (Brier score = 0.15; 95% CI, 0.13-0.17). Decision curve analysis showed the lack of a remarkable clinical usefulness of the RNN when predicting aggressiveness. CONCLUSIONS: The RNN cannot accurately predict histology in the setting of cT1a SRM amenable to PN.
INTRODUCTION/ BACKGROUND: The prediction of histology of SRM could be essential for their management. The RNN is a statistical tool designed to predict malignancy or high grading of enhancing renal masses. In this study we aimed to perform an external validation of the RNN in a cohort of patients who received a PN for SRM. MATERIALS AND METHODS: This was a multicentric study in which the data of 506 consecutive patients who received a PN for cT1a SRM between January 2010 and January 2013 were analyzed. For each patient, the probabilities of malignancy and aggressiveness were estimated preoperatively using the RNN. The performance of the RNN was evaluated according to receiver operating characteristic (ROC) curve, calibration plot, and decision curve analyses. RESULTS: The area under the ROC curve for malignancy was 0.57 (95% confidence interval [CI], 0.51-0.63; P = .031). The calibration plot showed that the predicted probability of malignancy had a bad concordance with observed frequency (Brier score = 0.17; 95% CI, 0.15-0.19). Decision curve analysis confirmed a poor clinical benefit from use of the system. The estimated area under the ROC curve for high-grade prediction was 0.57 (95% CI, 0.49-0.66; P = .064). The calibration plot evidenced a bad concordance (Brier score = 0.15; 95% CI, 0.13-0.17). Decision curve analysis showed the lack of a remarkable clinical usefulness of the RNN when predicting aggressiveness. CONCLUSIONS: The RNN cannot accurately predict histology in the setting of cT1a SRM amenable to PN.
Authors: Mark W Ball; Michael A Gorin; Sam B Bhayani; Craig G Rogers; Michael D Stifelman; Jihad H Kaouk; Homayoun Zargar; Susan Marshall; Jeffrey A Larson; Haider M Rahbar; Bruce J Trock; Phillip M Pierorazio; Mohamad E Allaf Journal: Urol Oncol Date: 2014-12-11 Impact factor: 3.498