A novel nomogram predicting lymph node invasion among patients with prostate cancer: The importance of extracapsular extension at multiparametric magnetic resonance imaging

Abstract

PurposeTo develop a novel risk tool that allows the prediction of lymph node invasion (LNI) among patients with prostate cancer (PCa) treated with robot-assisted radical prostatectomy (RARP) and extended pelvic lymph node dissection (ePLND). MethodsWe retrospectively identified 742 patients treated with RARP + ePLND at a single center between 2012 and 2018. All patients underwent multiparametric magnetic resonance imaging (mpMRI) and were diagnosed with targeted biopsies. First, the nomogram published by Briganti et al. was validated in our cohort. Second, three novel multivariable logistic regression models predicting LNI were developed: (1) a complete model fitted with PSA, ISUP grade groups, percentage of positive cores (PCP), extracapsular extension (ECE), and Prostate Imaging Reporting and Data System (PI-RADS) score; (2) a simplified model where ECE score was not included (model 1); and (3) a simplified model where PI-RADS score was not included (model 2). The predictive accuracy of the models was assessed with the receiver operating characteristic-derived area under the curve (AUC). Calibration plots and decision curve analyses were used. ResultsOverall, 149 patients (20%) had LNI. In multivariable logistic regression models, PSA (OR: 1.03; P= 0.001), ISUP grade groups (OR: 1.33; P= 0.001), PCP (OR: 1.01; P= 0.01), and ECE score (ECE 4 vs. 3 OR: 2.99; ECE 5 vs. 3 OR: 6.97; P< 0.001) were associated with higher rates of LNI. The AUC of the Briganti et al. model was 74%. Conversely, the AUC of model 1 vs. model 2 vs. complete model was, respectively, 78% vs. 81% vs. 81%. Simplified model 1 (ECE score only) was then chosen as the best performing model. A nomogram to calculate the individual probability of LNI, based on model 1 was created. Setting our cut-off at 5% we missed only 2.6% of LNI patients. ConclusionsWe developed a novel nomogram that combines PSA, ISUP grade groups, PCP, and mpMRI-derived ECE score to predict the probability of LNI at final pathology in RARP candidates. The application of a nomogram derived cut-off of 5% allows to avoid a consistent number of ePLND procedures, missing only 2.6% of LNI patients. External validation of our model is needed.