Abstract
Objectives
To develop and validate model predicting lymph node involvement(LNI) in men undergoing radical prostatectomy with/without suspicious magnetic resonance imaging(MRI) with/without genomic classifiers (GC).
Methods
Retrospective analysis of patients that underwent extended pelvic lymphadenectomy(ePLND) during robot-assisted radical prostatectomy(RARP). ePLND was defined as removal of obturator, internal and external iliac and distal part of common iliac lymph nodes. Based on preoperative work-up, imaging, and GC testing, we stratified patients into three cohorts. Cohort I with suspicious MRI (n = 2172), cohort II with nonsuspicious MRI (n = 1233) and cohort III with GC irrespective of MRI findings (n = 1003). Logistic regression analysis performed to create nomogram for predicting LNI. Receiver operative characteristics (ROC) and decision curve analysis (DCA) were performed to evaluate net benefit. Statistical analyses were performed using R 4.3.3. We also utilized artificial neural network (ANN) for calculating LNI risk by using binary classification model.
Results
Overall 138 (6.4%), 49 (3.9%) and 69 (6.8%) patients had LNI in cohort I,II and III respectively. Multivariable analysis showed prostate specific antigen (PSA), biopsy Gleason Grade Group (GGG), number of positive cores, MRI LNI were significant predictors of LNI in all cohorts; MRI lesion size, MRI T stage (cohort I), MRI prostate volume (cohort II) and biopsy GC (cohort III) were significant. ROC for predicting LNI were 0.92, 0.84 and 0.91 for cohort I,II and III respectively. Using the ANN, we calculated ROC curves were 0.90,0.82 and 0.91 for cohort I, II and III, respectively. DCA showed a clinical benefit for the model detection of LN metastases for each cohort.
Conclusions
We developed the nomogram that integrate clinical, radiological, histological and genomic parameters to predict lymph node metastases during prostatectomy. This will avoid unnecessary lymphadenectomy at cost of missing of few metastases.