Pelvic lymph node (LN) metastasis is a prognostic factor for prostate cancer; however, pelvic LN dissection (PLND) is associated with morbidity. Existing nomograms lack the integration of imaging variables as predictors of LN metastasis. We developed a model comprising clinical and imaging variables to predict LN metastasis. Data of patients who underwent robot-assisted radical prostatectomy for PLND were retrospectively analyzed. Preoperative variables included prostate-specific antigen (PSA) level, PSA density (PSAD), International Society of Urological Pathology (ISUP) grade, magnetic resonance imaging (MRI) T stage, and LN status assessed using both computed tomography and MRI. Logistic regression was performed to identify predictors. Model performance was evaluated using receiver-operating characteristic (ROC) analysis, calibration, bootstrap validation, and decision curve analysis. Among 123 patients, LN metastases were observed in 14 (11.4%). ISUP grade 5 (odds ratio OR, 6.84; p = 0.002), MRI T stage ≥T3 (OR, 6.22; p = 0.004), and PSAD (OR, 3.91 per 0.5 ng/mL 2 increase; p = 0.008) were identified as independent predictors. The model integrating MRI T stage, ISUP grade, PSAD, and MRI LN status achieved an area under the ROC curve (AUC) of 0.903 and optimism-corrected AUC of 0.891. The LN positivity rate in a subgroup of low-risk patients, defined as MRI T stage ≤2, ISUP ≤2, PSAD ≤0.15, and MRI LN-negative (n = 13), was 0%. At a 6% predicted probability threshold, the model would avoid PLND in 69.7% of patients while missing one LN-positive case (1.2%). The AUC of the model (90.3%) was higher than that of nomograms developed by Partin, Briganti, and the Memorial Sloan Kettering Cancer Center in our cohort (84.3%, 83.5%, and 88.8%, respectively). The integrated model exhibited excellent predictive accuracy. This approach may enable the safe omission of PLND, reducing morbidity without compromising outcomes.
Yoon et al. (Wed,) studied this question.