Predicting Lymph Node Invasion in Prostate Cancer – Validating the Briganti Nomogram and Developing a Novel Machine Learning based Nomogram for UK Clinical Practice

Introduction According to EAU guidelines, the indication for extended pelvic lymph-node dissection(ePLND) at the time of radical-prostatectomy (RP) should be based on validated nomograms. The Briganti 2017(B17) nomogram is considered the most appropriate to predict lymph-node invasion (LNI) in our cohort because most patients were diagnosed via systematic biopsies. We aim to perform the first UK-based external validation of B17 and improve its accuracy using machine learning approaches through development of an updated UK-tailored Briganti Nomogram 2025 (UKB25). Methods From a retrospective cohort of 1551 prostate cancer patients who underwent RP, 286 had ePLND between 2010 and 2021 in a tertiary referral centre. The B17 variables - preoperative PSA; clinical stage; biopsy Gleason grade group; percentage of positive cores with highest-grade PCa; and percentage of positive cores with lower grade PCa - were analysed with univariate and multivariate logistic-regression models predicting LNI. All variables with an OR > 1 were included in the subsequent model. The UKB25 nomogram was developed using penalised logistic-regression of the Briganti variables with elastic-net-regularisation. An optimal threshold was selected to maximise accuracy without increasing false negative rate. Performance metrics were compared between UKB25 and retrospective application of the B17 nomogram. Results Of 286 patients, 23(8.0%) had LNI. Application of the B17 nomogram, with a 7% threshold, to our population results in 8.7% of patients below the cut-off having LNI. This is unsatisfactory as per Briganti’s definition. Instead, using a 5.8% cutoff, the refined machine-learning-based UKB25 nomogram outperforms the Briganti 2017 nomogram in classification accuracy (McNemar’s test statistic χ2 = 34.2, p < 0.001). Importantly, UKB25 would have spared 52 patients from unnecessary ePLND (162 vs 214) with a lower rate of missed LNI (4.2%), thereby enhancing specificity (38.4% vs 18.6%) and PPV (11.9% vs 8.9%), while maintaining NPV (99% vs 96.1%), sensitivity (95.6% vs 91.3%) and area under ROC (0.785 vs 0.763). Conclusions Although the B17 predicts LNI reliably in other populations, its accuracy is limited in the UK. The updated UKB25 nomogram shows superior predictive performance in the UK, reducing the number of unnecessary ePLNDs by one quarter without missing additional LNI-positive patients. Further external validation is warranted.