External validation of nomograms including PSMA PET information for the prediction of lymph node involvement of prostate cancer

Background: Novel nomograms predicting lymph node involvement (LNI) of prostate cancer (PCa) including PSMA PET information have been developed. However, their predictive accuracy in external populations is still unclear. Purpose: To externally validate four LNI nomograms including PSMA PET parameters (three Muehlematter models and the Amsterdam-Brisbane-Sydney model) as well as the Briganti 2012 and MSKCC nomograms. Methods: Patients with histologically confirmed PCa undergoing preoperative MRI and PSMA PET/CT before radical prostatectomy (RP) and extended pelvic lymph node dissection (ePLND) were included. Model discrimination (AUC), calibration and net benefit using decision curve analysis were determined for each nomogram. Results: A total of 437 patients were included, comprising 0.7% with low-risk disease, 39.8% with intermediate-risk disease, and 59.5% with high-risk disease. Among them, 86 out of 437 (19.7%) had pN1 disease. The sensitivity and specificity of PSMA PET/CT for the detection of LNI were 47.7% (95% CI: 36.8–58.7) and 95.4% (95% CI: 92.7–97.4), respectively. Among predictive models, the Amsterdam-Brisbane-Sydney model achieved the highest discrimination (AUC: 0.81, 95% CI: 0.76–0.86), followed by Muehlematter Model 1 (AUC: 0.79, 95% CI: 0.74–0.85), both with good calibration but slight systematic overestimation of risks across all thresholds. The MSKCC and Briganti 2012 models had AUCs of 0.68 (95% CI: 0.61–0.74) and 0.67 (95% CI: 0.61–0.73), respectively, and both had moderate calibration. Decision curve analysis indicated that the Amsterdam-Brisbane-Sydney model provided superior net benefit across thresholds of 5–20%, followed by the Muehlematter Model 1 nomogram showing benefit in the 14–20% range. Using thresholds of 8% for the Amsterdam-Brisbane-Sydney nomogram and 15% for Muehlematter Model 1, ePLND could be spared in 15% and 16% of patients, respectively, without missing any LNI cases. Conclusion: External validation of the Muehlematter Model 1 and Amsterdam-Brisbane-Sydney nomograms for predicting LNI confirmed their strong model discrimination, moderate calibration, and good clinical utility, supporting their reliability as tools to guide clinical decision-making.