In this study an external validation of nomograms incorporating PSMA PET data for LNI was performed in a multicenter, international cohort of patients who underwent RARP and ePLND following preoperative staging with PSMA PET. Both the Amsterdam-Brisbane-Sydney model and Muehlematter Model 1 demonstrated strong discriminatory performance, achieving an AUC of 0.81 and 0.79, respectively, outperforming the MSKCC, Briganti 2012, and the two remaining Muehlematter (Model 2 and 3) models, which had AUCs of 0.67, and 0.68, 0.66, and 0.67, respectively. Furthermore, both calibration and net benefit analyses favored the Amsterdam-Brisbane-Sydney and Muehlematter Model 1.

Employing a conservative approach to avoid missing any LNI cases, applying thresholds of 8% for the Amsterdam-Brisbane-Sydney nomogram and 15% for Muehlematter Model 1 could spare ePLND in 15% and 16% of patients, respectively, without missing any LNI-positive cases. Using a higher threshold of 11%, the Amsterdam-Brisbane-Sydney model could spare 26.3% of the PLNDs at the cost of missing 4.7% of LNI cases. This underscores the potential for refining ePLND indications of these tools in patients preoperatively staged with PSMA PET, based on individual patient characteristics and preferences. In comparison, to achieve similar rates of spared ePLND using the MSKCC and Briganti 2012 nomograms, an 8% threshold with MSKCC would spare 16% of patients while missing 7% of LNI cases, and a 5% threshold with Briganti 2012 would spare 17% of patients while missing 4.7% of LNI cases. It should be noted that the range of predicted probabilities differs between nomograms. The distribution of Muehlematter Model 1 is narrower (ranging from 13.0 to 99.0%) compared to that of the Amsterdam-Brisbane-Sydney model (2.7–99.0%), which is consistent with the distribution observed in the original study [4]. These differences should be considered when establishing a threshold value for clinical decision-making. Nevertheless, our external validation confirms the favorable utility of models incorporating PSMA PET information, as previously reported in their respective development studies [21, 22].

In this study, we evaluated a broad population of PSMA PET staged patients undergoing RARP and ePLND, including patients with miN1-2 disease. Inclusion of patients with miN1-2 disease may explain the more favorable performance observed for the Amsterdam-Brisbane-Sydney tool, as this nomogram was developed on a cohort also including these patients. Additionally, based on the distribution of coefficients, miN1 status is a key determinant of predicted LNI risk in both the Amsterdam-Brisbane-Sydney and Muehlematter 1 nomograms [21, 22]. Other important contemporary predictors for LNI, all included in the Amsterdam-Brisbane-Sydney and Briganti 2019 and Briganti 2023 nomograms, are MRI T stage, percentage of positive systematic biopsy cores and biopsy GG. Multivariable logistic regression analysis using backward elimination with current study data identified these four predictors as the strongest factors associated with LNI risk prediction. These factors outperformed PSMAvol, serum PSA, age, and clinical stage assessed by DRE (Supplementary Table S7), suggesting their superior prognostic relevance.

Our findings are complimentary to those of other external validation studies, revealing that available clinical nomograms predicting LNI had suboptimal performance in patients staged miN0M0 [30, 31]. The very recent external validation study showed that the Briganti 2023 outperformed other available tools, including the Briganti 2019 and the Amsterdam-Brisbane-Sydney nomograms, for this patient subgroup [31]. The Briganti 2023 nomogram is specifically developed for patients with miN0M0 disease, and includes biopsy GG, clinical stage at mpMRI, maximum diameter of the index lesion, preoperative PSA and percentage of positive cores at systematic biopsy [30]. The fact that this nomogram outperformed other tools that were not specifically designed for miN0M0 populations highlights the importance of using models tailored to the target population. Apart from the fact that a portion of the patients in the current cohort was also included in the development of both the Briganti 2019 and Briganti 2023 nomograms, not all patients underwent the biopsy strategies required by these nomograms. Therefore, this cohort was unfortunately not suitable for a reliable external validation of these tools.

However, our subanalysis, which included only patients staged as miN0, revealed that the performance of both the Amsterdam-Brisbane-Sydney and Muehlematter 1 nomograms was substantially lower in this subset. In clinical practice, patients opting for surgery with miN1-2 would in the vast majority of cases also undergo ePLND. This is supported by the overall data of this multicenter project including patients preoperatively staged with PSMA PET, treated with RARP with or without ePLND. Among all patients staged as miN1-2, only in one patient (1.4%) ePLND was omitted, which was due to perioperative complications. To conclude, since nomogram-based predicted risk would mostly drive treatment decisions in patients staged miN0, the clinical relevance of both Muehlematter 1 and Amsterdam-Brisbane-Sydney could be suboptimal in their current forms. We would therefore advise to update both of these tools specifically for miN0 patients, to improve their clinical utility in this patient subgroup.

In addition, the generally high reported specificity of PSMA PET for the detection of pelvic nodal metastasis, mostly exceeding 90% in prior studies, it can be further argued if nomogram-based risk calculation provides added value in cases of miN1-2 disease [2, 3, 5]. Based on the observed PPV of 71.9% in this cohort, which aligns with prior studies reporting PPVs between 70% and 81%, incorporating additional clinical prognostic factors may however enhance risk stratification and reduce potential overtreatment. This may be particularly relevant for patients meeting intermediate-risk criteria with miN1-2 findings on PSMA PET, as the observed PPV in the intermediate-risk group for detecting LNI was 58.8%, compared to the high-risk group, where a PPV of 77.5% was observed (Tables S8 and S9).

A notable strength of this study is the validation of the [68Ga]Ga-PSMA-11 PET–based Muehlematter Model 1 nomogram using [18F]F-PSMA-1007 PET data. Given that this nomogram incorporates the SUV-based quantitative parameter PSMAvol, our findings suggest that the nomogram’s predictive accuracy is maintained regardless of the specific PSMA inhibitor used (PSMA-11 vs- PSMA-1007) and regardless the specific isotope (68Ga vs. 18F). This indicates potential interchangeability between different PSMA-PET techniques in this predictive model. Other factors that may have contributed to the favorable findings in the [18F]F-PSMA-1007 cohort included higher median number of resected nodes (21 vs. 14) and lower biopsy downgrading rates (31% vs. 38%) compared with the [68Ga]Ga-PSMA-11 subgroup (Tables S1, S10–S12). Additionally, the majority (93.6%) of patients staged with [18F]F-PSMA-1007 were treated at a single institution, which also may partially explain the favorable fit observed.

Since the introduction of PSMA PET for primary staging, the detection rates of both pelvic nodal lesions and distant metastases have significantly influenced treatment selection strategies [32, 33]. Despite its advantages, with a sensitivity of only 40–50% for detecting LNI, ePLND remains the most accurate method currently available for nodal staging. While incorporating PSMA PET findings into clinical prediction models may improve patient selection, our systematic analysis of LNIs detected and missed suggests ePLND will still need to be performed in a substantial number of node-negative patients to ensure that clinically significant rates of node-positive cases are not missed.

A critical question remains whether ePLND can be omitted in all patients undergoing radical prostatectomy with preoperative miN0M0 PET findings. This uncertainty is particularly significant as the undetected nodal lesions are typically small, millimetric in size, and their impact on long-term outcomes remains unclear [34]. Notably, PSMA PET-guided metastasis-directed therapy in the recurrence setting could serve as a reliable safety net, potentially mitigating the consequences of undetected nodal disease. The results of the ongoing Dutch national randomised controlled trial, PSMA SELECT, are highly anticipated. This trial randomised patients with an LNI risk > 5% to either a nomogram-based ePLND approach (ePLND performed universally) or a PSMA PET-guided strategy (ePLND omitted in miN0 patients and performed only in miN1 patients) [35]. These findings are expected to provide important insights into the optimal integration of PSMA PET in primary staging and treatment planning.

This study has a number of limitations. First, only the highest biopsy ISUP GG per prostate lobe were available in this study; and all positive cores on the ipsilateral side were regarded as ISUP GG ≥ 2 in case this was also the highest ipsilateral GG reported. This could have led to overestimation of the predicted probabilities for the Amsterdam-Brisbane-Sydney nomogram. However, a sensitivity analysis performed, counting a maximum number of 1 positive core per prostate lobe with GG ≥ 2 did not alter the study’s conclusions (Tables S13 and S14, Figs. S3 and S4). Second, although all patients underwent ePLND, encompassing the obturator fossa and the internal and external iliac arteries, variations in surgical templates were permitted across different hospitals and surgeons. Unfortunately, rates of template modifications were not available, which could limit the generalizability of our findings. Third, there was no central review of MRI, PSMA PET, or histopathological evaluation. However, inter-reader variability of these parameters can also be seen as a strength, as it enables more robust estimates and reflects real-world clinical practice. Fourth, among the 86 patients classified as pN1, 7 (8.1%) were classified as pN1 based on PSA persistence and presence of the suspicious node on postoperative PSMA PET. The lack of histopathological confirmation in these patients should be considered as a limitation. Fifth, the unavailability of certain parameters included in the Briganti 2017, Briganti 2019 and Briganti 2023 nomograms prevented their external validation in this study [16, 30, 36].

Lastly, the present cohort consisted of selected cases for both PSMA PET and ePLND, which could have introduced selection bias.