PD10-10: Improving Prediction of Local Stage by PSMA-PET. Development of a Novel Integrated Tool for Extracapsular Extension and Seminal Vesicle Invasion Combining Clinical and Imaging Features in Localized Prostate Cancer
Introduction: Validated models predicting extracapsular extension (ECE) and seminal vesicle invasion (SVI) based on clinical parameters and mpMRI are available for prostate cancer (PCa) patients undergoing radical prostatectomy (RP). However, their performance in the PSMA era are still unknown, where the inclusion of PSMA-PET might improve their discrimination. Methods: 195 patients staged with PSMA PET scan and treated with RP ± pelvic lymph node dissection (pLND) at 9 referral centers between 2016 and 2022 were identified. Men with nodal or extra-nodal spots at PSMA PET and those who received neoadjuvant treatments were excluded. Two existing models predicting ECE and SVI based on preoperative PSA, clinical stage, ISUP Grade Group at mpMRI, ECE and SVI at mpMRI, the maximum diameter of the index lesion at mpMRI, and the percentage of cores with significant PCa at systematic biopsy were externally validated. Logistic regression analyses tested whether intraprostatic SUVmax was associated with ECE and SVI at final pathology. We then developed a new model predicting pathological ECE (Model 1) and SVI (Model 2) based on the coefficients of the previously mentioned variables with the addition of SUVmax at PSMA PET. Calibration plots, ROC-derived AUC, and decision-curve analyses (DCAs) determined the calibration, discrimination, and net benefit of the new models. Results: The median SUVmax was 12. Overall, 128 (66%) and 34 (18%) patients had ECE and SVI after RP. The existing model’s discrimination was 72% for ECE and 78% for SVI at external validation. Nonlinear associations between SUVmax and ECE and between SUVmax and SVI were observed using non-parametric curves. Two nomograms based on the new model including SUVmax considered as a restricted cubic spline showed higher discrimination for ECE (81 vs. 72%) and SVI (90% vs. 78%), and a higher net benefit compared to the available ones. Using a nomogram cut-off of 80%, the sensitivity and specificity for ECE were substantially higher for Model 1 (44% and 96%) than for mpMRI alone (31% and 94%). Moreover, for Model 2, a 40% cut-off led to higher sensitivity for SVI than mpMRI (61% vs. 52%, respectively), with similar specificity (92% vs. 95%). Conclusions: Two novel predicting models including SUVmax values showed higher performances in the prediction of ECE and SVI after RP than the currently available ones. These models showed increased sensitivity for ECE and SVI when compared to mpMRI alone. SOURCE OF Funding: None
