Development and external validation of a multivariable [ 68 Ga]Ga-PSMA-11 PET-based prediction model for lymph node involvement in men with intermediate or high-risk prostate cancer.

Purpose: To develop and evaluate a lymph node invasion (LNI) prediction model for men staged with [ ⁶⁸ Ga]Ga-PSMA-11 PET.
Methods: A consecutive sample of intermediate to high-risk prostate cancer (PCa) patients undergoing [ ⁶⁸ Ga]Ga-PSMA-11 PET, extended pelvic lymph node dissection (ePLND), and radical prostatectomy (RP) at two tertiary referral centers were retrospectively identified. The training cohort comprised 173 patients (treated between 2013 and 2017), the validation cohort 90 patients (treated between 2016 and 2019). Three models for LNI prediction were developed and evaluated using cross-validation. Optimal risk-threshold was determined during model development. The best performing model was evaluated and compared to available conventional and multiparametric magnetic resonance imaging (mpMRI)-based prediction models using area under the receiver operating characteristic curves (AUC), calibration plots, and decision curve analysis (DCA).
Results: A combined model including prostate-specific antigen, biopsy Gleason grade group, [ ⁶⁸ Ga]Ga Ga-PSMA-11 positive volume of the primary tumor, and the assessment of the [ ⁶⁸ Ga]Ga-PSMA-11 report N-status yielded an AUC of 0.923 (95% CI 0.863-0.984) in the external validation. Using a cutoff of  ≥ 17%, 44 (50%) ePLNDs would be spared and LNI missed in one patient (4.8%). Compared to conventional and MRI-based models, the proposed model showed similar calibration, higher AUC (0.923 (95% CI 0.863-0.984) vs. 0.700 (95% CI 0.548-0.852)-0.824 (95% CI 0.710-0.938)) and higher net benefit at DCA.
Conclusions: Our results indicate that information from [ ⁶⁸ Ga]Ga-PSMA-11 may improve LNI prediction in intermediate to high-risk PCa patients undergoing primary staging especially when combined with clinical parameters. For better LNI prediction, future research should investigate the combination of information from both PSMA PET and mpMRI for LNI prediction in PCa patients before RP.
(© 2023. The Author(s).)