Development and validation of a preoperative ¹⁸F-PSMA-1007 PET/MR prognostic model for predicting biochemical recurrence in prostate cancer: a multicentre study.

Abstract

To develop and externally validate a preoperative ¹⁸F-PSMA-1007 PET/MR model for biochemical recurrence (BCR) after radical prostatectomy and assess its incremental prognostic value beyond clinical risk scores. This multicentre retrospective study included 210 patients from three centres (training, n = 150; external validation, n = 60). Predictors were selected by 10-fold cross-validated LASSO-Cox regression using the one-standard-error criterion and entered into a multivariable Cox model. Performance at 1-3 years was assessed using time-dependent AUCs, calibration, IPCW Brier scores, and decision-curve analysis. Incremental value beyond CAPRA-S was evaluated using likelihood-ratio tests (LRTs). Sensitivity analyses assessed the potential influence of persistent postoperative PSA and secondary treatment initiated before BCR. Seventy-five patients experienced the study endpoint of biochemical recurrence. The final model included ADCmin-TBR, TMR, and PET/MR T-stage. Time-dependent AUCs were 0.703-0.858 in the training cohort and 0.766-0.846 in the validation cohort. DeLong tests showed no significant AUC differences from CAPRA-S; however, adding the PET/MR model improved fit at 2 and 3 years in the derivation cohort (both LRT p < 0.001) and at 3 years in the external validation cohort (LRT p = 0.031). DCA showed positive net benefit across most clinically relevant thresholds. ADCmin-TBR and TMR remained directionally stable across sensitivity analyses, whereas the T-stage effect was attenuated after treatment censoring. The ¹⁸F-PSMA-1007 PET/MR model provided externally validated prognostic information. It also showed incremental value beyond CAPRA-S at 3 years. Prospective validation is required before it can guide secondary-treatment decisions. ClinicalTrials.gov (NCT06604377) retrospectively registered on September 17, 2024.