Prediction of biochemical recurrence after radical prostatectomy from primary tumour characteristics

Objectives: To construct and externally calibrate a predictive model for early biochemical recurrence (BCR) after radical prostatectomy (RP) incorporating clinical and modern imaging characteristics of the primary tumour. Patients and Methods: Patients who underwent RP following multiparametric magnetic resonance imaging, prostate biopsy and prostate-specific membrane antigen-positron emission tomography/computed tomography (PSMA-PET/CT), from two centres in Australia and the Netherlands. The primary outcome was biochemical recurrence-free survival (BRFS), where BCR was defined as a rising PSA level of ≥0.2 ng/mL or initiation of postoperative treatment per clinician discretion. Proportional hazards models to predict time to event were developed in the Australian sample using relevant pre- and post-surgical parameters and primary tumour maximum standardised uptake value (SUVmax) on diagnostic PSMA-PET/CT. Calibration was assessed in an external dataset from the Netherlands with the same inclusion criteria. Results: Data from 846 patients were used to develop the models. Tumour SUVmax was associated with worse predicted 3-year BRFS for both pre- and post-surgical models. SUVmax change from 4 to 16 lessened the predicted 3-year BRFS from 66% to 42% for a patient aged 65 years with typical pre-surgical parameters (PSA level 8 ng/mL, Prostate Imaging-Reporting and Data System score 4/5 and biopsy Gleason score ≥4 + 5). Considering post-surgical variables, a patient with the same age and PSA level but pathological stage pT3a, RP Gleason score ≥4 + 5 and negative margins, SUVmax change from 4 to 16 lessened the predicted 3-year BRFS from 76% to 61%. Calibration on an external sample (n = 464) showed reasonable performance; however, a tendency to overestimate survival in patients with good prognostic factors was observed. Conclusion: Tumour SUVmax on diagnostic PSMA-PET/CT has utility additional to commonly recognised variables for prediction of BRFS after RP.