Establishment and Validation of a Novel Prediction Model for Early Natural Biochemical Recurrence After Radical Prostatectomy Based on Post-Operative PSA at Sixth Week.

Abstract

Based on post-operative PSA at 6th week (PSA6w) after radical prostatectomy to establish an optimal model for predicting natural biochemical recurrence (BCR). A total of 742 patients with post-operative PSA6w from PC-follow database, between January 2003 and October 2022, were included. All the patients had not received any hormone therapy and radiotherapy before operation and BCR. Of these patients, 588 cases operated by one surgeon were enrolled for modelling and another 154 cases operated by other surgeons were for external validation. After screened by Cox regression, the post-operative PSA6w, pathological stage, Gleason Grade and positive surgical margins were adopted for modelling. The R software was used to plot the nomogram of the prediction model for BCR. C-index and calibration curve were calculated to evaluate the new model. Finally, integrated discrimination improvement was adopted to evaluate the prediction performances of the new nomogram model and the classical Kattan nomogram. The C-index of the new model was 0.871 (95% CI: 0.830-0.912). The calibration curve of the new model demonstrated superior consistency between the predicted and actual value. The C-index of the external validation group was 0.850 (95% CI: 0.742-0.958), which demonstrated perfect universality. The integrated discrimination improvement showed a 12.61% improvement in prediction performance over that of the classical Kattan nomogram (P < 0.01). Based on the new nomogram, patients were divided to high and low BCR group with a 3 year BCR-free cutoff probability as 74.72%. Low-risk patients, accounting for 77.89% of the patients, have no need to follow up frequently with a false-negative rate only 5.24%, which will save medical resources to a large extent. Post-operative PSA6w is a sensitive risk biomarker for early natural BCR. The new nomogram model could predict BCR probability with a higher accuracy and will further simplify the clinical follow-up strategies.