Development and Internal Validation of a Nomogram for Predicting Biochemical Recurrence After Radical Prostatectomy: A Retrospective Cohort Study

Background Biochemical recurrence (BCR) remains an important clinical outcome after radical prostatectomy (RP) for prostate cancer. Established postoperative tools such as the Cancer of the Prostate Risk Assessment Postsurgical (CAPRA-S) score provide useful risk stratification, but model calibration and predictor effects may vary across populations. This study aimed to develop and internally validate a clinicopathological nomogram for BCR prediction and to compare its performance with CAPRA-S and the conventional TNM staging system. Methods A total of 218 consecutive patients who underwent RP for clinically localized or locally advanced prostate cancer between January 2019 and December 2025 at our institution were retrospectively enrolled and randomly divided into a training cohort (n = 153) and a validation cohort (n = 65). LASSO-penalized Cox regression followed by multivariable Cox regression was used for model development. Performance was assessed using Harrell's C-index, calibration, time-dependent discrimination, the 3-year Brier score, and decision curve analysis. CAPRA-S was calculated according to the original postsurgical scoring algorithm. Additional sensitivity analyses evaluated surgeon and surgical-approach effects, postoperative adjuvant treatment, and bootstrap internal validation in the complete cohort. Results After a median follow-up of 24.0 months, 119 patients (54.6%) experienced BCR. Six predictors were retained in the final model: preoperative prostate-specific antigen (PSA), biopsy Gleason score, positive surgical margin, seminal vesicle invasion, extracapsular extension, and lymph node metastasis. The nomogram C-index was 0.812 (95% CI, 0.762–0.862) in the training cohort and 0.785 (95% CI, 0.714–0.856) in the validation cohort, compared with 0.781 (95% CI, 0.729–0.833) and 0.758 (95% CI, 0.682–0.834), respectively, for CAPRA-S. The differences in C-index were modest and not statistically significant (training: delta = 0.031, P = 0.118; validation: delta = 0.027, P = 0.311). The 3-year Brier score was 0.168 for the nomogram, 0.183 for CAPRA-S, and 0.224 for TNM. In full-cohort bootstrap validation, the optimism-corrected C-index was 0.791 and the calibration slope was 0.93. Conclusions The proposed nomogram provides individualized BCR risk estimates using routinely available clinicopathological variables. Its discrimination was numerically higher than CAPRA-S in this cohort, although the incremental difference was not statistically significant. The model should therefore be viewed as an internally validated, population-specific risk tool requiring independent external validation and longer follow-up before routine clinical use, particularly for 5-year prediction.