Development and validation of an online predictive model for biochemical recurrence after radical prostatectomy in elderly patients

TL;DR

This study created a web-based tool to predict biochemical recurrence in elderly prostate cancer patients after surgery, using factors like Gleason score and PSA levels.

Contribution

A novel nomogram and online tool for predicting biochemical recurrence in elderly prostate cancer patients after radical prostatectomy.

Findings

The nomogram showed excellent predictive accuracy with AUCs of 0.857, 0.915, and 0.916 for 2, 3, and 5-year BCR-free survival in the training set.

The model demonstrated good calibration and clinical net benefit confirmed by decision curve analysis.

Abstract

To develop and validate a novel model for predicting biochemical recurrence (BCR) in elderly prostate cancer (PCa) patients after radical prostatectomy (RP) and to create an accessible online tool for its clinical application.

This retrospective study included patients who underwent RP at two independent medical centers. The initial cohort included 450 patients (2015-2022), which were randomly divided into a training set (n = 315) and an internal validation set (n = 135) at a 7:3 ratio. An independent cohort of 175 patients (2013-2023) was used as the external validation set. Potential predictors were screened via univariable Cox regression. The independent prognostic factors for BCR were subsequently identified via multivariate Cox regression. A predictive nomogram was developed on the basis of these independent factors. The model performance was assessed via time-dependent ROC curves, calibration curves, decision curve analysis (DCA), and Kaplan–Meier (KM) curves.

Cox multivariate regression analysis revealed that Gleason score (GS), lymph node metastasis (LNM), seminal vesicle invasion (SVI), and free prostate-specific antigen (fPSA) were independent risk factors for BCR after RP in the elderly population (all P < 0.05). The nomogram exhibited excellent time-dependent discriminative ability: the AUCs for 2-year, 3-year, and 5-year BCR-free survival were 0.857, 0.915, and 0.916, respectively, in the training set; 0.810, 0.846, and 0.856, respectively, in the internal validation set; and 0.698, 0.679, and 0.715, respectively, in the external validation set. Calibration curves demonstrated good agreement between the predicted BCR risk and actual incidence, and DCA confirmed that the model provides substantial clinical net benefit. We further developed an online tool (https://bcrnomapp.shinyapps.io/bcr-risk/) for personalized BCR-risk prediction.

We developed a validated nomogram based on four independent risk factors—the Gleason score, lymph node metastasis, seminal vesicle invasion, and free PSA—for predicting BCR in elderly prostate cancer patients after radical prostatectomy. This model demonstrated robust predictive performance across multiple validation sets. The accompanying web-based tool facilitates rapid and individualized risk assessment, aiding in clinical decision-making.

Figures