What does this research mean for the field?

Interpretable machine learning models, particularly Random Forest, can accurately predict biochemical recurrence in prostate cancer patients after radical prostatectomy using routinely available clinical and pathological data. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

The study aims to develop and validate machine learning algorithms for predicting biochemical recurrence in prostate cancer patients after radical prostatectomy.

June 12, 2026Open Access

Prediction of biochemical recurrence in prostate cancer: The contribution of machine learning for achieving patient-centered interventions

Key Points

The study aims to develop and validate machine learning algorithms for predicting biochemical recurrence in prostate cancer patients after radical prostatectomy.
Retrospective analysis of 375 Colombian prostate cancer patients treated with radical prostatectomy (2016–2025).
Utilized 7 machine learning algorithms and a 70/30 train-test split with 10-fold cross-validation.
Assessed model interpretability with SHapley Additive exPlanations.
Biochemical recurrence occurred in 103 (27.5%) patients post-surgery.
Random Forest algorithm showed an accuracy of 87% and an AUC of 0.84, outperforming other models.
Key predictors included Gleason score, extracapsular extension, and lymph node involvement.

Abstract

Background: Prostate cancer (PCa) is the second most common malignancy in men worldwide. Radical prostatectomy (RP) is a curative option; however, up to 30% of patients develop biochemical recurrence (BCR), defined as rising prostate-specific antigen levels, which increases the risk of metastasis and mortality. This study aimed to develop and validate 7 machine learning (ML) algorithms for early BCR prediction. Methods: We retrospectively analyzed 375 Colombian patients with PCa treated with RP (2016–2025). The BCR was defined as 2 prostate-specific antigen values >0.2 ng/mL. The ML algorithms (logistic regression, random forest RF, support vector classifier, K-nearest neighbors, Gaussian naive Bayes, gradient boosting, and XGBoost) were trained using a 70/30 split with 10-fold cross-validation. Model development followed the sustainability, accuracy, fairness, and explainability principles. Interpretability was assessed with SHapley Additive exPlanations. Results: BCR occurred in 103 (27.5%) patients. The predictors significantly associated with recurrence included a higher Gleason score, extracapsular extension, seminal vesicle invasion, lymph node involvement, and positive margins. RF demonstrated the best balance of accuracy and interpretability, with an area under the receiver operating characteristic curve of 0.84, accuracy of 87%, and F1 score of 0.68. SHapley Additive exPlanations analysis identified lymph node involvement, Gleason score, and tumor invasion as the most influential features. Ensemble methods did not improve performance beyond RF. Conclusions: Interpretable ML models, particularly RF, can accurately predict BCR after RP using routinely available clinical and pathological data. These cost-effective tools may enhance patient stratification, guide timely interventions, and be integrated into health systems and digital pathology initiatives to support the evidence-based management of PCa.

Read Full Paperexternally

Ask AI

Helpful

Bookmark

View Full Paper