Incorporating artificial intelligence in urology: Supervised machine learning algorithms demonstrate comparative...

Area of Science:

• Urology
• Oncology
• Data Science

Background:

• One-third of patients experience biochemical recurrence (BCR) after radical prostatectomy (RP), linked to metastasis and mortality.
• Predicting BCR is crucial for patient management and treatment planning.

Purpose of the Study:

• To employ machine learning (ML) algorithms for predicting BCR post-RP.
• To compare the performance of ML models against traditional regression models and established nomograms.

Main Methods:

• Utilized a prospective uro-oncology registry with 1130 patients undergoing RP.
• Applied three ML models (Naïve Bayes, Random Forest, Support Vector Machine) and compared them with regression models and Kattan, CAPSURE, and John Hopkins nomograms.
• Trained and validated models using clinicopathological parameters, predicting BCR at 1, 3, and 5 years.

Main Results:

• ML models demonstrated robust accuracy (AUC > 0.82) and good calibration in predicting BCR.
• Naïve Bayes, Random Forest, and Support Vector Machine models showed strong predictive performance at 1, 3, and 5 years.
• ML models outperformed traditional nomograms (Kattan, JHH, CAPSURE) in BCR prediction (p < 0.001).

Conclusions:

• Supervised ML algorithms provide accurate predictions of BCR after RP.
• ML models outperform existing nomograms, offering a more precise tool for risk stratification.
• These findings can guide tailored patient care and identify candidates for multimodal therapy.