Machine learning and explainable artificial intelligence to predict pathologic stage in men with localized prostate...

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Accurate prediction of prostate cancer pathologic stage is crucial for treatment planning.
• Existing nomograms have limitations in predicting disease extent.
• Machine learning (ML) offers potential for improved predictive accuracy.

Purpose of the Study:

• To develop and validate ML models for predicting prostate cancer pathologic stage.
• To compare the performance of ML models against established nomograms.
• To assess the clinical utility of ML-based predictions.

Main Methods:

• Seven ML models were trained using data from the National Cancer Database for patients with prostate adenocarcinoma.
• Models predicted organ-confined disease, extracapsular extension, seminal vesicle invasion, and lymph node involvement.
• Performance was evaluated using area under the curve (AUC) and decision curve analysis (DCA) on internal and external validation datasets.

Main Results:

• The extreme gradient boosted trees ML model demonstrated superior performance.
• ML models achieved higher AUC values compared to MSK nomograms for all predicted stages.
• Model performance was consistent across internal and external validation datasets.

Conclusions:

• ML models provide more accurate predictions of prostate cancer pathologic stage than existing nomograms.
• Improved prediction accuracy can guide oncologists and patients in choosing definitive treatment options.
• These findings support the integration of ML tools in clinical decision-making for prostate cancer management.