Artificial neural network model to predict biochemical failure after radical prostatectomy

Area of Science:

• Oncology
• Urology
• Artificial Intelligence in Medicine

Background:

• Biochemical failure (rising PSA or adjuvant therapy) post-radical prostatectomy (RP) is an adverse prognostic factor for localized prostate cancer (CAP).
• Predicting biochemical failure is crucial for guiding treatment decisions in CAP management.

Purpose of the Study:

• To develop and validate an artificial neural network (ANN) model for predicting biochemical failure in patients treated with RP for CAP.
• To provide a tool that assists clinicians and patients in choosing among definitive treatment options for CAP.

Main Methods:

• Utilized clinical and pathologic data from 175 patients who underwent RP.
• Selected predictive variables included age, PSA, Gleason grade, and Gleason sum.
• Employed bootstrap training and validation sets (80% training, 20% validation) for model development.

Main Results:

• Forty-four percent of patients experienced biochemical failure, with a median follow-up of 2.5 years.
• The developed ANN model achieved an area under the ROC curve of 0.80 for prediction.
• The model demonstrated good predictive performance with a sensitivity of 0.74 and specificity of 0.78.

Conclusions:

• Artificial neural network models can effectively predict PSA failure using preoperative variables.
• These predictive models can support shared decision-making between patients and physicians regarding definitive CAP therapy.