From planning to prognosis: predicting renal function after minimally-invasive partial nephrectomy with artificial...

Area of Science:

• Nephrology
• Urology
• Artificial Intelligence in Medicine

Background:

• Minimally-invasive partial nephrectomy is a standard treatment for kidney tumors.
• Predicting postoperative renal function decline is crucial for patient management.
• Current methods for predicting renal function after surgery have limitations.

Purpose of the Study:

• To develop and validate a machine learning model for predicting renal function decline after minimally-invasive partial nephrectomy.
• To identify key patient, tumor, and surgical variables influencing postoperative estimated glomerular filtration rate (eGFR) drop.
• To assess the performance of a Random Forest Regressor model in this prediction task.

Main Methods:

• A dataset of 556 patients undergoing minimally-invasive partial nephrectomy between 2015 and 2023 was utilized.
• Patient demographics, tumor characteristics, and intraoperative surgical details (clamping strategy, resection technique, renorrhaphy type) were included as features.
• A Random Forest Regressor model was trained and evaluated for its ability to predict the 3-month postoperative eGFR drop.

Main Results:

• The Random Forest Regressor model achieved a prediction accuracy of 89.29%.
• The model demonstrated a mean absolute error of 8.09 mL/min/1.73 m² in estimating eGFR drop.
• A strong correlation (r=0.904, P<10⁻⁴²) was observed between predicted and actual outcomes.

Conclusions:

• Machine learning, specifically a Random Forest Regressor, can accurately predict renal function decline post-partial nephrectomy.
• The model's high accuracy supports its utility in personalized surgical planning for nephron-sparing surgery.
• AI-driven predictions can enhance the management of patients undergoing kidney-sparing procedures.