Prediction of Postoperative Creatinine Levels by Artificial Intelligence after Partial Nephrectomy

Area of Science:

• Nephrology
• Artificial Intelligence
• Surgical Outcomes

Background:

• Predicting postoperative renal function after partial nephrectomy (PN) is challenging due to dynamic, intertwined pre-, peri-, and postoperative factors.
• Acute kidney injury (AKI) is a significant concern following PN, impacting functional outcomes.

Purpose of the Study:

• To develop an artificial intelligence (AI) model to predict residual renal function and AKI incidence post-PN.
• Utilize perioperative factors for accurate prediction of postoperative serum creatinine (Cr) levels.

Main Methods:

• Retrospective study of 785 patients undergoing PN (open or robotic).
• AI model trained using 44 perioperative features; XG-Boost and genetic algorithms employed.
• Primary outcome: immediate postoperative serum Cr; Secondary outcome: AKI incidence (eGFR < 60 mL/h).

Main Results:

• AI model achieved a low Mean Absolute Error (MAE) of 0.03 mg/dL for predicting serum Cr.
• High correlation (R² = 0.9669) between predicted and measured postoperative serum Cr.
• Excellent sensitivity (85.5%–100%) and specificity (99.7%–100%) for AKI prediction across training and test sets.

Conclusions:

• The developed AI model accurately predicts postoperative serum Cr levels and AKI likelihood after PN.
• Model accuracy supports the need for personalized guidelines to optimize pre- and postoperative care strategies.
• AI-driven predictions can enhance multidisciplinary planning for improved patient management following PN.