Development and validation of an explainable machine learning model for predicting acute kidney injury after...

Area of Science:

• Nephrology
• Urology
• Artificial Intelligence in Medicine

Background:

• Robot-assisted partial nephrectomy (RAPN) is a minimally invasive treatment for renal masses.
• Acute kidney injury (AKI) is a common complication following RAPN, associated with poor prognosis.

Purpose of the Study:

• To develop and validate an interpretable machine learning model for predicting AKI after RAPN.
• To enable individualized risk assessment for patients undergoing RAPN.

Main Methods:

• Retrospective analysis of 325 patients for training and 146 for external validation.
• Utilized Boruta algorithm for feature selection and eight machine learning algorithms.
• Employed Shapley additive explanations (SHAP) for model interpretability.

Main Results:

• The Gradient Boosting Machine (GBM) model showed strong predictive performance (AUC 0.889 internal, 0.779 external).
• Key predictors identified: duration of renal artery blockade, preoperative serum creatinine, BMI, age, and gender.
• Specific thresholds for these predictors indicated increased AKI risk.

Conclusions:

• The interpretable GBM model effectively predicts RAPN-AKI, aiding early identification of high-risk patients.
• This tool can support timely interventions to potentially reduce AKI incidence and improve patient outcomes.
• Current applicability is limited to patients with low-risk or normal preoperative renal function.