Predicting acute kidney injury following open partial nephrectomy treatment using SAT-pruned explainable machine...

Area of Science:

• Nephrology
• Urology
• Data Science
• Machine Learning

Background:

• Acute Kidney Injury (AKI) is a significant complication following Partial Nephrectomy (PN), affecting renal function in up to 24.3% of patients.
• Predicting AKI preoperatively is crucial for patient management and improving outcomes after PN.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting AKI after PN using preoperative parameters.
• To identify key preoperative factors influencing AKI occurrence in patients undergoing PN.

Main Methods:

• A Random Forest (RF) model was developed using pre-operative data from 723 adult patients who underwent open PN.
• The RF model employed Boolean satisfaction-based pruned decision trees and greedy feature selection for binary classification (AKI vs. non-AKI).
• Model performance was optimized using hyper-parameter grid search and evaluated with fivefold cross-validation.

Main Results:

• The optimized RF model achieved an average precision and recall of 0.69 for classifying AKI and non-AKI groups.
• The model demonstrated a 0.75 probability of correctly classifying new predictions.
• The developed predictive model is accessible as an online calculator.

Conclusions:

• The developed machine learning model accurately predicts AKI following open PN with 75% accuracy.
• Future work includes external validation and adaptation of the model for minimally-invasive PN procedures.