Prediction of operation time in percutaneous nephrolithotomy (PCNL) patients: A machine learning approach.

Abstract

Purpose: To investigate the factors influencing the length of percutaneous nephrolithotomy (PCNL) procedures and identify predictive variables for operation time using machine learning models.

Materials and methods: A retrospective, observational cohort study was conducted at King Abdullah University Hospital, including 352 patients who underwent PCNL between January 2017 and September 2023. Data on preoperative and postoperative variables were collected from electronic health records. Four machine learning algorithms (Random Forest Classifier, AdaBoost Classifier, eXtreme Gradient Boosting Classifier, Logistic Regression) were employed to predict operation time, with features standardized using the StandardScaler module and Synthetic Minority Over-sampling Technique (SMOTE) used to address data imbalance. The dataset was split into training (80%) and testing (20%) sets. Model performance was evaluated using ROC curves, AUC scores, accuracy, precision, recall, and F1-score.

Results: Stone burden, gender, and hydronephrosis were significantly associated with longer operation times. Machine learning analysis identified stone-free status, stone burden, and gender as key predictors, with the eXtreme Gradient Boosting Classifier achieving the highest AUC (0.789). Patients with non-stone-free status had longer operation times (p < 0.001). Stone burden and specific stone locations also significantly impacted procedure duration.

Conclusion: Stone-free status followed by stone burden and gender are critical predictors of PCNL operation time. Achieving stone-free status significantly reduces procedure duration. Machine learning models, particularly eXtreme Gradient Boosting, provide valuable predictive insights, aiding in surgical planning and optimizing patient outcomes.