A retrospective cohort study on the use of machine learning to predict stone-free status following percutaneous nephrolithotomy: An experience from Saudi Arabia.

BACKGROUND

Machine learning techniques have been used extensively in the field of clinical medicine, especially when used for the construction of prediction models. The aim of the study was to use machine learning to predict the stone-free status after percutaneous nephrolithotomy (PCNL).

MATERIALS AND METHODS

This is a retrospective cohort study of 137 patients. Data from adult patients who underwent PCNL at our institute were used for the purpose of this study. Three supervised machine learning algorithms were employed: Logistic Regression, XGBoost Regressor, and Random Forests. A set of variables comprising independent attributes including age, gender, body mass index (BMI), chronic kidney disease (CKD), hypertension (HTN), diabetes mellitus, gout, renal and stone factors (previous surgery, stone location, size, and staghorn status), and pre-operative surgical factors (infections, stent, hemoglobin, creatinine, and bacteriuria) were entered.

RESULTS

137 patients were identified. The majority were males (65.4%; n = 89), aged 50 years and above (41.9%; n = 57). The stone-free status (SFS) rate was 86% (n = 118). An inverse relation was detected between SFS, and CKD and HTN. The accuracies were 71.4%, 74.5% and 75% using Logistic Regression, XGBoost, and Random Forest algorithms, respectively. Stone size, pre-operative hemoglobin, pre-operative creatinine, and stone type were the most important factors in predicting the SFS following PCNL.

CONCLUSION

The Random Forest model showed the highest efficacy in predicting SFS. We developed an effective machine learning model to assist physicians and other healthcare professionals in selecting patients with renal stones who are most likely to have successful PCNL treatment based on their demographics and stone characteristics. Larger multicenter studies are needed to develop more powerful algorithms, such as deep learning and other AI subsets.