A data mining approach to predicting patient-based laser machine settings for kidney stone treatment operation

Abstract

Urinary tract stone disease is a common health problem that a ects human health and quality of life. The main goal for the treatment of this disease is to reach a complete stone-free status by causing minimal damage to the patient. The advancement in technology has brought many modalities for the treatment of kidney stones. Selecting the best treatment option by considering patient characteristics is an important factor a ecting the success of the treatment. In this study, it is aimed to estimate patientspeci c machine settings for successful completion of an operation performed using the Retrograde Intrarenal Surgery method. With this motivation, the performance of linear regression, regression trees, random forest, and extreme gradient boosting methods in machine setting estimation are analyzed. The study is carried out using a dataset provided by the Urology Department of Istanbul Medipol University Hospital. Since the dataset has many missing values with only few complete observations and imputation methods do not provide good results, synthetic data is generated using the original dataset. Models constructed on synthetic data are tested on the original data. Models established on synthetic data have been found to give better estimates. In addition, the models trained on successful observations and the models trained on unsuccessful observations give di erent estimates, since the data groups they are trained on are di erent.