Predicting ESWL outcome using classification and regression tree (CART) analysis
Sources of Funding: None
Introduction
Previous studies have developed various predictive models for stone-free rate following extracorporeal shock wave lithotripsy (ESWL). However, these models have several limitations such as difficulty in usage and lack of information derived from CT imaging. In this study, we sought to develop a user-friendly predictive model for ESWL outcome using the classification and regression tree (CART) analysis._x000D_
Methods
We retrospectively reviewed charts of 245 patients who underwent ESWL for upper urinary tract calculi between January 2008 and January 2016. To create the CART decision tree, predictor variables were entered into the software (R version 3.2.2) to classify patients who achieved stone-free after a first session of ESWL. Variables included age, sex, hydronephrosis, urinary drainage, stone location, stone volume, number of stones and three indices based on CT attenuation of the stone, i.e., mean stone density (MSD), standard deviation of stone density (SDSD) and variation coefficient of stone density (VCSD). Stone-free status was determined as absence or residual fragments <4mm using NCCT within three months after a first session of ESWL._x000D_
Results
Overall stone-free rate was 47.8%. In CART analysis, VCSD had the greatest variable importance (100%) followed by MSD (56.7%), stone volume (52.7%) and SDSD (49.3%). A total of five subgroups of patients with distinct stone-free rate were produced by three predictive variables (Figure).
Conclusions
We have generated a first CART decision tree for ESWL outcome that included VCSD as a novel and most important factor, as well as previously reported predictors such as MSD and stone volume. This model provides clinician with practical bedside tool for predicting stone-free following ESWL._x000D_
Funding
None
Shimpei Yamashita
Takashi Iguchi
Satoshi Nishizawa
Akinori Iba
Kazuro Kikkawa
Isao Hara