[2M4-CC] Late Breaking Research Session - B

[2M4-CC-03] Prediction of ESWL Outcome Using Model-Based Recursive Partitioning

Daisuke Fujita¹, Shimpei Yamashita², Ryusuke Deguchi², Syoji Kobashi¹ (1. University of Hyogo, 2. Wakayama Medical University)

Among the treatments for urolithiasis, extracorporeal shock wave lithotripsy (ESWL) is frequently chosen in clinical practice due to non-invasive treatment. However, the uncertainty of ESWL outcomes in clinical practice often makes it difficult for physicians to decide on treatment strategies. In previous studies, decision trees and factor analysis were used to predict ESWL outcomes from clinical case data. Decision trees can visually represent the importance of factors, but are difficult to evaluate quantitatively. On the other hand, general statistical factor analysis can compare the importance of factors, but it cannot evaluate nonlinear relationships and does not provide a unified prediction model. In order to overcome the above problems, this study proposes an ESWL prediction model using Model-Based Recursive Partitioning. 344 cases of ESWL treatment are collected, which include age, gender, hydronephrosis, stone size and stone CT image features. Using this data, a model was trained and the model was evaluated for predictive accuracy by AUC of ROC analysis. The predictive model proposed in this study allows for ESWL outcome prediction with high interpretability.