Pd44-02 Machine Learning Algorithms for Predicting the Recurrence of Obstruction After Pyeloplasty: A Multicenter Retrospective Study

You have accessJournal of UrologyReconstruction: Ureteral Reconstruction (Including Pyeloplasty) and Bladder Reconstruction (Including Trauma-Related Fistula) II (PD44)1 May 2024PD44-02 MACHINE LEARNING ALGORITHMS FOR PREDICTING THE RECURRENCE OF OBSTRUCTION AFTER PYELOPLASTY: A MULTICENTER RETROSPECTIVE STUDY Shengwei Xiong, Lei Zhang, Zhihua Li, Hongjian Zhu, Peng Zhang, Bing Wang, Liqun Zhou, and Xuesong Li Shengwei XiongShengwei Xiong , Lei ZhangLei Zhang , Zhihua LiZhihua Li , Hongjian ZhuHongjian Zhu , Peng ZhangPeng Zhang , Bing WangBing Wang , Liqun ZhouLiqun Zhou , and Xuesong LiXuesong Li View All Author Informationhttps://doi.org/10.1097/01.JU.0001008800.83683.92.02AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Pyeloplasty represents the standard management of patients with ureteropelvic junction obstruction (UPJO). Recurrent UPJO, although infrequent, poses significant challenges in clinical practice. The objective of this study is to develop a machine learning (ML) model capable of predicting the risk of developing recurrent obstruction after pyeloplasty and the expected time to re-intervention. METHODS: Data were retrospectively collected from October 2009 to June 2023 at four Chinese medical centers. A total of 610 eligible patients who underwent laparoscopic or robot-assisted pyeloplasty were included in this study, with a median age of 27 years old. Recurrence of Obstruction was defined as the reappearance of symptoms and worsening hydronephrosis requiring reoperation. A total of 62 clinical features were utilized in the pre-processing phase, employing six feature selection algorithms. Subsequently, these features were used to train three ML models: random survival forests (RSF), survival support vector machines (S-SVM), and extreme gradient (XG) boost models. RESULTS: At a median follow-up of 48 months, 36 (5.9%) patients developed recurrent obstruction required re-intervention. Redo pyeloplasty, the absence of "ureter resting" preparations, and vascular variation were significant model features predicting the recurrence of obstruction. ML algorithms achieved higher c-index values than conventional models. The RSF models yielded the highest c-index values (0.814), followed by S-SVM (0.793) and XG boost (c-index 0.787). Decision curve analysis (DCA) demonstrated a significant advantage of all ML models compared with conventional models currently used in practice. CONCLUSIONS: ML models for predicting the recurrence of obstruction exhibited superior predictive performance compared to the conventional prognostic models currently employed in clinical practice. However, larger studies with a sufficient sample size are required to validate these findings. Source of Funding: None © 2024 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 211Issue 5SMay 2024Page: e904 Advertisement Copyright & Permissions© 2024 by American Urological Association Education and Research, Inc.Metrics Author Information Shengwei Xiong More articles by this author Lei Zhang More articles by this author Zhihua Li More articles by this author Hongjian Zhu More articles by this author Peng Zhang More articles by this author Bing Wang More articles by this author Liqun Zhou More articles by this author Xuesong Li More articles by this author Expand All Advertisement PDF downloadLoading ...