Mp23-18 Algorithms Using Operator Splitting and Machine Learning Outperform Humans in Prediction of Patient-Reported Outcomes to Botulinum Toxin for Overactive Bladder in an External Cohort

You have accessJournal of UrologyUrodynamics/Lower Urinary Tract Dysfunction/Female Pelvic Medicine: Female Incontinence (MP23)1 May 2024MP23-18 ALGORITHMS USING OPERATOR SPLITTING AND MACHINE LEARNING OUTPERFORM HUMANS IN PREDICTION OF PATIENT-REPORTED OUTCOMES TO BOTULINUM TOXIN FOR OVERACTIVE BLADDER IN AN EXTERNAL COHORT Glenn Werneburg, Eric Werneburg, Howard Goldman, Emily Slopnick, Ly Hoang Roberts, and Sandip Vasavada Glenn WerneburgGlenn Werneburg , Eric WerneburgEric Werneburg , Howard GoldmanHoward Goldman , Emily SlopnickEmily Slopnick , Ly Hoang RobertsLy Hoang Roberts , and Sandip VasavadaSandip Vasavada View All Author Informationhttps://doi.org/10.1097/01.JU.0001008776.99097.8a.18AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: A patient's perception of symptomatic changes following an overactive bladder (OAB) therapy is a challenging prediction problem, but its solution may lead to precision optimization of OAB management. We sought to develop an accurate model using machine learning for prediction of patient-reported treatment response to intravesical botulinum toxin injection. We hypothesized the algorithm would perform superiorly to other top algorithms and expert clinicians. METHODS: Algorithms using the mathematical principle "operator splitting" were developed to predict subjective (PGI-I score) response to botulinum toxin. Algorithms were trained using the ROSETTA trial cohort data and tested and validated using the ABC trial cohort. Algorithms were designed for accuracy, efficiency, transparency, and handling of missing data and small training data sets. The ROSETTA (training set) and ABC (validation set) trials differed in inclusion criteria, data completion rates, botulinum toxin doses, and accrual time period, and thus presented a challenging predictive scenario by design. Metrics including areas under the curve (AUC) of the novel algorithms were compared to the top publicly-available machine learning classifier XGBoost, logistic regression, and human experts in the out-of-sample test set from the ABC trial. RESULTS: 167 patients from ROSETTA were included for training, and 72 from ABC for out-of-sample validation. The operator splitting neural network had AUC 0.64 and outperformed XGBoost with DART (top publicly available machine learning classifier, AUC 0.62), and human experts (AUC 0.41 – 0.54) for prediction of patient-reported subjective response to treatment in the validation set (Figure 1). CONCLUSIONS: A new machine learning model outperformed other machine learning approaches and human experts in prediction of a patients' perception of overactive bladder symptomatic improvement following botulinum toxin in an independent validation cohort. Download PPT Source of Funding: Original trials were funded by NICHD and ORWH. Data were obtained from NICHD DASH repository. The authors thanks Dr. Anthony Visco and Dr. Cindy Amundsen, principal investigators for the trials © 2024 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 211Issue 5SMay 2024Page: e389 Advertisement Copyright & Permissions© 2024 by American Urological Association Education and Research, Inc.Metrics Author Information Glenn Werneburg More articles by this author Eric Werneburg More articles by this author Howard Goldman More articles by this author Emily Slopnick More articles by this author Ly Hoang Roberts More articles by this author Sandip Vasavada More articles by this author Expand All Advertisement PDF downloadLoading ...