Pd36-01 Interpretable Machine Learning Models for Personalized Prognostication in Intermediate-Risk Radical Prostatectomy Patients

You have accessJournal of UrologySurgical Technology & Simulation: Artificial Intelligence III (PD36)1 May 2024PD36-01 INTERPRETABLE MACHINE LEARNING MODELS FOR PERSONALIZED PROGNOSTICATION IN INTERMEDIATE-RISK RADICAL PROSTATECTOMY PATIENTS Umar Ghaffar and Robert J. Karnes Umar GhaffarUmar Ghaffar and Robert J. KarnesRobert J. Karnes View All Author Informationhttps://doi.org/10.1097/01.JU.0001008916.72488.6a.01AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Our objective was to utilize machine learning (ML) models to create a web application to predict survival outcomes in AUA intermediate risk patients after radical prostatectomy. METHODS: Patients diagnosed with prostate cancer were selected from the National Cancer Database (NCDB) between 2010 to 2020 to analyze overall survival at 12, 36 and 60 months in patients with AUA intermediate risk prostate cancer after radical prostatectomy. Data was preprocessed to handle missing values through MICE imputation. Continuous variables were scaled using robust scaler and normalized. Five machine learning models - Random Forest, XGBoost, LightGBM Gradient Boost, Support Vector and Multi-Layer Perceptron were employed and the best performing model was used to form a web based application. RESULTS: 433,267 patients with mean age 62 (±SD 7.0) and mean PSA 6.7 (±SD 10.1) were identified after excluding patients with missing clinical T stage and needle core biopsy based Gleason score. Performance evaluation indicated that the top-performing models for 12, 36 and 60 month mortality were random forest, LightGBM and LightGBM with area under receiver operating curve (AUROC) 0.9799, 0.9298 and 0.8722 respectively. (Table 1) A web based application was deployed to predict outcomes at http://bit.ly/ncdbprostatectomy using platform of huggingface.co. CONCLUSIONS: ML algorithms act as important tools in predicting survival outcomes for radical prostatectomy patients with AUA intermediate risk prostate cancer. These models may be incorporated as web applications to better prognosticate outcomes and potentially guide management. Source of Funding: N/A © 2024 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 211Issue 5SMay 2024Page: e792 Advertisement Copyright & Permissions© 2024 by American Urological Association Education and Research, Inc.Metrics Author Information Umar Ghaffar More articles by this author Robert J. Karnes More articles by this author Expand All Advertisement PDF downloadLoading ...