MP62-19 CROSS VALIDATION OF A PREDICTIVE ANALYTIC MODEL WHICH PREDICTS SUCCESS AND COMPLICATIONS OF SHOCKWAVE LITHOTRIPSY

Abstract

You have accessJournal of UrologyStone Disease: Shock Wave Lithotripsy1 Apr 2017MP62-19 CROSS VALIDATION OF A PREDICTIVE ANALYTIC MODEL WHICH PREDICTS SUCCESS AND COMPLICATIONS OF SHOCKWAVE LITHOTRIPSY Blake Hamilton, Ryan Seltzer, Donald Gleason, Stephen Nakada, and Glenn Gerber Blake HamiltonBlake Hamilton More articles by this author , Ryan SeltzerRyan Seltzer More articles by this author , Donald GleasonDonald Gleason More articles by this author , Stephen NakadaStephen Nakada More articles by this author , and Glenn GerberGlenn Gerber More articles by this author View All Author Informationhttps://doi.org/10.1016/j.juro.2017.02.1950AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookTwitterLinked InEmail INTRODUCTION AND OBJECTIVES Shockwave lithotripsy (SWL) is a primary treatment for nephrolithiasis that has been used widely for the past 3 decades. Recently, this technology has come under fire because of declining outcomes in the face of improving alternative technologies. Multiple authors have described pre-operative parameters that improve the success of SWL, including stone size, location, density, and skin-to-stone distance. Using these and other parameters, we present a predictive analytic model to help urologists select the most effective treatment modality with the highest likelihood of success and lowest likelihood of complication. METHODS We performed a random 70/30 split of 7,000 SWL treatment records for renal and ureteral stones from 2010-2016 to train and validate a generalized linear mixed model (GLMM) using statistical software (PROC GLIMMIX in SAS 9.4). This model uses 9 parameters: stone size, Hounsfield Units (HU), body mass index (BMI), stone location, anesthesia type, SWL machine type, anticoagulant use, age, and gender to predict treatment success, defined as stone free or fragments = 4mm, and to predict treatment complications. Actual treatment success and complications were obtained from self-reported physician follow-up surveys tied to the original SWL treatment record. Both treatment and follow-up data are housed in The Stone Disease Registry. RESULTS The training model was significantly related to treatment success, Likelihood Ratio (LR) Chi-square = 1136.02, p < .0001, Area under the curve (AUC) = .82. This model was in turn a good predictor of success in the validation dataset, AUC = .81. The training model was also significantly related to complications, Likelihood Ratio Chi-square = 538.75, p < .0001, AUC = .91. This model was a fair predictor of complication rate in the validation dataset, AUC = .77. CONCLUSIONS This novel predictive analytic model provides accurate prediction of treatment success and complications for SWL. Given the robust model fit to the validation data, we conclude that this model will be useful in prospectively predicting success for the treatment of urinary stones with SWL. This has the potential to assist urologists in prospectively making evidence-based decisions on which treatment modality will be most effective in maximizing success and minimizing complications and costs for treatment of urinary stones. © 2017FiguresReferencesRelatedDetails Volume 197Issue 4SApril 2017Page: e834 Advertisement Copyright & Permissions© 2017MetricsAuthor Information Blake Hamilton More articles by this author Ryan Seltzer More articles by this author Donald Gleason More articles by this author Stephen Nakada More articles by this author Glenn Gerber More articles by this author Expand All Advertisement Advertisement PDF downloadLoading ...