Prediction Of Stress Urinary Incontinence After Pelvic Organ Prolapse Surgery:A Machine Learning-Based Model

Background:About 8%to 40%of patients with pelvic organ prolapse may have stress urinary incontinence after prolapse surgery.However,no unified standard exists for predicting the occurrence of postoperative stress urinary incontinence.Previous prediction models could not be applied to patients with preoperative subjective urinary incontinence or those receiving colpocleisis or transvaginal mesh surgery.This study aimed to develop and validate a new machine-learning model to predict stress urinary incontinence 1 year after pelvic organ prolapse surgery,and compare it to previous models.Methods:Female patients who underwent pelvic floor reconstruction for stage Ⅱ-Ⅳanterior or apical prolapse between January 1,2015 and December 31,2019 at Peking Union Medical College Hospital were retrospectively enrolled.Prolapse surgery included LeFort/colpocleisis,sacrocolpopexy,native tissue repair,and transvaginal mesh surgery.The existing models were externally validated.Subsequently,the dataset was randomly divided into two sets at a 4:1 ratio.The larger group was used to construct and internally validate models of logistic regression,random forest,and XGBoost,which were then externally validated in the smaller group.The discrimination of prediction models was evaluated using the area under the receiver operating characteristic curve,while the calibration of the models was measured via the Spiegelhalter z test,mean absolute error,and calibration curves.Results:Overall,555 patients with pelvic organ prolapse were enrolled in this study,and 116(20.9%)experienced stress urinary incontinence in 1 year postoperatively.In the external validation,previous models revealed poor performance(areas under the curve 0.544 and 0.586,respectively;P values for the Spiegelhalter z test<0.001).In this study,three models were constructed using logistic regression,random forest and XGBoost methods.The areas under the curve of them were 0.595,0.842 and 0.714,respectively.However,only the XGBoost model exhibited good discrimination and calibration in both internal and external validations.Body mass index,C point of pelvic organ prolapse quantification stage,age,Aa point of pelvic organ prolapse quantification stage,and transvaginal mesh surgery were the five most important predictors in the XGBoost model.Conclusions:Previous models exerted poor discrimination and calibration among this Chinese population.Hence,we developed and validated an XGBoost model,which performed well irrespective of the preoperative subjective urinary incontinence and surgical methods.