AB312. SPR-39 The use of support vector machine in the prediction of stress urinary incontinence

OBJECTIVE: Stress urinary incontinence (SUI) is the involuntary leakage of urine due to an increase in abdominal pressure and it affects 30% of women over the age of 40. One of the primary risk factor is childbirth. The baby’s weight, head size and maternal age are just some of the variables clinicians can use to predict if women will develop SUI. Additional, previous family history of SUI is another predictor for development suggesting a genetic role in development of SUI. A new method used to create predictive models is a support vector machine (SVM) use in the field of cancer biology. The purpose of the study was to determine if a SVM algorithm could construct a model that can improve the performance of predicting SUI compared with previous methods. METHODS: Data was obtained from the Pelvic Floor Disorder Network Childbirth and Pelvic symptoms Study (CAPS). Only information from the Urinary Incontinence and general data forms were used (e.g., maternal age, baby weight, head circumferences). We compared our models efficiency to a previously published model. Based on the data, we employ SVM algorithm to construct a model for predicting SUI. The basic idea of SVM is to find an optimal hyper-plane which can separate the data of one class from another class. In our study, we first divided the preprocessed data into two subsets, one is training data set, and the other one is testing data set. The testing data set was utilized to train an optimal model, that is, to find an optimal hyper-plane. The testing data set was employed to test the performance of the trained model. In order to obtain stable performance, we use 10 folds cross validation to train the model and to test its performance. RESULTS: An optimal hyper-plane was determined. The results indicate the accuracy of prediction is around 70 percent, which is a little better than that of previous methods at 69 percent. CONCLUSIONS: The proposed method in this study can predict SUI. Further investigation is needed to determine if limitation of risk factors from the model can improve its performance. FUNDING SOURCE(S): None