Abstract: Objective To To compare the value of risk assessment models based on decision tree C5.0 or logistic regression in predicting postpartum stress urinary incontinence. Methods A total of 505 females in postpartum recovery clinic of one tertiary grade-A hospital in Shandong Province from July 2020 to January 2021 were selected as research objects. They were surveyed for general information and the data of postpartum stress urinary incontinence. An EEG feedback device was used to test the function of pelvic floor. All the data were divided into training set (n=450) and test set (n=145). Risk assessment models established with decision tree C5.0 or logistic regression were established respectively and their predictive value was assessed in terms of specificity, sensibility, accuracy, negative predictive value, positive predictive value, Youden index, and area under the curve (AUC) of receiver operating characteristic curve (ROC). Results In the training set, the accuracy, sensibility, specificity, positive predictive value, negative predictive value, Youden index and AUC of the two risk assessment models were 98.9% vs 85.6%; 94.7% vs 48.7%; 100.0% vs 95.4%; 100.0% vs 74.0%; 98.6% vs 87.4%; 94.7% vs 44.1% and 0.974 vs 0.721 respectively. The AUC of the two models indicated statistical significance (P<0.05). In test set, the accuracy, sensibility, specificity, positive predictive value, negative predictive value, Youden index and AUC of the two risk assessment models were 87.6% vs 82.8%; 78.6% vs 46.4%; 89.7% vs 91.5%; 64.7% vs 56.5%; 94.6% vs 87.7%; 68.3% vs 37.9% and 0.842 vs 0.689 respectively. The AUC of the two models showed statistical significance (P<0.05). Conclusion Decision tree C5.0-based risk assessment model presents better performance in predicting postpartum stress urinary incontinence than the model established with logistic regression.

Key words: decision tree C5.0, logistic regression, postpartum, stress urinary incontinence, predictive model