基于机器学习的女性压力性尿失禁发病风险预测模型建立及效能评价

Establishing a risk prediction model for the onset of female stress urinary incontinence based on machine learning

摘要

目的 运用K最近邻法（KNN）、支持向量机（SVM）、决策树（DT）及随机森林（RF）构建女性压力性尿失禁（SUI）发病的预测模型，并评估各模型效能，为SUI的早期诊断提供参考。方法 回顾性分析2019年10月-2023年10月石河子大学第一附属医院泌尿外科及妇产科治疗的女性SUI患者及同期行健康查体女性的临床资料，将产后42 d女性纳入产后组（n=611），围绝经期与绝经后女性纳入非产后组（n=409）。设置随机种子数并以7∶3的比例分为训练集与验证集。收集所有研究对象的相关临床资料，使用单因素及Lasso回归筛选有意义的变量，将其纳入KNN、SVM、DT及RF算法中并构建模型，分别计算模型的敏感度、特异度、准确度、曲线下面积（AUC）等，筛选出最优的模型。结果 产后组SUI患者为352例，占57.6%。根据单因素及Lasso回归，产后组筛选出有意义的变量为：年龄、身体质量指数（BMI）、快肌阶段最大值、孕次、膀胱颈移动度（BND）、尿道旋转角（URA）、会阴侧切、既往尿失禁史及便秘。在产后组验证集中KNN、SVM、DT、RF模型的AUC分别为0.881、0.878、0.750、0.905，RF模型的AUC、准确度、F1指数及Kappa值均最大。非产后组SUI患者为260例，占63.6%。根据单因素及Lasso回归，非产后组筛选出有意义的变量为：年龄、BMI、快肌阶段最大值及恢复时间、慢肌阶段平均值、后静息阶段变异性、阴道分娩、既往尿失禁史及便秘。在非产后组验证集中KNN、SVM、DT、RF模型的AUC分别为0.819、0.805、0.603、0.830，RF模型的AUC、准确度、Kappa值均最大。结论 本研究基于机器学习成功建立4种产后42 d女性，围绝经期及绝经后女性SUI发病的预测模型，其中采用RF算法的模型预测效率最佳。

Abstract

Objective To construct prediction models of female stress urinary incontinence (SUI), and evaluate the efficacy of each model, so as to provide reference for the early diagnosis of SUI. Methods Female SUI patients treated in our hospital during Oct. 2019 and Oct. 2023 and healthy women undergoing physical examination during the same period were involved. Women 42 days after delivery were included in the postpartum group (n=611), and perimenopausal and postmenopausal women were included in the non-postpartum group (n=409). The number of random seeds was set and the participants were divided into the training and verification sets in a ratio of 7∶3. Relevant clinical data were collected, and meaningful variables were screened using single factor and Lasso regression, which were then incorporated into the K-nearest neighbor method (KNN), support vector machine (SVM), decision tree (DT) and random forest (RF) algorithms. The sensitivity, specificity, accuracy and area under the receiver operating characteristic curve (AUC) of the models were calculated to screen out the optimal model. Results There were 352 SUI patients (57.6%) in the postpartum group. According to single factor and Lasso regression, significant variables included age, body mass index (BMI), maximum rapid muscle stage, parity, bladder neck mobility (BND), urethral rotation angle (URA), lateral perineal incision, past incontinence, and constipation. In the verification set, the AUC of KNN, SVM, DT and RF models were 0.881, 0.878, 0.750 and 0.905, respectively; the AUC, accuracy, F1 index and Kappa value of RF model were the largest. In the non-postpartum group, there were 260 SUI patients, accounting for 63.6%. The significant variables were age, BMI, maximum value and recovery time of fast muscle stage, mean value of slow muscle stage, post-resting stage variability, vaginal delivery, past incontinence, and constipation. In the verification set, the AUC of KNN, SVM, DT and RF models were 0.819, 0.805, 0.603 and 0.830, respectively; the AUC, accuracy, Kappa value of the RF model were the largest. Conclusion This study successfully established 4 prediction models for the incidence of SUI in women at 42 days postpartum, perimenopausal and postmenopausal women based on machine learning. Among them, the model adopting the RF algorithm had the best prediction efficiency.

关键词

压力性尿失禁 / 预测模型 / 机器学习 / 决策树 / 随机森林 / 支持向量机 / K最近邻法

Key words

stress urinary incontinence / prediction model / machine learning / decision tree / random forest / support vector machine / K-nearest neighbor method

参考文献

[1] 周艳艳， 丁家森， 高瞻， 等. 女性压力性尿失禁中医诊疗指南（2023）[J]. 中医杂志， 2024， 65（13）： 1408-1416.

[2] 金鑫悦， 黄锦华， 尤晓凤， 等. 近10年中国成年女性压力性尿失禁患病率及影响因素的Meta分析[J]. 中国疗养医学， 2024， 33（3）： 24-29.

[3] BEECKMAN D. A decade of research on incontinence-associated dermatitis (IAD): evidence, knowledge gaps and next steps[J]. J Tissue Viability, 2017, 26(1): 47-56.

[4] REIS AM, BRITO LGO, LUNARDI ALB, et al. Depression, anxiety, and stress in women with urinary incontinence with or without myofascial dysfunction in the pelvic floor muscles: a cross-sectional study[J]. Neurourol Urodyn, 2021, 40(1): 334-339.

[5] 王元毅. 新疆南疆地区女性压力性尿失禁的人工神经网络模型构建[D]. 新疆： 石河子大学， 2021.

[6] 郑睿颖， 刘根焰. 机器学习在感染性疾病临床预测模型中的应用进展[J]. 中国血吸虫病防治杂志， 2023， 35（3）： 317-321.

[7] 杨欣， 王建六， 孙秀丽， 等. 北京大学女性压力性尿失禁诊疗指南（草案）[J]. 中国妇产科临床杂志， 2012， 13（2）： 158-160.

[8] 中华医学会妇产科学分会妇科盆底学组. 女性压力性尿失禁诊断和治疗指南（2017）[J]. 中华妇产科杂志， 2017， 52（5）： 289-293.

[9] BI Y, XIANG D, GE Z, et al. An interpretable prediction model for identifying N⁷-methylguanosine sites based on XGBoost and SHAP [J]. Mol Ther Nucleic Acids, 2020, 22: 362-372.

[10] ZHANG D, ZHOU M, ZHANG M, et al. Development and external validation of clinical predictive model for stress urinary incontinence in Chinese women: a multicenter retrospective study[J]. BMC Womens Health, 2024, 24(1): 532.

[11] FU L, HUANG G, SUN Z, et al. Predicting the occurrence of stress urinary incontinence after prolapse surgery: a machine learning-based model[J]. Ann Transl Med, 2023, 11(6): 251.

[12] 孟祥勇， 秦嘉怡， 陈文生. 基于机器学习的早期胃癌淋巴结转移预测模型构建与验证[J]. 陆军军医大学学报， 2024， 46（21）： 2432-2442.

[13] 王厚清， 刘瑜， 谭夏秋， 等. 基于机器学习的预测前列腺癌生化复发的多基因模型的构建[J]. 临床泌尿外科杂志， 2024， 39（12）： 1107-1111.

[14] 胡洋， 楼叶琳， 张恒， 等. 盆底超声与盆底肌电生理参数评估产后女性压力性尿失禁的相关性研究[J]. 浙江医学， 2022， 44（10）： 1099-1103.

[15] 邓肖群， 龙晓莉， 陈浩， 等. 盆底超声参数及盆底肌电生理参数与女性产后尿失禁患者病情严重程度的相关性研究[J]. 临床和实验医学杂志， 2024， 23（8）： 868-872.

[16] 郝雨， 钱彪， 倪钊， 等. 初步探索新疆阿拉尔地区女性压力性尿失禁患者盆底超声的特征改变[J]. 吉林医学， 2020， 41（12）： 2890-2894.

[17] 于津， 谢军. 产后压力性尿失禁危险因素的探讨及其预测模型的构建[J]. 同济大学学报（医学版）， 2022， 43（3）： 421-426.

[18] 王莹， 文英， 马琳， 等. 会阴侧切分娩助产对经阴道自然分娩产妇产后盆底功能的影响[J]. 中国当代医药， 2021， 28（32）： 109-112.

基金资助

人才发展专项-兵团英才项目(CZ000821)