基于机器学习的糖尿病男性患者勃起功能障碍风险评估模型的开发与验证

Development and internal validation of a machine learning-based risk assessment model for erectile dysfunction in male patients with diabetes

摘要

目的 基于机器学习方法构建糖尿病男性患者勃起功能障碍的风险评估模型，并通过可解释性分析揭示关键风险特征。方法 本研究数据来源于2001-2004年美国国家健康与营养调查（NHANES）数据库，筛选符合纳入标准的498例糖尿病男性患者作为研究对象。按8∶2比例划分训练集与测试集，采用最小绝对收缩和选择算子（LASSO）回归筛选特征变量，并在训练集中采用SMOTE进行类别平衡处理。系统比较了12种机器学习算法的预测性能，以曲线下面积（AUC）、准确率、召回率、F1分数及Kappa系数等指标进行综合评估，筛选最优模型，并采用SHAP方法进行可解释性分析。结果 极端随机树模型在各项评估指标上表现最优，AUC为0.763，准确率为0.721，召回率为0.632，F1分数为0.635，Kappa系数为0.410。SHAP分析显示，对模型预测贡献最大的前5个特征依次为年龄、高血压病史、尿素氮、良性前列腺增生（BPH）及家庭收入贫困比（PIR）。年龄较高、有高血压病史、尿素氮升高及合并BPH与勃起功能障碍风险升高相关，而较高的PIR及血红蛋白水平对勃起功能障碍具有保护作用。结论 本研究构建的基于极端随机树算法的预测模型具有较高的判别能力和临床可解释性，所纳入的预测特征均为临床常规可及指标，可为糖尿病男性患者勃起功能障碍的早期识别和主动干预提供参考。

Abstract

Objective To develop a machine learning-based risk assessment model for erectile dysfunction（ED）in male patients with diabetes mellitus, and to reveal key predictive features through interpretability analysis. Methods Data were obtained from the National Health and Nutrition Examination Survey（NHANES）2001-2004 cycles. A total of 498 male diabetic patients meeting the inclusion criteria were enrolled. The dataset was split into training and test sets at an 8∶2 ratio. Least absolute shrinkage and selection operator（LASSO）regression was employed for feature selection, and synthetic minority over-sampling technique（SMOTE）was applied to the training set for class balancing. Twelve machine learning algorithms were systematically compared, with model performance comprehensively evaluated using the area under the receiver operating characteristic curve（AUC）, accuracy, recall, F1-score, and Kappa coefficient. The optimal model was selected and further interpreted using the SHAP method. Results The Extra Trees model demonstrated the best overall performance, achieving an AUC of 0.763, accuracy of 0.721, recall of 0.632, F1-score of 0.635, and Kappa coefficient of 0.410. SHAP analysis revealed that the top five contributing features were age, history of hypertension, blood urea nitrogen, benign prostatic hyperplasia, and poverty income ratio （PIR）. Higher age, presence of hypertension, elevated blood urea nitrogen, and presence of benign prostatic hyperplasia were associated with increased ED risk, whereas higher PIR and hemoglobin levels exhibited protective effects against ED. Conclusion The risk assessment Extra Tress model for ED in diabetic patients demonstrated acceptable discriminative ability and clinical interpretability. All predictive features included in the model are routinely accessible clinical indicators, providing a practical reference for early identification and proactive intervention of ED in diabetic patients.

关键词

勃起功能障碍 / 糖尿病 / 机器学习 / SHAP / 美国国家健康与营养调查（NHANES） / 模型 / 风险评估

Key words

erectile dysfunction / diabetes mellitus / machine learning / SHAP / National Health and Nutrition Examination Survey / model / risk assessment

参考文献

[1] 吕坤龙， 郑涛， 张天标， 等. 阴茎振动阈值测定能帮助诊断勃起功能障碍的严重程度[J]. 现代泌尿外科杂志， 2024，29（2）: 119-121，167.

[2] KESSLER A, SOLLIE S, CHALLACOMBE B, et al. The global prevalence of erectile dysfunction: a review[J]. BJU Int, 2019, 124(4): 587-599.

[3] KÖHLER TS, KLONER RA, ROSEN RC, et al. The princeton IV consensus recommendations for the management of erectile dysfunction and cardiovascular disease[J]. Mayo Clin Proc, 2024, 99(9): 1500-1517.

[4] DEFEUDIS G, MAZZILLI R, TENUTA M, et al. Erectile dysfunction and diabetes: a melting pot of circumstances and treatments[J]. Diabetes Metab Res Rev, 2022, 38(2): e3494.

[5] HOSTNIK B, TONIN G, JANEŽ A, et al. Erectile dysfunction in diabetes mellitus: a comprehensive narrative review of pathophysiology, genetic association studies and therapeutic approaches[J]. Endocrinol Diabetes Metab, 2025, 8(5): e70099.

[6] KITAW TA, ABATE BB, TILAHUN BD, et al. The global burden of erectile dysfunction and its associated risk factors in diabetic patients: an umbrella reviews[J]. BMC Public Health, 2024, 24(1): 2816.

[7] DEFEUDIS G, MAZZILLI R, DI TOMMASO AM, et al. Effects of diet and antihyperglycemic drugs on erectile dysfunction: a systematic review[J]. Andrology, 2023, 11(2): 282-294.

[8] 王佩佩， 侯钊， 马慧， 等. 基于机器学习的肾癌患者术后复发风险预测模型的构建与评价[J]. 现代泌尿外科杂志， 2025，30（3）: 240-247.

[9] GUO C, LIU Z, FAN H, et al. Machine-learning-based plasma metabolomic profiles for predicting long-term complications of cirrhosis[J]. Hepatology, 2025, 81(1): 168-180.

[10] CHEN XY, LU WT, ZHANG D, et al. Development and validation of a prediction model for ED using machine learning: according to NHANES 2001-2004[J]. Sci Rep, 2024, 14(1): 27279.

[11] 达尼亚尔·努尔德别克， 刘彼得， 南玉奎， 等. 体圆指数与急迫性尿失禁的相关性:一项基于NHANES的横断面研究[J]. 现代泌尿外科杂志， 2025，30（12）: 1084-1089.

[12] WANG G, NI C. Association of erectile dysfunction and peripheral arterial disease in NHANES 2001-2004: a cross-sectional study[J]. Front Endocrinol (Lausanne), 2024, 15: 1439609.

[13] RAO Z, YANG W, YANG Y, et al. Revolutionizing Wilson disease prognosis: a machine learning approach to predict acute-on-chronic liver failure[J]. J Transl Med, 2025, 23(1): 999.

[14] NWANOSIKE EM, CONWAY BR, MERCHANT HA, et al. Potential applications and performance of machine learning techniques and algorithms in clinical practice: a systematic review[J]. Int J Med Inform, 2022, 159: 104679.

[15] 陈卫宏， 杨玲， 陈悦， 等. 中国男性勃起功能障碍患病率的Meta分析[J]. 中国性科学， 2024，33（7）: 12-18.

[16] DILIXIATI D, WAILI A, TUERXUNMAIMAITI A, et al. Risk factors for erectile dysfunction in diabetes mellitus: a systematic review and meta-analysis[J]. Front Endocrinol (Lausanne), 2024, 15: 1368079.

[17] JIN M, YUAN S, WANG B, et al. Association between prediabetes and erectile dysfunction: a meta-analysis[J]. Front Endocrinol (Lausanne), 2022, 12: 733434.

[18] HSIAO W, BERTSCH RA, HUNG YY, et al. Tighter blood pressure control is associated with lower incidence of erectile dysfunction in hypertensive men[J]. J Sex Med, 2019, 16(3): 410-417.

[19] MANOLIS A, DOUMAS M, FERRI C, et al. Erectile dysfunction and adherence to antihypertensive therapy: focus on β-blockers[J]. Eur J Intern Med, 2020, 81: 1-6.

[20] PIZZOL D, XIAO T, YANG L, et al. Prevalence of erectile dysfunction in patients with chronic kidney disease: a systematic review and meta-analysis[J]. Int J Impot Res, 2021, 33(5): 508-515.

[21] MALIK MA. The role of kidney transplantation in the management of erectile dysfunction: a review of hormonal changes and psychosocial impacts in male patients[J]. Sex Med Rev, 2026, 14(1): qeaf079.

[22] ZHANG Y, ZANG N, XIANG Y, et al. A comprehensive analysis of erectile dysfunction prevalence and the impact of prostate conditions on ED among US adults: evidence from NHANES 2001-2004[J]. Front Endocrinol (Lausanne), 2025, 15: 1412369.

[23] FANG Y, DONG Z, HUANG T, et al. The role of socioeconomic status and oxidative balance score in erectile dysfunction: a cross-sectional study[J]. Heliyon, 2023, 9(11): e22233.

[24] IRFAN M, HUSSAIN NHN, NOOR NM, et al. Epidemiology of male sexual dysfunction in Asian and European regions: a systematic review[J]. Am J Mens Health, 2020, 14(4): 1557988320937200.

[25] CORONA G, RASTRELLI G, MONAMI M, et al. Body weight loss reverts obesity-associated hypogonadotropic hypogonadism: a systematic review and meta-analysis[J]. Eur J Endocrinol, 2013, 168(6): 829-843.