【摘要】 目的 开发基于机器学习算法的男性不育列线图风险预测模型，为男性不育患者的早期筛查提供依据。方法回顾性选取2023年1月1日至12月31日杭州市中医院收治的124例男性不育患者和来院检查的31名正常生育男性，收集并比较两组对象血常规、生化、生殖激素、维生素以及生殖抗体系列指标。将差异有统计学意义的指标纳入最小绝对收缩和选择算子（LASSO）回归，使用最小λ筛选预测变量。将整个数据集按照7∶3分成训练集（107例）和验证集（48例），使用逻辑回归（LR）、支持向量机（SVM）、随机森林（RF）、K近邻模型（KNN）和决策树（DT）算法构建评估模型。选择最佳算法建立列线图预测模型。结果 13个变量纳入到LASSO回归，使用λ最小值（λ=0.010 4）筛选获得10个预测变量。使用DT、KNN、LR、RF和SVM构建5个预测模型，其中LR算法构建的模型预测效能最佳，AUC为0.942，准确度为0.879，灵敏度和特异度为0.903和0.748。对10个预测因素进行多因素logistic模型筛选，最终筛选出前向运动精子、精子总数、维生素B₁、维生素B₃和25-羟基维生素D等5个因素，构建的列线图风险预测模型预测男性不育的AUC为0.893，准确度为0.842。校准曲线分析显示模型预测概率较好。结论 本研究基于不同机器学习算法构建男性不育的预测模型，通过列线图风险预测模型有效预测男性不育风险，可作为男性不育的早期筛查参考。更多还原

【Abstract】 Objective To develop and validate a nomogram risk prediction model for male infertility based on machine learning algorithms. Methods A retrospective study was conducted on 124 male infertility patients（infertility group） and 31 normozoospermic males（control group） from Hangzhou Traditional Chinese Medicine Hospital between January and December 2023. Data on blood routine, biochemical profiles, reproductive hormones, vitamins, and reproductive antibodies were collected and compared between the groups. Statistically significant variables were incorporated into least absolute shrinkage and selection operator（LASSO） regression to identify predictors using the minimum λ value. The dataset was randomly split into a training set（n=107, 70%） and a validation set（n=48, 30%）. Five prediction models were built using decision tree（DT）, K-nearest neighbor（KNN）, logistic regression（LR）, random forest（RF）, and support vector machine（SVM）algorithms. The best-performing algorithm was used to construct the nomogram. Results Thirteen variables were included in the LASSO regression, and 10 predictors were selected with the minimum λ（λ=0.010 4）. Among the five prediction models,the LR model demonstrated the best performance（AUC=0.942, accuracy=0.879, sensitivity=0.903, specificity=0.748）.Multivariate logistic regression analysis on the 10 predictors identified five final predictors: progressive motility, total sperm count, vitamin B₁, vitamin B₃, and 25-hydroxyvitamin D. The nomogram constructed with these factors achieved an AUC of0.893 and accuracy of 0.842 in predicting male infertility. Calibration curve analysis indicated good agreement between predicted and actual probabilities. Conclusion This study successfully constructed and validated prediction models for male infertility using machine learning algorithms. The developed nomogram risk prediction model effectively assesses the risk of male infertility and shows potential as a reference tool for early screening.更多还原