基于临床指标和机器学习的早期骨质疏松预测

Clinical Value of Machine Learning Models in Early Osteoporosis Prediction Based on Demographic and Routine Biochemical Indicators

目的:探讨机器学习基于人口学和常规生化指标预测骨质疏松的临床价值。方法:回顾性分析2053例50岁以上、接受低剂量CT扫描的健康受试者(女性906例,男性1147例)的人口学和常规生化指标。根据定量CT分析出的体积骨密度进行骨质疏松的诊断。将受试者按7:3的比例分为训练集和测试集,使用逻辑回归、支持向量机、决策树、随机森林和多层感知机共5种不同的算法构建模型,并评估模型性能。结果:在女性中,随机森林模型在训练(AUC=0.90)和测试集(AUC=0.80)中都是最佳模型,最重要特征是年龄,其次是碱性磷酸酶、甘油三酯和体重指数。在男性中,逻辑回归模型在测试集(AUC=0.81)中表现最好,人口学特征的重要性高于常规生化指标。结论:基于人口学和常规生化指标的性别特异性机器学习模型为体检等临床场景下的骨质疏松筛查提供了可能性。

Purpose:To investigate the clinical value of machine learning in predicting osteoporosis based on demographic and routine biochemical indicators.Methods:The demographic and routine biochemical indicators of 2053 healthy subjects(906 females and 1147 males)over 50 years who underwent low-dose CT scans were retrospectively analyzed.Diagnosis of osteoporosis is based on volumetric bone mineral density concluded from quantitative CT.The subjects were divided into training set and test set in a ratio of 7:3.Five different algorithms were used to build the model,including logistic regression,support vector machine,decision tree,random forest and multilayer perceptron.Results:In women,the random forest was the best model in both training(AUC=0.90)and test set(AUC=0.80),with age as the most important feature,followed by alkaline phosphatase,triglycerides,and body mass index.In men,the logistic regression performed best in the test set(AUC=0.81),and the importance of demographic characteristics was higher than that of routine biochemical indicators.Conclusion:The sex-specific machine learning model based on demographic and routine biochemical indicators provides a possibility for osteoporosis screening in clinical scenarios such as physical examination.