基于机器学习预测高血压患者发生临床前靶器官损害风险
Predicting the risk of preclinical target organ damage in hypertensive patients based on machine learning
王淼 1樊洪轩 1王蕾刚 1任兆煜 1梁斌1
作者信息
- 1. 山西医科大学第二临床医学院,太原 030001
- 折叠
摘要
目的 通过机器学习算法,探索高血压患者发生临床前靶器官损伤的独立危险因素,并开发和验证预测模型,为临床决策做参考.方法 选择已发表的文献中 2019 年 10 月至 2021 年 10 月关于塞拉利昂弗里敦西部地区城市的成年人进行的一项健康筛查调查,结局指标为是否发生临床前靶器官损害,本研究共纳入了2 373 例高血压患者,70.00%(1 661 例)的数据用于模型训练,30.00%(712 例)的数据用于模型验证.使用随机森林筛选死亡的独立危险因素,采用多元logistic回归分析在训练集中建立预测模型,然后测试集中进行验证.使用一致性指数(C-index)、受试者工作特征(ROC)曲线,校准图和决策曲线分析评估预测模型的辨别性、校准和临床有效性.结果 在 2 373 例患者中,发生靶器官损害的比例为 17.20%,其中训练集 17.30%,测试集 17.00%.经过筛选,最后纳入临床前靶器官损害的模型为舒张压、收缩压、低密度脂蛋白、甘油三酯与高密度脂蛋白的比值、腰臀比、肾小球滤过率分级、体质指数、糖尿病史及性别,经验证模型性能良好.结论 该模型能够很好地预测确诊高血压患者发生临床前靶器官损害的风险,为临床决策做参考.
Abstract
Objective To explore independent risk factors for preclinical target organ damage in hypertensive patients through machine learning algorithms,and to develop and validate predictive models for reference in clinical decision-making.Methods A health screening survey was conducted among adults residing in cities in the western region of Freetown,Sierra Leone from October 2019 to October 2021.The outcome measure was whether preclinical target organ damage occurred.A total of 2 373 hypertensive patients were included in this study,with 70.00%(n=1 661)of the data used for model training and 30.00%(n=712)used for model validation.We use random forest to screen for independent risk factors for death,use multiple logistic regression analysis to establish a predictive model in the training set,and then validate it in the test set.We use consistency index(C-index),receiver operating characteristic(ROC)curve,calibration chart,and decision curve analysis to evaluate the discriminability,calibration,and clinical validity of the predictive model.Results Among 2 373 patients,the proportion of target organ damage was 17.20%,with 17.30%in the training set and 17.00%in the testing set.Finally,there are nine risk factors after screening which are gender,diastolic blood pressure,systolic blood pressure,low-density lipoprotein,the ratio of triglyceride to high-density lipoprotein,waist hip ratio,glomerular filtration rate,body mass index and history of diabetes.The model performance was verified to be good.Conclusion Our model can effectively predict the risk of preclinical target organ damage in diagnosed hypertensive patients,providing reference for clinical decision-making.
关键词
高血压/机器学习/预测Key words
Hypertension/Machine learning/Forecasting引用本文复制引用
出版年
2024
NETL
NSTL
万方数据