不同机器学习算法对老年原发性高血压发病风险的预测价值

Predictive value of different machine learning algorithms for the risk of developing essential hypertension in the elderly

目的了解老年原发性高血压发病的危险因素,并利用机器学习算法构建老年原发性高血压发病风险预测模型。方法选取2021年1月至12月解放军第三〇五医院体检中心体检、血压正常且无高血压病史的人群为研究对象,并于2023年1月至12月观察其血压情况,根据血压结果将其分为血压正常组(1553名)和新发高血压组(428例)。分析老年原发性高血压发病的危险因素,使用不同机器学习算法(随机森林、决策树、支持向量机、K临近分类、多层感知机、逻辑回归)构建老年原发性高血压发病风险预测模型,受试者操作特征曲线评估模型的预测效能。结果两组年龄、白细胞计数、低密度脂蛋白胆固醇、总胆固醇、淋巴细胞百分比、球蛋白、收缩压、舒张压、糖化血红蛋白、体重指数、血红蛋白、血糖、直接胆红素、中性粒细胞百分比比较,差异有统计学意义(P<0.05)。年龄、收缩压、体重指数、血红蛋白是老年原发性高血压发病的危险因素(OR=1.209、1.204、1.243、1.218,P<0.05)。随机森林预测模型的准确度、灵敏度、特异度及约登指数、曲线下面积高于决策树、支持向量机、K临近分类、多层感知机、逻辑回归预测模型。结论年龄、收缩压、体重指数、血红蛋白是老年原发性高血压发病的危险因素;基于上述因素构建的老年原发性高血压发病风险预测模型中随机森林预测模型有更好的分类效果和判别能力。

Objective To understand the risk factors of developing essential hypertension in the elderly,and to construct a risk prediction model for developing essential hypertension in the elderly using machine learning algorithms.Methods People with normal blood pressure and no history of hypertension who had a physical examination at the Physical Examination Center,the 305th Hospital of the Chinese People’s Liberation Army from January to December 2021 were selected as study subjects,and their blood pressure was observed from January to December 2023,and they were divided into the normal blood pressure group(1553 people)and the new-onset hypertension group(428 cases)according to their blood pressure results.The risk factors of developing essential hypertension in the elderly were analyzed,different machine learning algorithms(random forest,decision tree,support vector machine,K-proximity classification,multi-layer perceptual machine,and logistic regression)were used to construct a predictive model of the risk of developing essential hypertension in the elderly,and the predictive efficacy of the model was assessed by receiver operating characteristic curve.Results There were significant differences in age,white blood cell count,low-density lipoprotein cholesterol,total cholesterol,lymphocyte percentage,globulin,systolic blood pressure,diastolic blood pressure,glycosylated hemoglobin,body mass index,hemoglobin,blood glucose,direct bilirubin,and neutrophils percentage between two groups(P<0.05).Age,systolic blood pressure,body mass index,and hemoglobin were the risk factors for essential hypertension in the elderly(OR=1.209,1.204,1.243,1.218,P<0.05).The accuracy,sensitivity,specificity,Jorden index,and area under the curve of random forest prediction model were higher than decision tree,support vector machine,K-proximity classification,multi-layer perceptron machine,logistic regression prediction model.Conclusion Age,systolic blood pressure,body mass index,and hemoglobin were the risk factors of developing essential hypertension in the elderly;the random forest prediction model has better classification effect and discriminative ability among the prediction models for the risk of developing of essential hypertension in the elderly constructed based on the above factors.